AI Tools, Especially Chatbots and Disabled People: Views of Critical Disability Studies Students

Abstract

AI tools, including chatbots, have become increasingly mainstream and could pose many problems for disabled people and their allies. Such problems could include the lack of accessibility of AI tools for disabled people, including disabled activists and disabled people as digital citizens. Chatbot outputs also may be inaccurate or of low quality due to biased or insufficient training data and might reflect biased and low literacy about the lived reality of disabled people amongst the people involved in setting up the training. Therefore, problems could also include a potential decrease in the public’s literacy about the lived realities of disabled people when AI tools such as chatbots are used without critically analyzing their outputs or using inappropriate prompts for inquiry. Critical disability studies students see themselves as allies of disabled people and want to make a positive difference in their lives. AI tools, including chatbots, are an emerging area that allies of disabled people have to think about because of their increasing impact on the lived reality of disabled people. Using an online survey hosted on a Qualtrics platform, we examined the perspectives of 89 critical disability studies students on the usability and impact of AI tools, including chatbots, on disabled people in areas including the following: accessibility; what knowledge is available and what and whose knowledge is reflected in the output of AI tools and chatbots; the impact of AI tools and chatbots’ output on the literacy of the users about the lived reality of disabled people; and the impact on ability judgment problems disabled people experience, such as ableism and disablism. We performed a thematic analysis of two open-ended questions and frequency (%) analysis of the other questions. Our 89 participants outlined many problems. One set of problems covered accessibility barriers and design-process shortcomings of AI tools, including chatbots. Another set was linked to content-related issues, such as chatbots not reflecting the lived realities of disabled people, the use of biased language and portrayals of disabled people, and the limited usefulness of chatbot-provided information. These issues are not only problematic for disabled people but also for their allies. Participants gave many suggestions regarding the practical impacts of not having data on the lived reality of disabled people, as well as how chatbots could be used, or what has to happen, for them to challenge ableist power structures rather than reinforcing it. All our findings suggest the need for various actions whereby AI ethics, “responsible AI”, and “AI for good” principles could be used to decrease the problems noted, as could be the concepts of AI justice and of ability judgment justice and other ability judgment-linked concepts.

1. Introduction

AI systems’ presence and use have become increasingly ubiquitous [1], and as such, are investigated extensively in the academic literature. For example, searching the abstracts of the academic database Scopus on 5 April 2026 generated 1,034,394 hits for the search string “artificial intelligence” OR “machine learning”. The term “chatbots” generated 20,266 hits, “AI-tools” 14,028 hits, “LLM” 35,168 hits, “generative AI” 17,539 hits, and “Natural Language Processing” 79,667 hits. The search string (“chatbot*” OR “Natural Language Processing” OR “large-scale language models” OR “LLM” OR “AI-tools”) generated 141,932 hits. Many problems related to AI are covered in the academic literature, often under the terms “AI ethics” (2463 abstracts in Scopus) and “responsible AI” (3160 abstracts). Most, if not all, of the problems discussed apply also to disabled people.

Many of these problems can be categorized under three sets of problems that disabled people1 face in relation to AI systems.

One category is the production of knowledge about disabled people and AI in the academic literature. Within this category, there are three issues that feed and amplify each other: (a) invisibility of disabled people period [2,3,4], (b) if covered, a biased coverage that manifests itself in an overrepresentation of medical and deficiency imagery of disabled people [2,3,4] and (c) missing data reflecting the lived social reality of disabled people [5,6,7]. Using Scopus abstracts again as sources to highlight the problem of invisibility, if one uses (“chatbot*” OR “Natural Language Processing” OR “large-scale language models” OR “LLM” OR “AI-tools”) and adds, for example, the terms “disabled people” OR “people with disabilities” to the search string, the hit count drops from 141,932 to 110. If one uses “ADHD” OR “autism” OR “attention deficit” OR “autistic”, the number drops to 357, whereby many of these abstracts have a medical angle. If one uses “neurodiv*”, a term normally indicating a non-medical coverage, the number drops to 43.

The second category of problems is the content and judgments produced by AI tools, such as chatbots and AI writing assistants. Given the gaps and biases in the available data outlined under the first category, these tools cannot provide an accurate, unbiased description of the lived reality of disabled people. Linked to this problem are biased judgments made by AI tools due to algorithmic bias related to disabled people [8,9,10,11,12,13]. This is part of a bigger problem, as “AI tools have been shown to reproduce and reinforce human biases” [14]. This has been attributed, in part, to “people providing biased training data, prompts, or feedback during the training process when creating these applications” [14] (p. 1).

The third category of problems is the usability of the AI tools and the lack of involvement of disabled people in their design [15], which is flagged by many policy documents, such as the 2025 Canadians Accessible and Equitable Artificial Intelligence Systems standard document [16].

Then there is a fourth category that is flagging problems using the concept of ability judgments, such as ableism and disablism and actions related to it [17]. Ability judgments are a foundational theoretical lens of disability studies and ability-based studies (ability expectation and ableism studies, short ability studies [18,19,20], studies in ableism [21,22,23], and critical studies of ableism [24,25]).

All these problems cause many issues for disabled people, ranging from a decrease in the quality of their lived reality, such as a decline in employment, to a decrease in the literacy of society around their lived reality, and an increase in negative ability judgments (disablism), whereby the decrease in literacy can increase, in particular, passive disablism where the “primary purpose was not to generate the disablement due to ability differences (perceived or real) but it is a side effect of not thinking, not being aware of the consequences of one’s ability expectations” [20].

Not only that, but disabled people also need allies to rectify or prevent policy decisions that negatively impact their lived reality, and the development and use of AI/ML tools is one such policy area. There are many expectations of allies, including expectations that they be activists, support social change, educate themselves constantly, and educate others [26,27]. As such, all four aforementioned categories of problems have to be potential targets for action by disabled people and their allies, but they also pose challenges for that activism for disabled people and their allies, such as burnout [27,28].

As to burnout, many causes of activists’ and allies’ burnout (due to their activist role) have been reported [27,28].

The lack of societal literacy on the lived reality of disabled people has already been flagged as a burnout problem for disabled people and their allies involved in activism [27]. The noted problems linked to AI systems, such as a decrease in literacy on the lived reality of disabled people, will increase activist and with that ally burnout and reduce the capacity of disabled people and their allies to prevent further negative impacts or to make a positive change.

Given the existing literature, our survey covered the first three categories in relation to AI tools and chatbots and the fourth one in relation to chatbots in areas including the following: accessibility; what knowledge is available and what and whose knowledge is reflected in the output of AI tools and chatbots; the impact of AI tools and chatbots output on the literacy of the users about the lived reality of disabled people; and the impact on the ability judgment problems disabled people experience, such as ableism and disablism.

Asking about accessibility issues is not new. The novelty of our survey lies in the parts that specifically asked questions about the content of chatbots and its impact on the literacy about disabled people, the impact of the content on essential activities, such as activism by disabled people and their allies, and the impact on ability judgments, such as ableism and disablism.

As disabled people need allies, we selected undergraduate critical disability studies students for our survey because they see themselves as allies of DP/PWD [29] and want to raise awareness about the problematic lived reality of DP/PWD [30]. As such, we were interested in understanding the views of critical disability studies students on the topic of AI tools and specifically chatbots in relation to disabled people.

1.1. AI Governance

AI ethics, responsible AI, and AI for good are three areas that aim to give guidance as to how AI should be developed and used.

Regarding AI ethics [31,32], it is stated that well-being can be reliably measured using the following measures:

• Community: belonging, crime and safety, discrimination and inclusion, participation, and social support;
• Culture: identity and values;
• Economy: economic policy, equality and environment, innovation, jobs, and sustainability;
• Natural resources and consumption and production and standard of living;
• Education: formal education, lifelong learning, and teacher training;
• Environment: air, biodiversity, climate change, soil, and water;
• Government: confidence, engagement, human rights, and institutions;
• Human settlements: energy, food, housing, information and communication technology, and transportation;
• Physical health: health status, risk factors, and service coverage;
• Psychological health: affect (feelings), flourishing, mental illness and health, and satisfaction with life;
• Work: governance, time balance, and workplace environment.

Then there is UNESCO’s 2021 AI-ethics document, which is described as

“a globally accepted normative instrument that focuses not only on the articulation of values and principles, but also on their practical realization, via concrete policy recommendations, with a strong emphasis on inclusion issues of gender equality and protection of the environment and ecosystems”.

[33] (p. 14)

This document contains a section called “Policy area 11: health and social well-being” and mentions disabled people specifically in various sections. Other provisions are also applicable to disabled people despite them not being specifically identified. To list only a few recommendations. “It holds that no person or community should be harmed or subordinated (physically, economically, socially, politically, culturally, or mentally) at any stage of the AI life cycle, and that new technologies should provide new means of advocating for and exercising human rights rather infringing them” (p. 18). The recommendations state that AI actors should have an enabling role in building peaceful, just, and interconnected societies, on the premise that the whole thrives only when all its constituent parts are able to thrive (pp. 19–20) and the promotion of participation and leadership of marginalized and vulnerable groups, including disabled people, in AI education programmes at all levels (p. 34).

Given the many problems noted around AI and disabled people, including around the term well-being [4], many actions to fulfill the demands of the document are needed.

As to responsible AI [34], the 2018 Montréal Declaration For A Responsible Development Of Artificial Intelligence lists the following principles:

• Well-being principle;
• Respect for autonomy principle;
• Protection of privacy and intimacy;
• Solidarity principle;
• Democratic participation principle;
• Equity principle;
• Diversity inclusion principle;
• Caution principle;
• Responsibility principle
• Sustainable development principle

Existing literature around AI and disabled people suggests that the AI governance discourse does not cover most of these principles in relation to disabled people.

Social good is mentioned in many policy documents linked to AI, such as “AI/IT/computer for social good” [35,36,37,38,39,40,41,42,43,44,45,46,47,48,49], “AI for common good” [50], “chatbots for social good, which are chatbots that provide scalable solutions to social and environmental challenges facing humanity while supporting human agency” [51] (p. 1), and “ChatGPT for Good” [52,53,54]. Seven factors are listed as essential for AI for social good [37]. Under number 6, situational fairness, it is stated, “AI developers typically rely on data, which may be biased in ways that are socially significant. This bias may carry across to the algorithmic decision-making that underpins many AI systems, in ways that are unfair to the subjects of the decision making process (Caliskan, Bryson, and Narayanan 2017) and, thus, may breach the principle of justice” [37] (p. 1786). And they state, “AI4SG designers should remove from relevant datasets variables and proxies that are irrelevant to an outcome, except when their inclusion supports inclusivity, safety, or other ethical imperatives” [37] (p. 1788). Fulfilling the goals and governance indicators for AI system advancements, such as the seven factors of “AI for social good”, requires engagement with disabled people, as they are experiencing well-being problems and injustices, as the UN Convention on the Rights of Persons with Disabilities [7] highlights. And if AI advancements are to work for disabled people, one needs an intricate understanding of disabled people and their lived reality.

All three fields mention well-being. The question is the following: do AI tools, including chatbots, increase or decrease the well-being of disabled people in these areas? Is data generated so one can answer this question? Existing academic literature around AI and well-being suggests a rather biased coverage of disabled people with many gaps [4].

1.2. AI Justice, Disability Justice, and Ability Judgment Justice

AI justice is a fairly new term. Although we could not find a formal definition, AI justice is described as examining how AI advancement and its governance influence social hierarchies [55], and AI justice is linked to relational equality, social accountability, and structural reform [55], procedural ethics, recognitional justice, distributive justice and capabilities justice [56], and the philosophy of Rawls [57]. It is argued that AI justice starts at the grassroots level [58]. A move from AI ethics to AI justice is proposed [55].

Given these premises, engaging with disabled people falls within the scope of AI justice. Disability justice was introduced as a distinctly intersectional lens [59,60,61,62,63,64] to analyze the lived reality of disabled people with its 10 principles: intersectionality; leadership of those most impacted; anticapitalism; cross-movement organizing; wholeness; sustainability; cross-disability solidarity; interdependence; collective access; and collective liberation [64]. Looking at AI justice through a disability justice definition would be useful.

Then there are the fairly new concepts of “Ability judgment justice” and “ability judgment injustice” or “ability justice” and “ability injustice” for short [17,20].

Ability justice is formally defined as: ability justice is achieved in a world “that eliminated irrelevant and/or arbitrary ability expectations, decreased ability privileges and disablism based on these expectations, and enabled the use of ability expectations and ableism to decrease ability judgment-based oppression, inequity, disablism and privilege” [17].

In plain language, “ability justice is achieved in a world where everyone is treated fairly, regardless of what they can or cannot do ability-wise. It occurs when society

• Removes unfair ability expectations;
• Reduces advantages linked to those expectations and
• Stops discrimination against people whose abilities are judged against unfair norms.

It also means using ideas about ability in ways that increase fairness and reduce ability-based discrimination (disablism) for everyone” [17] (for a formal and plain language versions of ability injustice and body/mind specific ability justice and injustice definitions see, Refs. [17,20]).

One premise of the ability justice and injustice definitions is that ability judgments are a general cultural reality and are employed in relation to disabled people, the intersectionality of disabled people and non-disabled marginalized identities and realities (which means that ability justice is intricately linked to disability justice [59,60,61,64]) but also shape the relationships between humans in general, humans and animals, humans and nature, humans and post/trans/cyborg humans, and humans and machines (sentient and non-sentient) [17,18,19,20,65]. Another premise is that ability judgments can be used to disable and enable [17]. As conceptualized, the definitions can cover AI justice in relation to non-disabled people and disabled people separately but also highlighting the commonality of ability judgment-based AI justice issues disabled and non-disabled people face (for more premises see [17]).

Ability judgments also influence and are influenced by ethical reasoning [66], and with that, AI ethics, responsible AI, and various ethics theories covered under AI justice could be engaged with through the lens of ability judgment justice, disability justice, disability studies, and ability-based studies.

1.3. AI Systems, AI Governance, and Disabled People

1.3.1. Policy Documents

Many problems related to AI systems and disabled people have been identified in various policy documents, and what has to happen has been outlined.

The “CAN-ASC-6.2:2025- Accessible and Equitable Artificial Intelligence Systems” document [67] from Accessibility Standards Canada, published in December 2025 states that “upholding the principles of the Accessible Canada Act when deploying AI requires that people with disabilities experience equitable benefits, do not experience inequitable harms or a loss of rights and freedoms, retain agency, are treated with respect, and have the right to choose equitable alternatives” and to ensure that:

• “AI is accessible to people with disabilities (Clause 10);
• AI systems are equitable to people with disabilities (Clause 11);
• Organizations implement processes needed to achieve accessible and equitable AI (Clause 12); and
• AI education supports systemic change needed to achieve accessible and equitable AI (Clause 13)”.

The same document defines AI literacy as an “understanding what AI is, how it works at a basic level, and how it can affect individuals and communities. It includes awareness of: a. how AI systems make decisions or generate outputs; and b. the potential for both immediate and long-term negative impacts, such as biased decisions in services or policies, or future consequences from poor predictions” [67].

Section 13, “Accessible Education, Training and Literacy on AI,” in the same document further requires that AI education, training, and literacy initiatives be accessible, include content about accessible and equitable AI, involve disabled people in their development and delivery, and support the agency and autonomy of people affected by AI-assisted decisions [67].

The policy document “The impact of the use of artificial intelligence on people with disabilities” [68] outlines 10 recommendations. These recommendations call for applying the principle of “Nothing About Us Without Us,” mandating full accessibility in AI development, requiring the inclusion and weighting of disability data, establishing standardized and sector-specific accessibility benchmarks, supporting hybrid AI architectures, investing in domain-targeted AI systems, guaranteeing non-AI alternatives and human-centric options, fostering ongoing community-led testing and feedback, bolstering transparency, documentation, and accountability, and strengthening legal and regulatory safeguards [68] (pp. 9–10).

“Building A Disability-Inclusive AI Ecosystem: A Cross-Disability, Cross-Systems Analysis Of Best Practices” [69] is another recent document highlighting problems and provides specific recommendations in relation to the topics of employment, education, benefits determinations, information and communication technology (ICT), healthcare, transportation, and the criminal justice system.

1.3.2. AI Governance and Disabled People: The Academic Engagement

AI Justice and Disabled People

AI justice is a foundational aspect of AI governance given how it is described. However, searching Scopus, the 80 databases of EBSCO-Host and Web of Science on 3 September 2026, using “AI Justice” OR “AI-justice” OR “artificial intelligence justice” and (“disabled” OR “disabilit*” OR “impair*” OR “deaf” OR “neurodiv*” OR “dyslexia” OR “ADHD” OR “autism” OR “ASD” OR “attention deficit” OR “autistic” OR “wheelchair*”) in abstracts generated no hits in any of the three databases.

In the book “AI for a Just World”, however, the following is argued as a premise of the book:

“AI justice will not emerge from better code alone. It requires redistributive politics, participatory governance, disability-centered policy, Indigenous sovereignty, and care-based design practices. This volume emphasizes leadership by those most impacted—disabled people, racialized communities, migrants, and colonized populations—not as consultees but as decisionmakers”.

[70] (p. xxxii)

Much more academic engagement with AI justice in relation to disabled people is needed.

AI and Disability Justice

Disability justice is a foundational concept to engage with the intersectionality of disabled people [71], and this term should be engaged with extensively in relation to AI, as the different intersectionalities are all impacted by AI differently.

However, the abstract searches of Scopus, the 80 databases of EBSCO-Host and Web of Science on 3 September 2026 generated for “disability justice” and “chatbot*” only 3/2/2 hits and for “disability justice” and “AI” still only 26/17/7 hits.

This suggests an under-engagement with the concept of disability justice within the AI academic literature we searched, which is flagged by others stating, “disability in/justice is often confined to discussions about technology access, with little engagement from critical disability studies on the social, ethical, and political dimensions of AI/ADM” ([72] (p. 4474)).

AI Ethics, AI for Good, AI for Social Good, Responsible AI and Disabled People

Searching Scopus, the 80 databases of EBSCO-Host and Web of Sciences on 3 September 2026, for “responsible AI” and (“disabled” OR “disabilit*” OR “impair*” OR “deaf” OR “neurodiv*” OR “dyslexia” OR “ADHD” OR “autism” OR “ASD” OR “attention deficit” OR “autistic” OR “wheelchair*”) in abstracts led to 48/22/23 hits in all three databases, and the numbers for “AI-ethic*” or “AI ethic*” were 33/41/16.

Given that these hits include duplicates between the 80 databases in EBSCO HOST and between the three main databases, the numbers suggest a limited coverage of AI ethics and responsible AI in conjunction with disabled people.

The result for (“AI for social good” OR “AI for common good” OR “chatbots for social good” OR “ChatGPT for Good” OR “AI for good” OR “IT for good”), with the disability terms, was 11 and 8 for Scopus and Web of Science (EBSCO does ignore words like “for” in a quote, so the hits do not reflect the phrases and are not listed).

As to the academic literature, the lack of engagement of AI ethics with disabled people has been pointed out for some time [2,73]. Newer sources still flag the same problem [11], for example, “disability—a key factor in mobility technology planning—remains overlooked in AI transport ethics” [74] (p. 6291). That disabled people are not knowledge producers but only consumers (Ref. [72]), citing [3], is something that goes against one of the 2021 UNESCO AI ethics document recommendations that demands that marginalized groups lead the discussions. It also goes against the leadership principle of disability justice [60,64].

In the same study, the following was found:

“For the dialogue group participants in our study, ethical reasoning was most strongly influenced by concerns about how ADM systems affect self-determination, caring relationships and identity recognition, and about the political–economic drivers of automation. The article contributes to AI ethics by empirically demonstrating that justice and ethics depend on the social relationships valued in different contexts and what is at stake, both personally and politically, in decisions aided by automation”.

[72] (p. 4473)

The numbers for the different “for good” phrases we list above are already very low. This fits with an earlier study that highlighted that “social good” and “for good” covering AI discourses were not engaging with disabled people [3].

Ability judgment terms are an essential lens to discuss social issues disabled people face. So how are ability judgment concepts used in relation to AI and disabled people?

1.3.3. Ability Judgment Terms and AI in Academic Literature

There is an increasing presence of ability judgment-based terms, such as ableism, ableist, disablism, and disablist, in abstracts of academic databases that cover AI. Searching 4327 abstracts containing the terms ableism or disablism or ableist or disablist, obtained from the academic databases Scopus, the 80 databases of EBSCO-Host and Web of Science on 25 December 2025, the term “AI” generated 122 hits, the term “artificial intelligence” 22 hits, and the term “machine learning” 10 hits [17]. Redoing the searches on 3 September 2026, for this study, the numbers of abstract hits for the three databases that contained the terms ableism or disablism or ableist or disablist in the abstracts were for AI 72/73/45; for “artificial intelligence” 53/66/31; “machine learning” 6/5/8; “chatbot*” 1/1/1 (all same article); LLM 6/4/2.

One covering “automated decision-making (ADM) systems and forms of artificial intelligence (AI) that learn to make predictions based on past data with little human intervention” laid out that “data infrastructures for classification and governance in the digital era has roots in eugenic data practices and ideas” [75] (p. 524) and “how automated systems perpetuate hierarchical divisions originally shaped by ableist eugenic race science” [75] (p. 523). Another one noted the danger of “automating ableism” [76] and that although AI can promote accessibility, it can also promote the ‘fixing’ of disabled people using arguments of “the eugenics movement” [77] (p. 1).

They list the following harms:

“1. Representational: stereotyping, demeaning, erasing, alienating, denying, reifying. Malice as modernization: Twenty first-century eugenics, AI, Project 2025, and disability. 2. Allocative: opportunity loss, economic loss. 3. Quality of service: alienation, increased labor, forced or rushed adoption. 4. Interpersonal: loss of agency or control, technology facilitated violence, diminished health and wellbeing, privacy violations, inability to verify output. 5. Societal harms: representation, culture, political, civic, macro socio-economic, legitimation of medical model (Shelby, et al., 2023; Wang, et al., 2025)”.

[77] (p. 5)

And they refer to others as follows:

“For disabled people, the specific negative impacts in gen AI include a revival of eugenics-based perceptions of disability (van Toorn and Soldatić, 2024), treatment of disabled people as social and societal outsiders (Trewin, 2018), new forms of surveillance (Nagy, 2024), loss of privacy (Berridge and Grigorovich 2022; Kamikubo, et al., 2023), misrecognition (Scully, et al., 2025), misdiagnosis (Keyes, 2020), and denial of benefits (van Toorn and Scully, 2023; van Toorn, et al., 2025), among alarmingly numerous other identifiable harms”.

(Wang, et al., 2025) [77] (p. 5)

It is also reported that AI models cannot identify ableism in a way that is needed [78] and that artificial intelligence decision-making processes “fail to reflect the lived experiences of individuals”, with the effect of creating “an ableist culture” and the “resurrection of eugenics-type discourses” [79] (p. 1). One study has a section on AI and eugenics [8].

One outlined problems with AI-based moderation of social media to deal with and identified ableist hate speech and has a section titled “AI and ableism” [80], on this topic see also [81].

One used the concept of “cripping AI” to engage with ableist judgments of AI [82], another flagged the biased ability judgment exhibited by electronic health records and clinical AI training [83]. Techno-ableism is often used to discuss problematic aspects of AI, e.g., [84].

One study found that “while many students recognized racism and classism at play in the potential negative impacts of AI systems, most students failed to recognize ableism as a driving social force for inequity” [85] (p. 67).

In one study, the authors outline various reasons why AI harms disproportionately impact disabled people:

“The harms caused by large language models and other AGI-driven systems disproportionately impact disabled people through various mechanisms, including inaccessible outputs, the perpetuation of stereotypes, and environmental effects that exacerbate existing oppression. For example, the bias in training data and algorithms can result in AI systems that fail to recognize or accommodate diverse communication styles, further marginalising neurodivergent people. Moreover, the resource-intensive nature of AGI development contributes to environmental degradation, which often disproportionately affects disabled communities that may be more vulnerable to the impacts of climate change. An intersectional lens reveals how these harms compound for disabled BIPOC (black, indigenous, and other people of colour) people, who face multiple, interlocking systems of oppression. The technosolutionist approach to AGI development, with its emphasis on efficiency and standardization, risks deepening these inequities by disregarding the complex, intersectional realities of marginalized communities”.

[8] (p. 6)

Four recent books cover many issues in relation to disabled people, disability justice, and AI and ability judgment concepts and AI [70,86,87,88].

AI, Ability Judgments, and Disability Justice

Searching Scopus, the 80 databases of EBSCO-Host and Web of Science on 3 September 2026 for abstracts that contained (ableism OR disablism OR ableist OR disablist) and “disability justice” generated 131/359/77 hits (duplicates not removed), highlighting the importance of this concept to discuss ability judgments. Adding “AI” OR “artificial intelligence” to the search led to 13/6/3 hits, suggesting an underuse of linking AI, ability judgments and disability justice.

This lack is, for example, noted in [89], and it is argued that a disability justice framework is needed to counter the production of ableism by AI [90], to challenge data ableism and techno-ableism [90], to question the medicalization effect of AI [90], to deal with AI use in war technology and its impact on disabled people [90] and to deal with “data injustice, medical colonialism, and the exclusion, violence, and erasure that shape the everyday experience of disabled people” [90] (p. 222). One study examined the term “sociotechnical ableism”, analyzing “AI tools such as chatbots, learning analytics, robotic tutors, and intelligent tutoring systems [8] (p. 1). One study outlined their visions and actions in order to achieve disability justice within AI by the four authors with the four themes: shaping intersectional disability justice systems through leadership and culture; improving disability data; reimagining medical records to reshape AI; synthesizing cross-disciplinary methods for disability-inclusive AI [76].

AI, Ability Judgments, and AI Ethics, Responsible AI, AI for Good, AI for Social Good, and AI Justice

“AI-ethics” searched together with the ability judgment concepts of “ableism” OR “disablism” OR “ableist” OR “disablist” has been shown to be rarely covered in the academic literature [17]. Performing the same abstract searches on 3 September 2026, for “ableism” OR “disablism” OR “ableist” OR “disablist” but replacing “AI ethics” with “AI justice” led to zero hits, with “responsible AI” to one hit, and with “AI for social good” OR “AI for common good” OR “chatbots for social good” OR “ChatGPT for Good” OR “AI for good” OR “IT for good” to zero hits (Scopus and Web of Science only, EBSCO not used as explained earlier).

Within this broader area, chatbots/AI writing assistants are one subarea of AI tools with special implications for disabled people and society at large.

1.4. Chatbots and Critical Thinking

Our search of Scopus on 12 June 2026 on AI and critical thinking generated 2700 hits. A total of 186 abstracts contained the terms “critical thinking” and “chatbot*”, indicating that critical thinking is one main topic of discussion around chatbots. Chatbots are seen to decrease the feeling of ownership and recall of data [91]. The homogenization of human perspective and reasoning and the decrease in cognitive diversity and collective intelligence are also flagged [92,93]. It is argued that we see an “epistemic collapse: a situation where diversity in knowledge, reasoning, or interpretation is lost, leading to uniform ways of thinking and a narrower collective understanding” [93] (p. 2). This makes sense given the biased judgments by AI tools due to algorithmic bias [8,9,10,11,12,13], whereby one reason is that “people providing biased training data, prompts, or feedback during the training process when creating these applications” [14] (p. 1). Another source outlined challenges found in the literature that all point to the importance of critical thinking:

“training data affects user learning and development; exhibiting of preferences for certain ethnic and gender groups or political views; a lack of the creativity and originality found in humans; the fabrication of data or results; the manipulation, alteration, or omission of data or results so that they do not accurately represent the research; hallucinations (text that seems semantically or grammatically reasonable but incorrect or meaningless); inaccuracies in generating citations; inaccurate facts or flawed logic; unreliable and poorly informed responses; outputs that merely copy or produce a combination of the training data; and socially prejudice, discriminatory, aggressive, or insulting answers”.

[94] (p. 9)

One recent study looked at the issue of artificial intelligence (AI) writing assistants making autocomplete suggestions to people as they write text [14]. It had as a premise that AI writing assistants can provide biased suggestions [14]. It furthermore argued that “as AI tools become more politicized, users may select assistants that align with their views” [14] (p. 8). The study found the following:

“(a) the attitudes participants expressed in a post task survey converged toward the AI’s position, (b) a majority of participants were unaware of the AI suggestions’ bias and their influence, (c) the influence of the AI writing assistant was stronger than the influence of similar suggestions presented as static text and (d) warning participants about assistants’ bias before or after exposure does not mitigate the attitude-shift effect”.

[14] (p. 1)

The “critical thinking in AI use scale” introduced in 2026 [95] defines critical thinking in relation to AI as “reflective thinking about what to believe or do when interacting with AI systems and AI-generated outputs, adapted from Ennis’ (1985, 2011) foundational definition of critical thinking”, and “it involves deliberately evaluating AI-generated content to make informed and responsible decisions” [95] (p. 2). Within their “critical thinking in AI use scale”, they measured “Verification (scrutinising and corroborating AI outputs), Motivation to understand AI (seeking to learn how AI systems work and where they fail), and Reflection (considering ethical, societal, and long-term implications of AI use)” [95] (p. 17). They also warned that using AI tools for verification can increase the problem of the negative impact on critical thinking [95].

Another recent article proposes five stages of AI literacy: stage 0—Not Yet Engaged; Stage 1—Uncritical Use; Stage 2—Informed Use; Stage 3—Critical Evaluation; Stage 4—Improvement [96]. At stage 2 they use the following indicators to judge users, “selects suitable AI applications”, “verifies facts and sources”, “recognizes hallucinations and bias”, and “documents AI use”. At stage 3, “ evaluates assumptions and uncertainties”, “assesses bias and disciplinary fit”, “considers consequences of error” and “critiques strengths, and limitations”. At stage 4, “ designs improved prompts and workflows”, “develops evaluation protocols”, “curates datasets”, and “builds/customizes AI systems” [96] (p. 8)

With respect to disabled people, a key question is whether the user has access to data on the lived reality of disabled people or knowledge of that reality through personal relationships that would allow them to successfully fulfill the requirements measured by the “critical thinking in AI use scale” [95] and to act on literacy stages 2–4 of [96].

1.5. Chatbots and Disabled People

As part of a teaching tool on chatbots that was given to students at the beginning of a fall 2025 course, a review was performed that looked at how disabled people were covered in relation to chatbots in the academic databases Scopus, the 80 databases of EBSCO-HOST, Web of Science, and ACM/IEE (journal) [97]. The review found 182 relevant abstracts out of 35,023 abstracts covering chatbots. The main themes within the 182 relevant abstracts were medical issues, accessibility of chatbots, and chatbots of use to disabled people [97]. The teaching document [97] notes that only four abstracts mentioned biased output of chatbots related to disabled people [98,99,100,101]. The problematic use of ability judgments was mentioned only twice [8,101] and “not one abstract covered the topic of literacy gained or not gained about disabled people by using the chatbots” [97] (p. 1), which the teaching document flagged as a gap in the academic literature [97] (p. 1).

One source identified “how narratives generated by ChatGPT reflect and reinforce societal biases, in the context of disability, ableism, and disablism” [102] (p. 7), which they argued enabled critical literacy on the topic (see also [103]). Another study noted “Gadiraju, et al. (2023) found LLM chatbots were reliant on harmful stereotypes and narratives about disability in their outputs, fixating on physical disabilities (especially wheelchair users), and created narratives that put disabled people in helpless positions, fell victim to inspiration porn, or described them as wanting to be fixed”. And then the authors gave examples where them using Chat GPT5 exhibited the same problem [77] (p. 5). Then, they make the case that “the potential positives and negatives for gen AI and disability exist within a much larger set of policy concerns for disabled people” [77] (p. 8).

1.6. Chatbots, Critical Thinking, and Literacy About Disabled People

The ability for critical thinking is an important issue in relation to disabled people given the many biases produced about them by various sources [2,3,4,104,105,106,107,108]. Representational harm is flagged as one harm caused by chatbots, whereby the content the chatbots provide amplifies existing bias instead of improving the literacy of the chatbot user [106]. As previously mentioned, a related concern is the potential of chatbots to decrease literacy about disabled people [15].

The danger of decreasing literacy about the lived reality of disabled people makes sense, as chatbots make use of existing data, and the biases in the literature (bias as in a) what data is available and b) what imagery of disabled people is prevalent) will be reflected in their output, which in turn will have an impact on what users think the lived reality of disabled people is. In one study, it is proposed that one should ask the following six questions of chatbot outputs:

“(1) How well does the statement accurately represent people’s lived experiences? (2) How objectifying is the statement towards disability (e.g., inspiration porn)? (3) How disregarding or marginalizing is the statement of one’s granular identity? (4) How assumptive is the statement? (5) How violating is the statement of one’s autonomy? (6) How discouraging is the statement towards one’s abilities?”.

[106] (p. 212)

The low quality of data, coupled with a decrease in critical thinking, epistemic collapse, and the decrease in literacy about the lived reality of disabled people, has many implications for disabled people and their allies. The low societal literacy on their lived reality, the need to constantly educate others who lack knowledge of even long-standing problems, and the attitudinal inaccessibility allies face when raising an issue others simply did not recognize as a problem are already flagged as a burnout problem for disabled people and their allies involved in activism [27]. The results described in [14], and the discussions around the biases and negative impact on critical thinking of chatbots, highlight the danger of a further decrease in literacy, which will lead to a greater impact on activism by disabled people and their allies due to increasing disabled activist burnout [28] and ally burnout [27]. It also might lead to a decrease in the number of allies, as people would not realize that there is a problem [27].

To summarize, disabled people face many problems in relation to AI systems, such as lack of production of knowledge about disabled people in the academic literature, content and judgment biases produced by AI tools, such as chatbots and AI writing assistants, problems with the usability of the AI tools, and the lack of involvement of disabled people in the design of AI tools. The lack and/or biased data on the lived reality of disabled people is of particular importance. Having data on the lived reality of disabled people is important for people to become literate in that reality and to engage in critical thinking in relation to disabled people. Having quality data is important for people to be allies, active citizens, and change agents for disabled people and for disabled people to be self-advocates. Being literate is also needed to implement AI ethics, responsible AI, AI for social good, and AI justice for disabled people, and to prevent further adding to (and not rectifying) the negative lived reality of disabled people. Few studies indicate the danger of a decrease in literacy on the lived reality of disabled people [15,97,106]. We could find no study that asked participants questions about this aspect. Our survey asked about the impact of AI tools, including chatbots. That aspect is not new. However, the novelty of our study lies in the parts that specifically asked questions about the content of chatbots and its impact on the literacy about the lived reality of disabled people, the impact of the content on essential activities such as activism by disabled people and their allies, and the impact on ability judgments, such as ableism and disablism.

2. Method

2.1. Study Design

This study was conducted in accordance with the ethics approval granted by the Institutional Review Board of the University of Calgary (protocol code REB24-1606 and 31 January 2025). It examines the perspectives of undergraduate critical disability studies students from one Canadian university on the topic of AI tools and chatbots. Our study can be classified as an online qualitative survey [109], as open-ended questions were used mostly to obtain data on the novel aspect of this study, which focused on the views of participants on AI tools and chatbots in relation to disabled people. The online qualitative survey was distributed via the Canadian university’s Qualtrics online platform.

2.2. Analytical Framework

Given that the participants are disability studies students, we interpret our findings through the epistemological framework (Nothing about us without us) and other premises of the field of critical disability studies [25,110,111,112,113], which investigates the social, lived experience of disabled people/people with disabilities. For one of her last musings about the field, see the views of the late critical disability studies scholar from India, Anita Ghai, a pioneer in disability studies/critical disability studies and the role of disabled women [114]. We will also draw from disability justice, which introduced a distinctly intersectional lens [59,60,61,62,63,64] to disability studies.

To interpret our findings, we also draw on some of the 50 ability judgment-based concept [20] such as disablism, the “discriminatory, oppressive or abusive behaviour arising from the belief that disabled people are inferior to others (although you won’t find a definition in a dictionary)” [115] (p. 9), coined within the disability rights movement, the field of disability studies [17,116,117,118], and the three strands of ability-based studies (ability expectation and ableism studies, short ability studies [18,19,20], studies in ableism [21,22,23], and critical studies of ableism [24,25]), which examine the role of ability-based expectations, judgments, norms, and conflicts.

We especially make use of the concepts of ability expectation and ableism justice and injustice (also called ability judgment justice and injustice, or ability justice and injustice for short) [17,20,97,119,120], which we argue are an essential part of the justice components mentioned in many documents linked to AI governance [34,49,121,122,123], including AI justice [55,56,57,58] and the justice principle of biomedical ethics [124], which is seen to be a main foundation of AI ethics [125].

2.3. Source of Participants

From our searches online, there seem to be no surveys that explicitly look at the questions of the impact of AI use, especially chatbots, on the literacy of people on the lived reality of disabled people, the utility of chatbots for activism, and the impact of chatbots on ability judgments, such as ableism and disablism.

As a first step to fill this gap, we report here results on the views of undergraduate critical disability studies students on these topics. This study was not about evaluating the impact of the class lecture on how the students would answer the survey. We did not give the survey before and after the topic was taught. We did choose this group as our first one because disability studies courses are based on furthering social justice for disabled people [126] and enhancing the lived reality of disabled people in all their facets [111]. Students in general are seen as change agents [127,128,129,130,131,132,133,134], and it is assumed that students joining a disability studies degree are concerned about the ability of disabled people to have a good life. Disability studies students are further assumed to (1) engage with the social problems disabled people face in order to make a positive difference in their lives [135], (2) see themselves as allies of disabled people [29], and (3) want to raise awareness about the problematic lived reality of disabled people [30].

Many expectations are attached to allyship, such as “supporting social change [12], lifelong learning [49,60], not being defined by the allyship [49], making use of one’s privilege [49,55], being aware of general privilege [12,60], giving voice to the marginalized [49,55,56,60,61,62], considering the power in language [61], understanding one’s positionality [12,61], doing research [55,61,62], and taking action [55,61,62,63]” [26] (p. 3) All the expectations from allies in general are also expected from allies of disabled people [26], and there are expectations that specifically use the ability judgment language, such as ableism and disablism, “grounded in collaboration with persons experiencing disability or disablism” [136] (p. 67) cited in [26] (p. 4), or “to develop a nuanced perspective on intersectional ableism and allyship [137,138], to bring internalized ableism and privilege to light [137,138], to be aware of ability privileges [139], dismantle ableism [137], challenge ableist practices [137]” cited in [26] (p. 3).

The literature highlights many causes of activist burnout, such as “internal and external to the activist and problems within and between movements [6,135,136]” [28] (p. 4). As allies ought to be activists [26], these factors can be seen to also lead to ally burnout. Many factors leading to the burnout of allies of disabled people are reported, such as the low level of literacy of society at large in relation to the social situation of disabled people, but also not being taken seriously by others (attitudinal inaccessibility), lack of support, and no changes happening [27]. Many of these ally burnout factors reported can be seen to be linked to the low level of literacy on the social reality of disabled people by society at large. Given the reality, such as the burnout allies face and the expectations of them, the questions we asked our potential allies of disabled people (critical disability studies students), especially the literacy aspect, are relevant to them and to disabled people.

Knowing the answers of participants to the questions asked might be useful to tailor educational material in disability studies classes but also in classes that cover social implications of AI/ML and STEM education. The answers might also be important beyond education, such as policy making, science and technology governance, AI ethics, AI justice, responsible AI, and AI for good discussions.

2.4. Survey: Question Development

The questions we present here are a subset of questions asked of students as assignments throughout the course. The questions were piloted with, and feedback on the wording of the questions was obtained from, a small sample of disabled and non-disabled students from within and outside of the critical disability studies degree within the university. These students were known to the authors for some time and were judged by the authors to be seen as allies of disabled people. Furthermore, these students saw themselves as allies of disabled people and others. Questions were slightly modified, and the final version of the questions, which are listed in Table 1, was used. The questions (with the exception of question 7) on purpose did not ask for different types of body/mind characteristics that put one in the “disability characteristic”. Disabled people are so diverse; to cover every possible variation, including intersectionalities and intrasectionalities, would have made such a survey very unwieldy, especially if this would have been expected also for the open-ended questions.

Table 1. Themes linked to the survey questions.

Theme	Sub-Theme	Questions Focused on AI Tools	Questions Focused on Chatbots
Accessibility: Physical/digital			Q7—To what extent do you feel chatbots are accessible to people with a range of ability differences from the norm? Q8—Do you think disabled people have been excluded or faced barriers when using a chatbot due to the due to their divergence from an ability norm? Q9—Do chatbot platforms offer meaningful customization options to support a diversity of communication needs (e.g., plain language, screen reader compatibility, emotional tone adjustment)?
Accessibility/impact: Attitudinal		Q2—To what extent do you feel AI tools support respectful and fair treatment of disabled people? Q5—How concerned are you about the following AI-related issues?
Knowledge	Whose knowledge/whose voices	Q4—Do you believe disabled people were consulted in the development or deployment of AI systems you use?	Q10—How often do chatbots reinforce design assumptions that center non-disabled users, so users who fit the ability norm? Q17—When you ask chatbots disability-related questions, do you think they center disabled voices and experiences? Q19—Do you believe disabled people’s knowledge, activism, and scholarship are adequately represented in chatbot outputs?
	What knowledge	Q3—Do you feel AI-based systems (e.g., for hiring, grading, scheduling) fairly reflect the needs and experiences of disabled people?	Q15—Do you feel chatbots accurately reflect the lived experiences of disabled people? Q16—When you ask chatbots disability-related questions, do you think they rely on medical or deficit-based models of describing the disabled person? Q18—Have you encountered examples where a chatbot reinforced harmful stereotypes or outdated terminology about disabled people? Q20—In your experience, do chatbots reflect a diversity of identities of disabled people, including intersectional perspectives (e.g., disability and race, gender, poverty, etc.)?
Literacy	Importance of data on the lived reality of disabled people		Q12—In your opinion, how important is it to collect data on the lived experience of disabled people?
	Impact of lack of data on the lived reality of disabled people		Q13—What do you think are the practical impacts of not having data on the lived reality of disabled people? Open-ended question
	Usefulness of knowledge produced by chatbots for advocacy purposes		Q21—How effective is the use of chatbots to obtain information so one can advocate for the rights of disabled people?
Ableism/disablism/ability judgment	Chatbot output impact on ableism/disablism and ability judgments		Q22—Do you think chatbot answers reinforce or disrupt ableist power structures?
	How to improve the impact of chatbots on ableism and disablism		Q23—How can chatbots be used to challenge ableism and disablism rather than reinforce it?

2.5. Data Collection

From September to December 2025, weekly survey questions were given via the university Qualtrics online platform to students enrolled in a junior-level disability studies course. These questions were related to the weekly topics of the class, and students were required to submit their responses by the Friday before the following Monday’s lecture for any given week.

The survey question results used in this study were linked to the week when we engaged with artificial intelligence. The survey answers we report in this article reflect the views of the students before they had access to the lecture material covering AI. As to the topic of this paper, we think, based on our knowledge of the degree program, that the students were not exposed to the topic of the social implications of AI and chatbots before the class in other courses within the degree they were in. We are also not aware of other courses outside the disability studies program that cover AI and disabled people in any depth.

In general, the instructor generated a report of the answers in Qualtrics every week and incorporated the answers into the lecture on that topic without identifying a given student, to stimulate discussions on the weekly topic in the class.

Students completed the questionnaire as part of a standard pass/fail marked course assignment (fail if not submitted). Students were expected to answer all questions; however, the system did not force them to do so. But students had to click through every question before they could click the submit button. The instructor could check if a student did not answer questions. In compliance with the ethics approval and procedures developed together with the institutional ethics board (IRB number to be added in case of acceptance), students saw this as a normal course assignment during the time they did the survey. After the final marks for all course members were added into the university grading platform by the instructor and the grades were approved by the university, the students were asked by email whether they would agree that the data be used without attributing the data to any person for a potential publication. Only a PDF of the data generated by the Qualtrics software, which had no names attached to any answers, was used for the analysis. The survey answers of any student who would not have agreed would have been removed from the dataset by the instructor before the analysis. None asked to be removed.

2.6. Data Analysis

As to our online survey, basic quantitative descriptive percent (%) data were extracted and analyzed using Qualtrics’s intrinsic frequency distribution analysis capability. The qualitative results of the open-ended questions were downloaded as one PDF with no mention of students’ names. We performed a directed thematic analysis [140,141] following the six-phase thematic analysis process outlined by Braun and Clarke [141].

As to trustworthiness measures for qualitative data [142,143,144]; as to confidence, this was a class assignment, and the instructor was deeply engaged with participants, and peer debriefing between the authors was employed. Differences in codes and theme suggestions of the qualitative data were few and discussed between the authors and revised as needed [143]. Confirmability is evident in the audit trail made possible by the comment function in Adobe Acrobat software [145]. As for transferability, the description of our method gives all the required information for others to decide whether they want to use our qualitative online survey in other settings. There are many other groups one can ask, such as disability studies students from other universities, especially in the Global South and from different stages of the degree. As disabled people need allies from all walks of life and areas of expertise, one should also ask the questions of potential allies, such as students in engineering, computer, and data science degrees and people involved in the areas of AI ethics, responsible AI, and AI for good.

2.7. Limitations

This study has several limitations. Students were not given the survey questions as a document beforehand and could not seek clarifications in real time as one could do in an in person/virtual interview. However, a student could have paused the survey and emailed the instructor in case they did not understand a question they were to answer. No email was sent.

Additionally, we did not collect demographic data, including disability data, because the survey was part of a standard graded assignment aimed at fostering engagement with course topics. Asking demographic questions might have disrupted the sense of anonymity within the group, as students might have been able to attach certain answers to certain subgroups that are so small that one could make an educated guess as to who the student was. This would have been detrimental to using the survey answers to stimulate discussions on the weekly topic in the class. Also, it would mean that the instructor would know who in the class identifies as a disabled person. As most students who fall under the disability category fall into the invisible disability category, many of whom do not identify themselves to others as disabled (e.g., many do not apply for accommodations), asking that question would have been problematic and not approved for use in the teaching. As for writing a publication, with such data, this would not only have gone against our ethics approval due to the danger of being identifiable by readers from this class but might have also potentially discouraged students from consenting to the survey to be used.

Given the purposive nature of this course, the results of this study are not meant to be generalized to the perspectives of all students, even within disability studies programs and within a given disability studies program. However, the questions provided should allow others to perform studies focusing on other groups, including other disability studies student cohorts from different years in their program and different geographical locations, students from other programs, and participants who are not students. Once more studies are done, one might be able to generate a meta-analysis on this topic based on answers given in various studies.

Social desirability is a possibility all the time. Students might assume the instructor wants to hear certain answers based on prior lectures of the instructor in this disability studies course. This social desirability might be a possibility for the Yes/No questions. However, it would be very hard for the students to assume what qualitative answers the instructor wants to see.

3. Results

In the result section we provide all tables (Table 2, Table 3, Table 4, Table 5, Table 6, Table 7, Table 8, Table 9, Table 10, Table 11, Table 12, Table 13, Table 14, Table 15, Table 16, Table 17, Table 18, Table 19 and Table 20) reflecting all the answers to the questions of our surveys we reported on.

Table 2. Q2—To what extent do you feel AI tools support respectful and fair treatment of disabled people?

Not at All	To a Small Extent	To a Moderate Extent	To a Great Extent	Completely	Total
2	26	34	5	2	69

Table 3. Q3—Do you feel AI-based systems (e.g., for hiring, grading, scheduling) fairly reflect the needs and experiences of disabled people?

Yes	No
25.88% (n = 22)	74.12% (n = 63)

Table 4. Q4—Do you believe disabled people were consulted in the development or deployment of AI systems you use? (n = 89 responses).

Yes	No	Not Sure
3.30% (n = 3)	49.45% (n = 45)	45.05% (n = 41)

Table 5. Q5—How concerned are you about the following AI-related issues?

#	Question	Not Concerned		Somewhat Concerned		Moderately Concerned		Very Concerned		Extremely Concerned		Total
1	Algorithmic decisions that exclude or harm disabled people	0.00%	0	2.35%	2	14.12%	12	23.53%	20	60.00%	51	85
2	AI tools designed without consultation with disabled people	0.00%	0	3.53%	3	16.47%	14	36.47%	31	43.53%	37	85
3	AI reducing human empathy in processes such as HR, education, or healthcare	0.00%	0	0.00%	0	5.88%	5	10.59%	9	83.53%	71	85

Table 6. Q7—To what extent do you feel chatbots are accessible to people with a range of ability differences from the norm?

#	Question	Not at All Accessible (Completely Inaccessible)		Slightly Accessible (Significant Barriers Exist)		Moderately Accessible (Some Barriers Exist)		Very Accessible (Minor Barriers Exist)		Fully Accessible (Barrier-Free/No Barriers		Total
1	Sensory blind	28.41%	25	36.36%	32	27.27%	24	6.82%	6	1.14%	1	88
2	auditory deaf	3.41%	3	11.36%	10	42.05%	37	34.09%	30	9.09%	8	88
3	Cognitive	7.14%	6	27.38%	23	47.62%	40	13.10%	11	4.76%	4	84
	Mobility	8.05%	7	20.69%	18	24.14%	21	34.48%	30	12.64%	11	87
	Psychiatric	6.02%	5	27.71%	23	42.17%	35	22.89%	19	1.20%	1	83

Table 7. Q8—Do you think disabled people have been excluded or faced barriers when using a chatbot due to their divergence from an ability norm?

Yes	No	Not Sure
72.62% (n = 61)	2.38% (n = 2)	25.00% (n = 21)

Table 8. Q9—Do chatbot platforms offer meaningful customization options to support a diversity of communication needs (e.g., plain language, screen reader compatibility, emotional tone adjustment)?

Not at All	Small Extent	Moderate Extent	Large Extent	Fully
5.95% (n = 5)	32.14% (n = 27)	52.38% (n = 44)	9.52% (n = 8)	0% (0)

Table 9. Q10—How often do chatbots reinforce design assumptions that center non-disabled users, so users who fit the ability norm? [Scale: 1 = Never, 5 = Always].

Never	Rarely	Sometimes	Often	Always
9.41% (n = 8)	17.65% (n = 15)	21.18% (n = 30)	35.29% (n = 30)	16.47% (14)

Table 10. Q12—In your opinion, how important is it to collect data on the lived experience of disabled people?

Not Important	Slightly Important	Moderate Important	Very Important	Essential
0% (n = 0)	0% (n = 0)	7.06% (n = 6)	40.00% (n = 34)	52.94% (45)

Table 11. Q13—What do you think are the practical impacts of not having data on the lived reality of disabled people? Open-ended question.

Theme (Negative Impact on)	Frequency	Sub-Theme	Frequency	Relevant Quotes (Please Note, the Quotes Do Not Reflect the Full Quotes but the Part of the Quote Reflecting the Theme; Therefore, the Below Are Mostly Not Complete Sentences
Negative impact on		Disability awareness/pride	1	disability awareness and pride
	129	Services	6	Services and technologies which were designed that ignore their needs; services become entirely impractical and useless; service that fail to meet their needs; for example, there are more reports on male pattern baldness than there is reports on debilitating endometriosis, the effects of this have been plain denial of symptoms for patients, lack of non-invasive treatments to offer to patients, as well as a lack of awareness and empathy in the general public; here wouldn’t be specialized services to accommodate people with any disability. Such as specialized inclusive public transit, specialized classes, aids/assistant and much more; increasing amount of poorly designed services and tools available to disabled people.
		Technologies	2	Services and technologies which were designed that ignore their needs; makes it harder to develop different strategies and technologies that can further help disabled individuals if there input and experiences are not being considered.
		Barriers/inaccessibility	25	Unfair barriers; barriers stay hidden; inaccessibility around the community; inequitable environments; inequalities in healthcare, financially and socially; furthering the world in a direction that is not built for people who are “different” rather against them; excluding them; barriers that people with disabilities face become hidden to the outside world; very hard to make society and life more accessible for people with disabilities; create a barrier; limits the accessibility they can get; Without having data on how disabled people live, it can exclude a minority from accessing the simple things in life; end up with systems and programs that are not inclusive and do not provide fair and equal opportunity/access; make it harder to have a fair community and an inclusive education system or an inclusive government system; environments in different social settings are less likely to be inclusive to disabled people; creates a major barrier towards funding, considering that without specific data, inadequate funds would be provided to healthcare or disability services; products and environments may be designed in ways that exclude disabled people; it is impossible to get a complete and accurate picture of the barriers and challenges faced by those individuals; certain people may experience certain challenges or barriers, and people wouldn’t know what tools to add to help them succeed; we cannot get rid of barriers that marginalizes these individuals; inaccessible areas; creates an unjustified harsh environment for individuals with disabilities as they navigate through everyday life because the needs and accommodations aren’t being met; raiser barriers for people with disabilities; Reinforcing and creating more societal barriers for people with disabilities because less things will be designed with them in mind; burden on healthcare as disabled persons’ social access can be hindered.
		Decrease in support	16	No proper support of their daily life; inequalities in healthcare, financially and socially; inequitable environments; to push off issues; Fewer resources; fewer opportunities; indifference, and ignorance towards the disabled community; impact the support that needs to be received in certain areas. and to be aware of such issues so it can be discussed and dealt with; lack of empathy; forgetting and leaving people with disabilities behind struggling and not providing with the help they may need; inability to provide the resources they need; lack of support from others; will not have the proper support; increased inequality and inequity in areas such as employment and health (not to mention greater ostracism of disabled people); allows people with disabilities to be marginalized; lack of depth in the program; People with disabilities likely won’t feel their needs have been met.
		Excluded	4	They are exempted and excluded; to exclude them or marginalize them; When people are not educated on the lived experiences and struggles of disabled individuals in our society, these citizens go unnoticed, unacknowledged and excluded; disabled people may be unable to secure a job position, receive services from stores/restaurants, access upper floors of a building, and navigate independently to their desired destinations. Without the collection and education on the reality of disabled people, this enormous exclusion and denial of disabled people’s rights occurs.
		Policies	16	Policies miss the mark; bad decisions because when we create solutions to solve our problems, we base them off of the assumptions and stereotypes we hear from the internet rather than having the knowledge from someone who has experienced being disabled themselves; furthering the world in a direction that is not built for people who are “different” rather against them; harder to remove the stigma disabled people face; harder to make the world more accessible; Change in society would also not happen if nobody was able to provide the accessibility struggles they deal with; to push off issues; opportunities, city buildings, or offices might not be designed for individuals with disabilities; build a world that shut them out and leaves them stranded on their own; no changes or improvements will be made in society which will only make the day to day of people with disabilities harder than it has to be; make it harder to have a fair community and an inclusive education system or an inclusive government system; for example, there are more reports on male pattern baldness than there is reports on debilitating endometriosis -> the effects of this have been plain denial of symptoms for patients, lack of non-invasive treatments to offer to patients, as well as a lack of awareness and empathy in the general public; inability to provide the resources they need; how people chose to create certain things in society; makes it practically impossible to ever create accommodations as well as inclusivity in our world.
		AI systems	4	AI systems reinforce exclusion; AI can’t really extrapolate answers when there are questions asked about the life of disabled people; cause AI to become biased and not be inclusive and accessible to everyone; bias regarding views and social norms as AI might not challenge these norms.
		Perception of disabled people	22	Unrecognized for their own unique talents; ay only be seen for their disability alone; harder to remove the stigma disabled people face; bias and stereotypes about people with disabilities; bias; leading individuals to possibly feel like their identity is the disability they struggle with, which is not true; would never know how people with disabilities live their respective lives every day, the things that they need in order to perform regular tasks, or their mindsets; would never know how to properly offer them the support that they require; everyone’s disabilities are different, without these data, we could never come up with options that works best for all; hard to fit the experience of a disabled person into a box. People have different experiences and views with their disability and that effects how they live in many different ways; creation of negative public stigmas around disabled communities; would not properly represent people with disabilities and dismiss proper knowledge of accessibility; other might hold presumptions on how they live their lives basely on the insufficient data they’ve encountered that might lead to placing stereotypes on people with disabilities; decisions may be based on stereotypes rather than actual experience; Stereotypes and misconceptions would play a big role as the media shapes narratives with inaccurate information instead of representing the experiences of the disabled community; people who don’t live with a disability make assumptions; a large stigma surrounding disabilities. This stigma involves individuals with disabilities being represented as incapable, inferior, unworthy and inadequate; perceptions of those who are able bodied can create faulty systems and unrealistic standards when it comes designing thing; Inaccurate assumptions, and “understandings” would be made about people with disabilities; then able-bodied people will be able to assume the needs of those with a disability without actually understanding how they function. Leading to an inaccurate representation of what actually needs to happen to help those who are disabled; other people are unable to be educated on living with disabilities.
		Knowledge	20	Many people may not know what the affects are of certain disabled peoples experiences; raising the amount of people who lack the knowledge of disabled people; false assumptions would be made based on stereotype; creating room for stereotypes; to create marginalization and/or even harm through false assumption of what adaptions could be useful; unreliable data; educational facilities not having the proper tools to teach individuals; oblivious to the reality of disabled people; overlook their problems; aren’t able to fully answer in a way that it considers the full extent of disabled people; lack of understanding towards what certain types of disabled people need, mentally, physically and socially; lack of understanding regarding the systems optimal for their needs and experiences; lead to misunderstandings; dismiss proper knowledge of accessibility; lack of understanding among the able bodied and neurotypical populations; wouldn’t know how they feel about certain things or what we need to do as a society to make them more comfortable; people wouldn’t know what tools to add to help them succeed; for example, there are more reports on male pattern baldness than there is reports on debilitating endometriosis -> the effects of this have been plain denial of symptoms for patients, lack of non-invasive treatments to offer to patients, as well as a lack of awareness and empathy in the general public; difficult to know what accommodations are necessary to support their daily life; people don’t understand and never will understand if they are not taught or hear about the experiences of disabled people; could not have the facts, observations, and details provided to meet their needs; ignorance within society; there are a multitude of issues like social or physical barriers that interplay with those with disabilities that can be overlooked; the perspectives of the individuals aren’t being accounted for, making it more difficult for society to understand what they’re going through; causing social invisibility to issues that affects disabled people; other people are unable to be educated on living with disabilities.
		Participation	7	Lack of opportunities for disabled people to be active members in society; undermined by many people who do not understand or empathize with them; being so closed-minded about the needs and struggles of people with disabilities; silencing the voices of the disabled community; extremely discriminatory; They are not included when decisions are made that affect their way of life. The decisions which are made will only be based on a portion of the population considered as the “norm”.
		Advocacy	5	Harder for them to advocate for their rights; limits us from presenting the truth since we don’t have anything to use and work off of; it is hard to help adapt different parts of the community and help make things more suited towards them. if we don’t take in the data then it will add more onto a make things harder for them; would be unable to be helpful, and create accommodations accurate to what is needed by those with the disabilities; People such as parents who have become new parents to a child with a disability would struggle a lot more if there was no data or information on how people with certain disabilities live.
		Research	2	There is often new research and new theories, but it doesn’t matter if there isn’t lived data that proves that the impacts actually have a positive/negative effect on the individuals; for example, there are more reports on male pattern baldness than there is reports on debilitating endometriosis, the effects of this have been plain denial of symptoms for patients, lack of non-invasive treatments to offer to patients.
Moving backwards	2			Causes society to move backwards because it just lets us continue to live our lives while forgetting and leaving people with disabilities behind struggling and not providing with the help they may need; Individuals/groups that do not have data on the lived experiences of people with disabilities may tend to feed into societal norms.
Unchanged	3			Leaves society unchanged, with continued marginalization and a lack of accessibility for their needs; surrounding people won’t be able adjust to the needs of disabled people; Society is not able to improve the reality of disabled people because there is no data to base decisions/designs on.
Not fitting the question	4			Probably privacy? Collecting data from anyone may be invasive, and though it can be helpful, data in the wrong hands can be used maliciously; I think in regard to what is seen as practical, many companies, governments, and organizations avoid this data because they see it as a way to save time and money. If they ignore the problem, they take the easy way out and use less resources; however, this impacts the disabled community in a negative way. To truly demonstrate support for the community groups need to be willing to take the extra step in taking in the data and making the necessary changes so that everyone can reach their full potential; the world is created to work around the abilities and needs of those without disabilities which forces those who do not, and cannot fit into the mold to be marginalized and experience an unneeded amount of struggle to live and function in society. This can have lasting impacts on mental health for those individuals and their families and caregivers. A lack of effort put into creating an accessible world also results in extra pressure on their families/caregivers.
Perception of the world	1			How we see the world
Bias	1			Bias
Promote a mindset without empathy	1			Promote a mindset without empathy

Table 12. Q15—Do you feel chatbots accurately reflect the lived experiences of disabled people?

Never	Rarely	Sometimes	Often	Always
9.41% (n = 8)	62.35% (n = 53)	27.06% (n = 23)	1.18% (n = 1)	0% (0)

Table 13. Q16—When you ask chatbots disability-related questions, do you think they rely on medical or deficit-based models of describing the disabled person? 85.

Never	Rarely	Sometimes	Often	Always
1.18% (n = 1)	14.12% (n = 12)	43.53% (n = 37)	37.65% (n = 32)	3.53% (3)

Table 14. Q17—When you ask chatbots disability-related questions, do you think they center disabled voices and experiences? 85.

Never	Rarely	Sometimes	Often	Always
16.47% (n = 14)	55.29% (n = 47)	28.24% (n = 24)	0% (n = 0)	0% (0)

Table 15. Q18—Have you encountered examples where a chatbot reinforced harmful stereotypes or outdated terminology about disabled people? 76.

Never	Rarely	Sometimes	Often	Always
38.82% (n = 33)	28.24% (n = 24)	18.82% (n = 16)	3.53% (n = 3)	0% (0)

Table 16. Q19—Do you believe disabled people’s knowledge, activism, and scholarship are adequately represented in chatbot outputs? 85.

Yes	No
4.71% (n = 4)	95.29% (n = 81)

Table 17. Q20—In your experience, do chatbots reflect a diversity of identities of disabled people, including intersectional perspectives (e.g., disability and race, gender, poverty, etc.)?

Yes	No	Not Sure
7.04% (n = 5)	69.01% (n = 49)	28.00% (n = 21)

Table 18. Q21—How effective is the use of chatbots to obtain information so one can advocate for the rights of disabled people?

Not at All Effective	Slightly Effective	Moderately Effective	Mostly Effective	Extremely Effective
12.94% (n = 11)	52.94% (n = 45)	27.06% (n = 23)	4.71% (n = 4)	2.35% (2)

Table 19. Q22—Do you think chatbot answers reinforce or disrupt ableist power structures?

Disrupt	Reinforce
29.76% (n = 25)	64.29% (n = 54)

Table 20. Q23—How can chatbots be used to challenge ableism and disablism rather than reinforce it?

Theme		Frequency	Sub-Theme	Frequency	Relevant Quotes (Please Note, the Quotes Do Not Reflect the Full Quotes but the Part of the Quote Reflecting the Theme; Therefore, the Below Are Mostly Not Complete Sentences
Not sure/do not know		1			Direct word
Inclusion		1			Direct word only
Increase accessibility		16			Making them more accessible; improving performance in accessibility; offering customization; screen reading supports, adjustable text or audio and accessible features that include all varieties of disabled people; giving those who are differently abled opportunities to have increased access to support services; to make it more inclusive for more people such as by adding features such as speech-to-text making it more accessible for those who need/want to use it; should be more accessible to everyone which it currently is not because of those who run the program; Accessibility and the awareness should be the key focus when creating a tool that should be available for everyone; being accessible for everyone to use; through intentional programming, allowing disabled users to access the resource efficiently and safely. This may include features like text to speech, using simple words over complex, imaging over text, and allowing customization of the platform in order to better suit needs of people with disability’s; chatbots is built with the experiences and needs of the disabled community in mind; can be made in a way that they are inclusive to everyone’s needs making the online space more accessible; more accessible and actually considering the needs of people with disabilities; Making the platforms accessible and easy to use for all individuals with different needs; Including more inclusive designs with the help of input from disabled individuals and their lived in experiences.
Increase research by designers in relation to disabled people		2			Doing more research; with the right research and involvement, that it could very easily help get rid of derogatory terms and advocate for a more well rounded understanding society.
Improve knowledge by designers on disabled people		2			Forming more knowledge on disability; used resources made by people who are part of the disabled community.
Increase diversity of designers		2			Created with the influence of both people who do not have disabilities and people who do; incorporating people with disabilities into companies that work with enhancing chatbot responses.
Change bias of designers		2			Those who manage AI have an ableist mindset when creating new updates or changes in the software, disregarding others, such as disabled people, when running this program; chat bots need to be programmed differently and should be able to receive information from a more objective lens.
Training chatbots		13			The main issue enabling chatbots to reinforce ableism is through inexperience. AI chatbots are not designed to respond to a user with ability difficulties. These chatbots may be able to provide a user with history and knowledge defining disabilities and measures that can be taken when engaging with communities of people with disabilities, however, these systems struggle to curate systems that can best suit a disabled person and take into account their particular needs which may vary across different users; if programmed right can be used to educate people on disability’s;, therefore, becoming more inclusive towards people that have disabilities. Disablism can be challenge by chatbots in making it more accessible to people with disabilities such as it being customized to the disability of a person; through intentional programming, allowing disabled users to access the resource efficiently and safely. This may include features like text to speech, using simple words over complex, imaging over text, and allowing customization of the platform in order to better suit needs of people with disabilities; train the chatbots with literature and experiences of the disabled community; if programmed right can be used educate people on disabilities; can be trained not to use ableist language through reprogramming the chatbots; earning from trustworthy and proper sources that can actually properly inform people; flag misinformation and pull from peer reviewed and accredited resources to support it; Providing chatbots with information with perspectives of different people, and views. Providing views to challenge ableism, rather than consistent support and one sided views; having AI chatbots to be trained on prioritizing inclusivity, providing accessibility in getting information and challenge these systems to learn stereotypes and normalize that disability is a part of human diversity; instructed it to specifically render an inclusive design; If properly developed, chatbots can give based answers that aren’t pulled off of random medical data, but rather real life experiences. Chatbots can discourage hate instead of subconsciously reinforcing it. If it is unbiased, like it is supposed to be, it’ll allow people to use it as a legitimate tool to do research. It is difficult to develop such things because we, as people, are always biased in one way or another, and no one is truly neutral.
Change design	79		Designed to represent disabled people	2	Chatbot designed to represent individuals with disabilities; getting information about disability from respectable sources.
			Become bias free	1	Should have a bias free way of looking at things.
			Representation	1	It might not recognize those with disabilities as the norm when talking about scenarios involving situations with people.
			To educate users	16	Designed in a way that educates users; educate their users on the topic of disabilities; document lived in experiences by disabled people who can carefully explain why and how certain barriers exist against them and why certain things that they do may be reflected by their behaviour rather than someone who isn’t “disabled” assuming those things for them; focus on how not every individuals with disabilities will face the same challenges, as they are smart enough to learn things with the right technique and education; do not focus only on what the disabilities can enable; leading users to consider things beyond themselves and realize their way of life isn’t the only way of being; making chatbots more accessible to data that can inform people about the challenges those with disability face; educates users of chatbot while correcting false information or assumptions being made about them; as chatbots gain more access to issues regarding ableism and disablism, I would say they can inform people more about these issues, which could start conversations and hopefully bring awareness to try and stop that structure; can be used to educate others on the experiences of other people. But in order for that to happen, chat bots need to be programmed differently and should be able to receive information from a more objective lens; if we put input the voices and experiences of people with disability we can use that to challenge ableism and disablism rather than reinforcing it. This way the chat bot will be informed and can defend or educate others, eventually leading to a more fair society; by educating users on the topic and providing a clearer insight on the topic; chatbots can work on defining terms that often have a misconception and teach individuals about appropriate language, such as the difference between person-first language and identity-first language; AI can teach people about their [disabled people’s] experiences to fully understand what they go through sometimes; People can ask chatbots on ways to make a change and impact towards society, and can provide answers or knowledge; can provide people on ways to challenge ableism and how they can act on it; to spread awareness and break down barriers; to present information that explains misconceptions of disabilities and why those misconceptions are false; Chatbots can be used to challenge ableism by educating oneself about disabilities that people with disabilities have.
			Promoting awareness towards accessibility and inclusivity	2	Promoting awareness towards accessibility and inclusivity; promoting the importance of accommodations that promote inclusivity in the environment can help raise awareness.
			Promotes fairness, respect, and understanding	3	Promotes fairness, respect and understanding; if we put input the voices and experiences of people with disability we can use that to challenge ableism and disablism rather than reinforcing it. This way the chat bot will be informed and can defend or educate others, eventually leading to a more fair society; can be reprogrammed using the feedback of those with disabilities so that a more inclusive and fair solution is created.
			Change language	10	Uses language that avoids any stereotypes; including identity-affirming language; inform their user to start promoting inclusion, and empowerment for disabled people rather than reinforcing stereotypes or barriers in society; Chatbots often do not put person-first and use incorrect terms about certain disabilities. This leads to users not knowing how to properly address people with disabilities and get false information from these systems; By using the right terminology and using it when responding to topics regarding disability; being respectful and aware of the language it uses; using respectful language; by using inclusive language; use proper, appropriate and respectful terms in referring to people with disabilities; adjusting language used.
			Designed with empathy, inclusion, and transparency	1	Designed with empathy, inclusion, and transparency.
			Centering voices of disabled people/people with disabilities	26	Centering disabled voices; including the voices of disabled people; amplifying disabled voices; opinion of disabled users is more present, the output will align more with their opinions as well; these models should also be overlooked [looked at] by individuals with disabilities to help ensure the product is working to the best of its ability used resources made by people who are part of the disabled community; need to be generated with real life stories of people in communities like the disabled community. With that said, it does not mean it will completely disrupt but it is one way of encouraging and shedding light on the perspectives of people with disabilities and how they are affected in society; asking for their options and real life experiences so they can gain more real information; chatbots if programmed right can be used to educate people on disability’s; with the right program and input from individuals with disabilities they can become an effective tool for all; if we put input the voices and experiences of people with disability we can use that to challenge ableism and disablism rather than reinforcing it; can be reprogrammed using the feedback of those with disabilities so that a more inclusive and fair solution is created; If the facility who develop chatbots allow people with disabilities to share their lived experiences so they can reinforce it with real data or from real life experiences; train the chatbots with literature and experiences of the disabled community; chatbots can expose useful resources that are linked with experiences lived by disabled individuals such as blogs, articles, documentaries, and other resources where people with disabilities are able to share their perspective rather than projecting nondisabled perspectives; promoting the importance of accommodations that promote inclusivity in the environment can help raise awareness; centering disabled voices more and advocate for them; created with disabled individuals in mind; should have data from people who live with disabilities. So the chatbots can think of their needs, accessibility and more ideas so the people with disabilities can thrive; listening to the lived experiences of people with disabilities and respecting and accepting their experiences; To allow those with lived experience to have a hand in programming and updating their software’s. ensuring reliable data and programs are being used to give good information; Including more inclusive designs with the help of input from disabled individuals and their lived in experiences; Designing it using voices of those with disabilities and inputting multiple opinions and experiences; Developed by people with disabilities; Using the experiences and perspectives of disabled individuals; adding data from people with a disability into the AI’s algorithm; use scientific resources or resources form actual disabled people rather than the entire internet as a whole.
			Design that supports advocacy	2	Supporting advocacy; They can advocate multiple perspectives when giving answers. Having a well rounded focus would help; centering disabled voices more and advocate for them.
			Generate new data	7	Giving a new perspective practically, that is by getting new data to create new information thus so forth. Meaning groups and compagnies to create awareness in order to create this as a priority; need to be generated with real life stories of people in communities like the disabled community. With that said, it does not mean it will completely disrupt but it is one way of encouraging and shedding light on the perspectives of people with disabilities and how they are affected in society; asking for their options and real life experiences so they can gain more real information; more data is given to generative models, it become a tool to use towards spreading awareness about the lived experiences people with disabilities have on an everyday basis; If the facility who develop chatbots allow people with disabilities to share their lived experiences so they can reinforce it with real data or from real life experiences; Providing chatbots with information with perspectives of different people, and views. Providing views to challenge ableism, rather than consistent support and one sided views; to present information that explains misconceptions of disabilities and why those misconceptions are false.
			Comparing data	1	Pulling data from multiple sources to compare and challenge ideas.
			Be vague	1	By providing a very vague understanding of the negative impacts it has on people with disabilities.
Being an advocate to change chatbots		1			Advocating for more stories about how disabled live in society and inputting that data to train the AI would help to give it a better and logical understanding of how different kinds of people live within society.
How users are using it		2	Proper usage	1	It usually provides accurate information if used properly.
			Used as a tool for empowerment	1	Utilizing AI to provide space in between the gaps of what disabled people require to improve their quality of life in the everyday might earn them societal respect and lower the rates of discrimination.
Have chatbot counter biased input		1			Trying to effectively beat stigmas and stereotypes when users ask biased questions.
Structural power imbalance in chatbots/AI		1		1	From either perspective chatbots can be used to fight the opposing party, however, in these cases the disabled community is placed at a deficit as there are more biases and falsehood present for them. Those with typical ability are placed at a high advantage as things are typically more accessible for them. Both party’s can utilize chatbots to fight against one another.
Perceptions of/attitudes towards chatbot				5	Usually it uses respectful languages and includes many perspectives, highlighting social barriers rather than framing disability as a personal deficit; focus on how not every individuals with disabilities will face the same challenges, as they are smart enough to learn things with the right technique and education; Chatbots are respectful, generally accessible and mostly bias aware. I think these programs need to be improved, but can be very useful for challenging ableism and disablism; Chatbots are useful for challenging ableism and disablism because they have tools that can be used to help a person with a disability maintain their independence. Tools such as text to speech and auto captions are a some examples of this; Chat bots can be used to challenge ableism and disablism rather than reinforce because it can break the stereotypes that are being used in everyday life.
Being objective	1				Maybe it can create new perspectives. Since AI is purely objective, it could help bring to light perspective that are shot down due to people being subjective.
Danger of chatbot	1				It can bring back old attitudes against ableism that is commonly seen on the internet because AI chatbots get their source of information from what is available on the internet. which can be negative at times.
Against use in general but others will use it	1				I personally am quite against AI bots, so I wouldn’t use an AI as a way to challenge disability discrimination, but I’m sure many will use it as a tool.
Can be used	1				They can be used to educate the general population and gather data as evidence of lived experiences to change policies and perspectives.
Not biased	1				Because chatbots focus on all perspectives and aren’t bias to one side, meaning that there can be many ideas and concepts around the topics of ableism and disablism that can challenge other individuals ideas.

4. Discussion

In short, participants highlighted accessibility barriers, including attitudinal barriers; they flagged the importance of data on the lived reality of disabled people and acknowledged the lack of it at present; and they identified problems with the position of disabled people as knowledge producers. They also flagged problems with the content of the chatbot output for activism, in part because chatbot answers reinforce ableist power structures. They gave many examples of practical impacts of not having data on the lived reality of disabled people and suggestions on how chatbots can be used to challenge the negative use of ableism (disablism) rather than reinforce it.

In Table 1, we clustered the questions covered in relation to AI tools and chatbots under four themes (accessibility, knowledge production, literacy, and ability judgments). Table 21 below recaps the key results of the survey by themes.

Table 21. Summary of key survey results by these themes.

Theme	Q Number	Question (Short)	Key Result	n
Accessibility: Attitudinal	Q2	AI tools support respectful/fair treatment?	Moderate extent most common (49.3%); 40.6% say not at all/small extent	69
Accessibility/impact: Attitudinal	Q5	Concern about AI-related issues	Very/extremely concerned: excl./harm 83.5%; no consultation 80.0%; reduced empathy 94.1%	85
Accessibility: Physical/digital	Q4	Disabled people consulted in AI dev/deployment?	No (49.5%); not sure (45.1%); yes only 3.3%	89
Accessibility: Physical/digital	Q7	Chatbot accessibility across disability groups	Blind/low-vision least accessible (64.8% report significant/complete barriers); other groups moderate	83–88
Accessibility: Physical/digital	Q8	Excluded/faced barriers using a chatbot?	Yes (72.6%)	84
Accessibility: Physical/digital	Q9	Chatbots offer meaningful customization?	Moderate extent most common (52.4%); 38.1% report little/no customization	84
Knowledge production	Q3	AI systems fairly reflect needs/experiences?	No (74.1%)	85
Knowledge production	Q10	Chatbots reinforce non-disabled-centered design?	Often/always combined 51.8%	85
Knowledge production	Q15	Chatbots reflect lived experience?	Rarely most common (62.4%)	85
Knowledge production	Q16	Chatbots rely on medical/deficit model?	Sometimes/often combined 81.2%	85
Knowledge production	Q17	Chatbots center disabled voices?	Rarely most common (55.3%)	85
Knowledge production	Q18	Encountered harmful stereotypes from chatbot?	Never/rarely combined 67.1%; sometimes/often/always combined 32.9%	85
Knowledge production	Q19	Disabled knowledge/activism/scholarship represented?	Yes 4.71% and no 95.29%	85
Knowledge production	Q20	Chatbots reflect diverse/intersectional identities?	Yes 7.04%; no 69.01% and not sure 28.00%	75
Literacy	Q12	Importance of collecting data on lived experience?	Very important/essential combined 92.9%	85
Literacy	Q13	Practical impacts of not having such data? (open-ended)	Top themes: barriers/inaccessibility (25 mentions), perceptions of disabled people (22), knowledge gaps (20)	129 codes
Literacy	Q21	Effectiveness of chatbots for advocacy information?	Slightly effective most common (52.9%)	85
Ableism/disablism	Q22	Chatbot answers reinforce or disrupt ableist power structures?	Reinforce (64.3%)	79
Ableism/disablism	Q23	How can chatbots challenge ableism/disablism? (open-ended)	Top themes: centering disabled voices (26 mentions), increase accessibility (16), educate users (16), train chatbots (13)	79 codes

All problems identified suggest that a lot of work is needed to be in compliance with the “CAN-ASC-6.2:2025—Accessible and Equitable Artificial Intelligence Systems” document [67] from Accessibility Standards Canada, published in December 2025.

The findings also suggest a problem for the “‘Nothing About Us Without Us’ [146,147,148,149,150,151,152] sentiment that disabled people want to be heard on topics that impact them [153,154,155,156,157,158,159,160] but think they are often ignored [159,161,162,163,164,165,166]”, cited from [167] (p. 4), a sentiment also recommended to follow in the policy document “The impact of the use of artificial intelligence on people with disabilities” [68].

We discuss in the next four sections the four categories (accessibility, knowledge production, literacy, and ability judgments).

4.1. Accessibility (AI Tools in General and Chatbots)

As to physical/digital accessibility for the question asking whether disabled people had been consulted in the development or deployment of AI systems the participants used (Q4), responses split almost evenly between “no” and “not sure”, with almost no one answering “yes” (3.3%). Specific to chatbots, the majority felt that disabled people face physical, digital, and attitudinal problems in using chatbots due to chatbots tailored to ability norms (Q7-Q9). As to attitudinal accessibility, which includes how one treats and talks about disabled people [168], the majority felt that AI tools in general do not support respectful and fair treatment of disabled people (Q2), do not reflect the needs and experiences of disabled people (Q3), and were concerned about algorithmic decisions that exclude or harm disabled people: AI tools designed without consultation with disabled people and AI reducing human empathy in processes such as HR, education, or healthcare (Q5).

Specific to chatbots, the majority felt that disabled people face physical, digital, and attitudinal problems in using chatbots due to chatbots tailored to ability norms (Q7-Q9).

The lack of physical, digital, and attitudinal accessibility flagged by our participants aligns with problems raised in “Building A Disability-Inclusive AI Ecosystem A Cross-Disability, Cross-Systems Analysis Of Best Practices” [69] and “The impact of the use of artificial intelligence on people with disabilities” [68]. They also fit with the Canadian Accessibility Act, which flagged attitudinal barriers as a main accessibility issue [169], and with existing literature that flags biased content, such as the overrepresentation of a medical and/or deficiency focus of disabled people [3,4], and a study that directly modelled barriers to accessing online tools/services for disabled people in Canada and the physical/digital/attitudinal accessibility barriers they encounter [170].

Inaccessibility in any of these forms decreases the well-being of disabled people across many well-being indicators, which marks a failure against the very goals of AI ethics [31,32], responsible AI [34], and AI for Social Good narratives, as all note the importance of well-being of people, which should include disabled people. The findings are also problematic given that governments increasingly have the ability expectations of their citizens to be digital citizens, which includes being digital activists [171], because not being able to perform this ability expectation negatively impacts the well-being of disabled people.

4.2. Knowledge Production: Who Produces and What Knowledge

Two distinct problems fall under knowledge production: who is enabled to produce knowledge, and what knowledge gets produced.

As to who produces the knowledge, one consequence of inaccessibility (physical, digital, and attitudinal) already mentioned is that disabled people face barriers in being knowledge producers. That participants felt that disabled people were not the knowledge producers was also present in that the majority felt that chatbots do not center the voices of disabled people (Q17) and that disabled people’s knowledge, activism, and scholarship are not adequately represented in chatbot outputs (Q19) and that chatbots reinforce design assumptions that center non-disabled users, so users who fit the ability norm (Q10).

The answers of our participants clearly saw disabled people’s voices missing in the co-design of AI tools and the production of content visible in the output of AI tools, such as chatbots.

The concerns of our participants fit with prior literature. It has been stated for a long time that disabled people are experts of their lived experiences but that they are not involved in the knowledge production about their own lives [6,172,173] (see Nothing about us without us, which was already mentioned). Numerous articles engage with the epistemic exclusion [174] and epistemic injustice [175,176,177,178,179,180,181] experienced by disabled people as knowledge producers.

It has also been shown that the main theoretical lens of ability judgments, such as ableism and disablism, was rarely present in the AI governance literature, including AI ethics [17] and AI for good and responsible AI (data produced for the background section of this study). It also fits with studies that indicate a lack of visibility of disabled people as an issue of inquiry and as influencers of AI development and use [2,3]. The answers of our participants also pose problems for the goal of “AI/IT/computer for social good” [35,36,37,38,39,40,41,42,43,44,45,46,47,48,49], “AI for common good” [50], and “AI for good” [182], which require that the voices of disabled people and their concerns be present.

As to what knowledge is produced, the majority of participants felt that AI tools do not reflect the needs and experiences of disabled people (Q3); that chatbots do not reflect the lived reality of disabled people (Q15); that chatbots rely sometimes or often on medical or deficit-based models of describing the disabled person (Q16); and that chatbots do not reflect a diversity of identities of disabled people, including intersectional perspectives (Q20). That our participants stated that intersectionality related data is not reflected in chatbot output makes sense, as studies noted that intersectionality of disabled people is under-researched, including how to teach about the intersectionality of disabled people in intersectional pedagogy [183]. This also fits with our search performed for this article that showed a lack of coverage of AI through a disability justice lens, a lens that focuses on the intersectionality of disabled people. Teaching content has been generated recently on how and what to teach about the intersectionality of disabled people using the intersectional pedagogy framework [71]. However, it is not clear whether this document would have already been put forward by chatbots at the time our study took place. Our participants’ responses suggest a problem in achieving literacy on the intersectionality issues disabled people face, which in turn severely hinders the achievement of disability justice, which has as a cornerstone intersectional reality [59,60,61,62,63].

Interestingly, although for the question of whether chatbots reinforced harmful stereotypes or outdated terminology about disabled people (Q18), the “never” or “rarely” were the dominant answers, at the same time 43.53% for “sometimes” and 37.65% for “often” were the main answer choices selected for Q16, “When you ask chatbots disability-related questions, do you think they rely on medical or deficit-based models of describing the disabled person?” This suggests that the medical/deficiency framing was not seen in terms of “reinforced harmful stereotypes or outdated terminology about disabled people.

The sentiment of our participants that data on the lived reality of disabled people are missing also fits the literature. It is well reported that data on the social situation of disabled people are missing [5,6,7]. Where such content is produced at all, its producers are concentrated in a few countries, even for topics that are at the core of disabled people’s lives [17].

Given the answers to who produces the knowledge, the answers to what knowledge is produced make sense. For example, the answers to Q16 fit with the literature that flagged the overrepresentation of a medical, deficiency focus of disabled people [3,4]. The bias in what data (text and imagery) about disabled people are produced by AI tools and chatbots shapes what users (human beings and AI actors) come to believe the lived reality and imagery of disabled people is.

4.3. The Issue of Literacy (Questions Asked in Relation to Chatbots and a General Question)

On the literacy issue, in response to Q12, participants indicated that it is very important/essential to collect data on the lived experience of disabled people. Questions linked to the impact of chatbots found that participants felt that the use of chatbots to obtain information so one can advocate for the rights of disabled people is not at all effective/slightly effective (Q21).

With respect to the open-ended responses to the question “What do you think are the practical impacts of not having data on the lived reality of disabled people” (Q13), participants overwhelmingly described the negative impact (n = 129), with the main sub-themes being barriers and inaccessibility (n = 25), perception of disabled people (n = 22), decrease in support (n = 16), policies (n = 16), knowledge (n = 20), participation (n = 7), services (n = 6), and advocacy (n = 5).

Literacy is seen as an important goal in many AI content discussions, as is the issue of critical thinking. The combination of low-quality data, a decrease in critical thinking amongst AI users, and the risk of epistemic collapse (the loss of diversity in knowledge and reasoning) carries many implications for disabled people and their allies. Users might not realize that there is a problem in how disabled people are represented, which in turn might decrease the number of allies of disabled people on a given topic, including AI itself, as they would not realize that there is a problem. Furthermore, it can make the task of educating others, which is expected from allies, more difficult. Indeed, the need to constantly have to educate others and the others’ lack of knowledge even on issues that have existed for a long time were identified as one main factor in ally burnout, as was the attitudinal inaccessibility allies faced when they raised an issue and the others simply did not see it as a problem [27]. It is argued that due to digital echo chambers and biased use of digital media, the digital citizen is “increasingly exposed to bias undermining the key assumptions of liberal democracy” [184] (p. 212). The digital echo chamber is also a problem for disabled people given the biased perception and content generation around disabled people. Given the ever-increasing digital world and the ever-changing ability requirements of digital citizens in their role of being activists, it is important to generate data on the reality of disabled digital citizen activists, including the physical, digital, and attitudinal accessibility issues they face to become digital citizens [167].

More specifically, in terms of addressing chatbots, many sources outlined challenges related to chatbots,

“such as training data, affects user learning and development, exhibit preferences, e.g., certain ethnic and gender groups, political viewpoints, lack the creativity and originality like humans, fabricating data or results. Manipulating, altering, or omitting data or results so that they do not accurately represent the research, hallucinations, text that seems semantically or grammatically reasonable but incorrect or meaningless, inaccuracies in generating citations, inaccurate facts or flawed logic, unreliable and poorly informed responses, outputs are copy or combination of the training data, and social prejudice, discriminatory, aggressive, insulting answers”.

[94] (p. 9)

Many types of biases against disabled people are produced by various sources [2,3,4,104,105,106,107,108]. Representational harm is flagged as one harm caused by chatbots, whereby the content the chatbots provide is seen to amplify bias instead of improving the literacy of the user [106]. Some authors propose that one should ask “the following six questions of chatbot outputs:

(1) How well does the statement accurately represent people’s lived experiences? (2) How objectifying is the statement towards disability (e.g., inspiration porn)? (3) How disregarding or marginalizing is the statement of one’s granular identity? (4) How assumptive is the statement? (5) How violating is the statement of one’s autonomy? (6) How discouraging is the statement towards one’s abilities?”.

[106] (p. 212)

As chatbots make use of existing data, the bias our participants and the literature flagged will be reflected in their output, which in turn will have an impact on what users think the lived reality of disabled people is. As such, it is important for disabled people and their allies to counter the many problems noted [14,67,68,69], especially the problem of decreased or lack of critical thinking and literacy [93,94,95,96], in order to decrease and counter the danger of a declining literacy about disabled people [15] (p. 85). Indeed, if we look at the answers to Q19, “How effective is the use of chatbots to obtain information so one can advocate for the rights of disabled people?”, not at all effective 12.94%, (n = 11), slightly effective 52.94%, (n = 45), moderately effective 7.06%, (n = 23), mostly effective 4.71%, (n = 4), and extremely effective 2.35% (n = 2), this can only become worse with a decreasing literacy about disabled people.

Referring back to the aforementioned definition of critical thinking in AI [95], a question arises: if disabled people are not knowledge producers, if data on the lived reality of disabled people are missing, and if data favour certain biases such as a medical view of disabled people, then who would have the capacity to “evaluate AI-generated content to make informed and responsible decisions” [95] (p. 2)?

Our participants, who, as disability studies students, one assumes, might be disabled people themselves or, more likely, were already exposed to the lived reality of disabled people and therefore have already been conditioned to be allies of disabled people, might be able to question certain content. Yet even they can only notice what is missing if they have encountered a different narrative somewhere else first. If they were never exposed to the differentiated views disabled people have on AI, especially outside the academic literature, as most very likely did not have this topic in their classes before this survey, how would they make the various critical thinking linkages to be able to evaluate the positive or negative impact on disabled people, and with that, also differentiate the impact based on different categories of disabled people? Even if they have been exposed to disabled people before, this very likely was limited to certain disabilities, for example, their sibling or parent. How diverse a group of disabled people would a high school student see in their classes or in their school? In general, a user who has never been exposed to a topic has no reference point against which to note its absence from a chatbot’s answer. The gap is invisible the most to people who need it most to be filled. Verification (scrutinizing and corroborating AI outputs), which the critical-thinking literature treats as essential, cannot operate on an absence the reader cannot perceive. It is stated that “structured prompting can decrease the negative effect on critical thinking” [95] (p. 3). But structured prompting (a) still needs available data and (b) one needs to be taught which prompts to use.

The verification problem also exists at the level of access to sources. A recent review of the coverage of ableism, disablism, and other ability judgment terms in the English language academic literature covering Scopus, Web of Sciences, and the 80 databases accessible under EBSCO-Host [17] found that 75% of the data published are not open access. Furthermore, it was noted [17] that the list of countries from which the authors originated and the journals that published them were limited in their diversity. Indeed, it has been argued for a long time that data through the voices of disabled people from the Global South on the reality of disabled people is even more missing than voices from the Global North [113,185,186,187,188,189,190,191,192,193,194,195]. So, what data and written by whom, did the participants of our study and students in general have access to? Will students have heard in high school of AI in relation to disabled people? Will students in STEM (science, technology, engineering, and math) degrees learn about AI ethics, responsible AI, AI for good, and AI justice in relation to disabled people beyond a techno-optimistic angle? For example, a study performed in 2021 looked at the views of STEM students on the impact of AI on the ability to have a good life. Participants could choose from various social groups. Participants selected the groups of disabled people and elderly people to be the ones most positively impacted by AI now and even more so in the future [108]. How can that be, given the many problems outlined for disabled people around AI that we cite in the background section? Were they not exposed to the critical voices of disabled people on the topic? The article reporting on the study [108] refers to other studies that flag the techno-optimistic and techno-deterministic teaching in STEM education, and the article notes that the coverage of AI in relation to disabled people is so biased towards the positive that this shaped the answers. It would be interesting what the answers would be today.

Verification depends on corroborating an output against sources; if most of the relevant sources are behind paywalls, the many users who get their information from chatbots cannot perform the very check that critical thinking requires. We suggest that the situation, as is, will pose general problems for closing the literacy and data gap, and even more severe problems for closing this gap between the Global South and the Global North and for increasing the voices of disabled people from the Global South, which are goals of the field of Global Disability Studies [196,197,198,199,200].

None of the AI guidance, standards, and policy documents we covered engaged explicitly with the impact of the existing data on the literacy about disabled people; all looked at increasing the literacy of disabled people in using the AI tools, which misses a big area of literacy.

4.4. The Issue of Ability Judgments

Being negatively ability-judged is a main concern for disabled people, as is the disablism (social discrimination) they experience, as one consequence of being negatively ability-judged [17,115]. Therefore, we asked our participants about that topic in relation to chatbot content a user would receive. As to the question of whether chatbot answers reinforce ableist power structures (Q22), 29.76% (n = 25) felt that they disrupted them and 64.29% (n = 54) felt that they reinforced them. This sentiment fits with various articles we cite in the background section, reflecting the view that AI is based on and produces negative ability judgments related to disabled people. For example, many of the studies we cite in the background section covering AI and disabled people use the term “eugenics” [8,75,77,79,88]. This makes sense. Eugenics, whether positive or negative eugenics, is, at its core, about the negative use of ability judgments to justify a social hierarchy. It depends on arbitrary, false and irrelevant ability expectations narratives set by people who have the power to set ability judgment narratives that suit their group. To quote Galton,

“We greatly want a brief word to express the science of improving stock, which is by no means confined to questions of judicious mating, but which, especially in the case of man, takes cognisance of all influences that tend in however remote a degree to give to the more suitable races or strains of blood a better chance of prevailing speedily over the less suitable than they otherwise would have had. The word eugenics would sufficiently express the idea…”.

[201] (p. 25)

Eugenics is reflecting all aspects of the ability judgment injustice definition.

“Ability expectation/ableism Injustice, short ability judgment Injustice and even shorter ability Injustice, is present in a world that promotes irrelevant and/or arbitrary ability expectations, and ability privileges and disablism based on these expectations, and does not use ability expectations and ableism to decrease ability judgment-based oppression, inequity, disablism and privilege”.

[17,20]

As to the question “How can chatbots be used to challenge ableism and disablism rather than reinforce it (Q23), the dominant theme was change design (n = 79), which included, for example, the sub-themes “educate users” (n = 16), “change language” (n = 10), “centering voices of disabled people/people with disabilities” (n = 26), and themes, such as “increase accessibility” (n = 16), “train chatbots” (n = 13), and “generate new data” (n = 7). They interpreted the question to mean what has to change for the content to be useful for disrupting purposes.

To follow through with the suggestions the participants posed requires dealing with all the issues our participants flagged as problems in conjunction with AI tools in general and chatbots specifically, problems one could categorize as forms of disablism, as discussed in the last three sections, and problems that were also identified in policy documents and academic literature we used as guidance documents in our article. Furthermore, although only seven participants specifically mentioned the generation of new data, many of the other action items noted have as a prerequisite the generation of new data.

The answers to Q22 and Q23 suggest that people involved with chatbots on all levels, from design to training and use, should familiarize themselves and engage with the over 50 ability judgment-related concepts [17,18,19,20,21,22,23,24,25,115,116,117,118] and specifically with the concepts of ability justice and injustice [17] in order to increase their critical thinking ability in relation to the cultural reality of ability judgments.

Our participants flagged various ability justice and injustice issues, although without using the terms of ability justice or ability injustice, such as the concerns they agreed with in answering Q5. The problems our participants flagged could be worded in many ability judgment-linked concepts, such as ability oppression, ability inequity, disablism, and ability privilege (definitions at [20]). For example, the ability inequity definition, “Ability inequity is a normative term denoting an unjust or unfair distribution of access to and protection from abilities generated through human interventions” [202] (p. 99), reflects perfectly many of the problems discussed around AI in the moment and for many concerns our participants voiced.

To engage with the world of ability judgments is important because how to achieve ability justice and to avoid ability injustice for any given topic differs for disabled people of different body/mind characteristics and whether their identity intersects with another marginalized identity [71,183], or lived reality (e.g., Global South). To engage with the ability judgment concepts is also of benefit to other marginalized groups, as ability judgments are also used to justify the marginalization of other groups [17]. The concept of ability obsolescence [203,204] can be used to ascertain emerging ability judgment-based marginalized groups of humans, and already marginalized groups whose sets of abilities will be judged even more in a negative way and who have no way to obtain new abilities needed.

Furthermore, as ability justice also asks for using ability judgments to enable the ability of a good life [17], the many different discourses that have as one goal to enable a good life (e.g., the capability approach by Sen, Anand, and Nussbaum [205,206,207,208,209,210,211]), and the many indicators of well-being from the Better Life Index [212], the Canadian Index of Wellbeing [213], the World Health Organization-initiated Community-Based Rehabilitation (CBR) Matrix [214], and the Social Determinants of Health (SDH) [215,216,217]) demand differentiated actions, based on many different parameters.

Engaging with the ability judgment concepts increases literacy on how ability judgments can generate and defend various social hierarchies but also how they can be used to enable a good life.

As such, all ability judgment concepts, including ability judgment justice and injustice, are of use to all AI governance discourses, AI ethical reasoning and AI justice.

Ability judgment-based literacy and critical thinking have to include areas such as how ability judgments shape science and technology advancements and governance, and how science and technology advancements and governance shape ability judgments, so one can design prompts that allow one to see potential impacts of these discussions on society at large, including disabled people and to judge the quality of the output of AI tools, such as chatbots.

5. Conclusions, Future Research, and Implications

This study set out to examine how critical disability studies students, as allies of disabled people, view AI tools and chatbots in relation to disabled people, with particular attention to the impact of chatbot content on the literacy about disabled people, on activism and on the disablism disabled people face. Our participants outlined many problems ranging from the design of AI tools, including chatbots, to the content of chatbots. Participants highlighted the negative impact of chatbot use on the literacy of the lived reality of disabled people and the low utility of chatbots for activism, and participants gave various suggestions to enable AI tools and chatbots to decrease the disablism disabled people experience.

Our participants were bachelor of critical disability studies students at a single Canadian university, a purposive group whose views are not meant to be generalized even to disability studies students as a whole or in a given program. Within disability studies, the questions could be asked of cohorts at other universities, at different stages of their program, and in other regions. Beyond disability studies, the questions we developed should be given to other groups, such as disabled people from various backgrounds, with various disabilities, and various self-understandings and should cover disabled people who are activists and who are not. Because disabled people need allies from every field, the same questions should also be put to potential allies, such as students in engineering, computer science, and data science, and people already working in AI ethics, responsible AI, AI for good, and AI justice whose choices directly shape the tools. As studies accumulate, meta-analysis across them could ascertain where views are different and similar within disability studies and beyond.

Furthermore, one could also do the survey before and after the content of AI has been covered in a class to see whether something changes. One could also give the same survey to students at the beginning of the given year of class and again at the end of the year to see whether the students answer the survey differently.

Our study did not specify the AI tools, including specifying the chatbot or a given disability. Related studies can focus on specific disabilities, or specific intersectionalities between the disability identity and another marginalized identity and lived reality or on specific AI tools and chatbots. Answers might differ for some questions, and the style might differ between the chatbots (we routinely compare ChatGPT, Kimi, and Claude AI), which might already lead different users to different reactions to the content. However, AI tools also constantly evolve, whereby the answers become more polished; however, if the missing and biased data situation is not rectified, the quality of the answers will not fundamentally change, but for people not knowledgeable about the social reality of disabled people, the pitfalls of the answers might become even more difficult to detect.

Our study results demand actions from various groups.

For AI developers and designers, what would help is inclusion at the source; designing chatbots that challenge rather than reinforce disabling power structures; centring the voices of disabled people in design and training processes; generating new data grounded in disabled people’s lived experience; and building tools that allow users to learn about the lived reality of disabled people.

For educators, our findings suggest the need to equip students with prompting strategies that allow students to gain some insight into the quality of the output of chatbots. Exposing students to ability judgment concepts could be one strategy to enable the critical thinking and prompting skills of students and others. For example, having chatbots answer the same question but through different lenses (non-disabled people, a marginalized group, disabled people, disability rights, disability studies, disability justice, ability justice and injustice (using the definitions from [17]), and other of the over 50 ability judgment-related concepts [20] fitting a given topic) should reveal various issues and gaps. Another useful strategy is exposing students to the BIAS FREE Framework (Building an Integrative Analytic System for Recognizing and Eliminating InEquities), which was developed by Burke and Eichler [218,219] to identify and prevent biases arising from social hierarchies. One could use this tool to further the critical thinking skills of students and others on the lived reality of disabled people and others by unmasking the social hierarchies disabled people and others experience. One could also use the BIAS FREE to analyze ability judgment justice/injustice issues caused by AI development and implementation discourses.

For policy and standards bodies, our participants’ evaluations suggest that current AI, as we experience it, does not yet meet the requirements of the CAN-ASC-6.2:2025 standard [45], amongst other standards at the national and international levels, and that the standard’s own conception of AI literacy needs widening to include literacy about the lived reality of disabled people, not only disabled people’s literacy in using AI.

Across all areas, the underlying issue is the same one our participants and the “Nothing About Us Without Us” principle both point to, which is that there cannot be useful AI content about disabled people without disabled people as knowledge producers and leaders.

Finally, various tools still need to be developed to ascertain the impact of AI use on the lived reality of disabled people, such as questionnaires that allow one to ascertain the impact of AI use on the literacy on a given topic in relation to disabled people and strategies that allow one to ascertain whether this literacy decreases or increases over certain time periods in general or due to the use of AI tools.