From Digital Natives to AI Natives: Emerging Competencies and Media and Information Literacy in Higher Education

Abstract

The arrival of artificial intelligence (AI) is transforming the informational and epistemic landscapes of higher education institutions. This study examines the skills that students believe they have developed through AI use and considers the media and information literacy (MIL) skills required for its ethical and critical application. A total of 3120 responses from students at two Latin American universities were analyzed using the ALCESTE method, supported by IRAMUTEQ software 0.7 alpha 2 2020. The analysis identified four competencies: assisted writing, enhanced self-management of learning, faster academic output, and methodological meta-reflection. The findings suggest that although students note improvements in performance, the educational value of these skills depends on critical engagement. Risks such as dependence, misinformation, and loss of agency were recognized. In response, four key MIL-AI competencies are proposed: critical discernment, academic integrity, cognitive independence, and qualitative judgment. The conclusion emphasizes that universities must actively serve as ethical laboratories for the responsible use of AI, fostering students who can navigate technology with awareness and judgment.

1. Introduction

The rise of artificial intelligence (AI) is reshaping informational and knowledge ecosystems with ontological, epistemic, communicational, and ethical impacts (McLuhan & Fiore, 1967; Sanchez-Acedo et al., 2024). In a highly connected society, where people aged 16 to 24 spend an average of 7 h and 30 min online daily (Kemp, 2025), AI is often integrated into daily life without much scrutiny: it mimics agency and emotional responses, automates tasks, saves time, and boosts content creation, even developing discursive tools that can establish credibility and authority among users (Natale, 2021). In a university context, this integration sparks fascination and creates ambiguous boundaries where the line between functional yet artificial interactions and those that may undermine critical thinking and independence becomes unclear (Umbrello & Natale, 2024). This study examined the skills related to the use of artificial intelligence that students believe they have developed in their academic work. It also discusses the media and information literacies (MIL) needed to promote the ethical and responsible use of these tools in the classroom. The goal is to update MIL to include MIL-AI, helping prevent students from being harmed by failing to distinguish between reliable information and biased or fake content (Carli & Calvaresi, 2023; Kerdvibulvech & Jiang, 2025). Moving toward AI-related media literacy is a civic skill (Hristovska, 2023) because training students to recognize and resist information manipulation supports education, lifelong learning, and democracy in algorithmic environments (Tiernan et al., 2023).

2. From Media and Information Literacy (MIL) to MIL-IA for University Education

Digital transformation is altering social interaction: we are shifting from individuals mainly consuming information to becoming producing-curating agents capable of shaping truths and influencing content that impacts decision-making. This transition necessitates updating Media and Information Literacy (MIL) (Tiernan et al., 2023) to identify informational agencies that arise from the interaction between humans and algorithms (Braidotti, 2019; Juelskjær & Schwennesen, 2012; Trejo-Quintana & Sayad, 2024). These agencies, combined with information overload, create conditions that foster hoaxes or fake news, echo chambers, cognitive biases, and other advanced strategies for deception or information predation (Begby, 2024; Gunn, 2021; Munroe, 2024), which diminish the ability to evaluate the reliability of the digital media (Rohman et al., 2025).

Until 2022, according to Tiernan et al. (2023), AI appeared only tangentially in national frameworks such as the Canadian Digital Competence Framework (2019) and the Spanish Professional Framework for Teaching in the Digital Age (2019), mainly as a technological tool rather than as an epistemic infrastructure for education. DigiComp 2.2 (2022) connects it with digital evaluation and navigation, recognizing its influence on the creation, positioning, and consumption of information; however, its approach remains neutral regarding models and does not delve into the specific practices required by generative systems (e.g., hallucination control and verification of the provenance of generated content). Meanwhile, although the normative response is delayed, the presence of AI is quietly becoming normalized, shaping emotions, trust, and decisions (Natale, 2021; Umbrello & Natale, 2024). Based on this, digital literacy is called upon to promote critical thinking and responsible use of information (Ghodoosi et al., 2023; Van Audenhove et al., 2024), to break the bubble effect (Pariser, 2017), and to avoid socio-technological binaries (emancipatory technologies versus oppressive elites). It is imperative that we diversify with social change, emphasizing that there is no single perspective for understanding it or for evaluating the level of literacy a person can have in the 21st century (Špiranec et al., 2019).

In line with the above, UNESCO (AI and Education: Guidance for Policy-Makers; UNESCO, 2021; Guidance for Generative AI in Education and Research; UNESCO, 2023, 2019) promotes advancing MIL toward MIL-AI, stressing algorithmic citizenship and cognitive self-defense. It emphasizes identifying AI presence in everyday information, combating filter bubbles, disinformation, fake news, and microsegmentation, which threaten epistemic agency (Coeckelbergh, 2023; Mochizuki et al., 2025). The goal is to develop a citizenry capable of co-designing solutions focused on the common good, repairing inequalities, and being vigilant about the potential erosion of social fabric by algorithmic infrastructures (Ahmad et al., 2022; Cox, 2024; Marushchak et al., 2024; Sriprakash et al., 2025). In operational terms, this transition requires that MIL-AI enable, among others, the following capabilities: (i) Detecting the invisible presence of AI by monitoring technical clues such as real-time inferences and data contrasts, thereby fostering informed suspicion. (ii) Understanding the meaning of learning, reasoning, or perceiving that AI must avoid reducing the world to mere textual patterns. (iii) Evaluating cost-benefit and risk-mitigation relationships before delegating academic or institutional decisions to AI. (iv) Engaging in debates about AI control and its purposes to enhance critical thinking in civic education. (v) Gaining insight into how the AI black box functions to develop critical explanations of its responses. (vi) Promoting data sovereignty for those who generate data, not just those who use it. (vii) Leveraging the power of algorithms to address historical exclusions by recognizing differences and the humanity within education. (viii) Developing alternatives for coexistence with technology that go beyond solutions imposed by large corporations, which position technology as a disruptive commodity.

A distinctive feature of the shift from classic MIL to MIL-AI is the move from critiquing messages to critiquing infrastructure. It is no longer just about analyzing media but about understanding, auditing, and transforming sociotechnical systems that extract, process, and decide based on personal and collective data (Couraceiro et al., 2025; Mochizuki et al., 2025; Sriprakash et al., 2025). This shift involves four key displacements: from content to system, from technical skills to law and civic responsibility, from instrumental to critical pedagogy, and from the isolated individual to the subject in relation. The urgency arises because AI infrastructures enable mass data collection and decision-making automation with public impact, so symbolic participation is replaced by technopolitical participation. Within this framework, the study does not challenge existing frameworks (Digicomp, MIL/UNESCO) but instead operationally redefines their descriptors, exploring how students claim to have their competencies when interacting with AI are practiced and verified. For example, this shift may involve moving the “search-filter-evaluate” triangle toward model-aware discernment (reducing hallucinations, verifying sources, triangulating, and dating claims (Ghodoosi et al., 2023; Stewart & Rodgers, 2025); viewing creation as orchestrated authorship (prompt interaction and justification, tool chaining, version control) (Bender, 2024); and incorporating cognitive autonomy (criteria for when not to use AI and thresholds for delegation that prevent overdependence on technology) (Grotlüschen et al., 2024).

In short, given AI infrastructures that operate with opacity and limited accountability, and whose deployment is reshaping the public sphere by automating attention and decision-making, a critically aware innovation agenda is essential. This agenda must empower agencies capable of challenging and decolonizing the algorithmic knowledge produced in the Global North and its extractivist logics (Saliu, 2024), while promoting an epistemological democracy that aims to identify system tensions, open up its rationales, and create conditions for reformulating it from diverse perspectives.

At the university level, this agenda requires changing curricula and governance: basic awareness is not enough; it is necessary to incorporate data epistemologies and algorithmic auditing that encourage conscious interaction with systems, prevent inequalities, and enhance students’ ability to ethically question and interpret their actions with AI (Fridman et al., 2025; Kerdvibulvech & Jiang, 2025). In this way, competency frameworks would define their role as teachable and auditable practices, aligned with the guidelines of DigComp 2.2 and UNESCO for the critical and responsible use of digital technologies of AI, thereby strengthening the academic integrity of higher education.

3. Research Methodology

This study used a research approach based on lexicometric analysis employing the ALCESTE method (Lexical Analysis by Context of an Ensemble of Text Segments; Reinert, 1990, 2000, 2003). Its level is interpretive and comprehensive because it examines the lexical behavior of the corpus to explore students’ discursive practices, revealing their repertoire of meanings, representations, and experiences with artificial intelligence.

3.1. Units of Analysis

An ad hoc corpus was created from the responses of 3120 university students from the University Center for Economic and Administrative Sciences at the University of Guadalajara, Mexico (2051), and the Faculty of Business Sciences at the Technical University of Machala, Ecuador (1069). A Google form was sent to their institutional email addresses with the following question: “As a student, what can you do to improve your knowledge of the text? Have you conducted things differently or better since you started using artificial intelligence tools?“ The participants did so voluntarily and provided informed consent. The questionnaire was available from 16 to 20 June 2025.

3.2. Corpus Composition

A corpus of short, open-ended responses (Gee, 2014; Patton, 2014) was created in Spanish with the following features: 3120 responses spread across 678 elementary context units (ECUs), with an average of 35.86 forms per ECU. The corpus included 2520 and 24,317 occurrences (tokens). Hierarchical Top-Down Classification identified four lexical classes, with 64.45% of ECUs classified. Following Reinert (1990) and Lebart et al. (2010), the corpus’s compositional features allow the identification of dominant semantic fields and their vocabularies, differentiate discourses, define functional typologies of skills based on lexical co-occurrences, ensure the reliability of classified lexical segments, and support the link between quantitative analysis and qualitative interpretation.

Regarding heterogeneity and analysis criteria, a single broad question was intentionally used to maximize the ecological validity of the student repertoire and prevent framing that would steer responses toward predetermined categories. This broad approach naturally leads to lexical diversity; therefore, the analysis was conducted with homogeneous segmentation into SCUs (~40 words), lemmatization in Spanish, and conservative thresholds for selection (frequency ≥ 5; χ² ≥ 6.63, df = 1), focusing on identifying dominant lexical fields. In this setup, 64.45% of the classified SCUs are considered an internal indicator of stability aligned with broad guidelines and are interpreted alongside characteristic forms and typical segments, ensuring clarity without limiting the diversity of reported practices.

3.3. Corpus Analysis Using the ALCESTE Methodology

The ALCESTE methodology focuses on identifying words that appear together in different parts of a corpus. Using these co-occurrences, the text segments are grouped into homogeneous lexical categories (Reinert, 1990). This process was carried out in six consecutive stages. (i) Corpus normalization: The entire text was automatically converted to lowercase, preserving Spanish diacritics. Punctuation and non-alphabetic characters were eliminated, and apostrophes and hyphens were replaced with spaces. Spaces were standardized, and the **** separator was used to delimit the responses. Exact duplicates and empty records were removed. (ii) Corpus segmentation was carried out using 40 words as the minimum unit to identify blocks of information with a reference context and thus uncover latent meanings. (iii) Lemmatization and categorization of forms to create consistent and stable classes. This process reduced words to their base form (Akhmetov et al., 2020; Rodriguez-Bazan et al., 2023) to help identify genuine semantic patterns (Uddin et al., 2022) without losing the contextual meaning of the words (Malik et al., 2024), leading to better identification of corpus divisions (Karousos et al., 2024). (iv) Estimation of the lexical co-occurrence matrix, which determines the presence and absence of words in each identified segment, by creating a large table that records the frequency of words and their closeness to each other. (v) Top-down hierarchical classification: During this process, the corpus is repeatedly broken down into segments or classes based on how closely the words are related. Each split maximizes the lexical difference between groups and promotes lexical cohesion within them. It is important to note that the match between the words and the class is determined by the chi-square value. (vi) Selection of lemmas to represent the class: the inclusion of characteristic lemmas was based on the 2 × 2 χ² association test (presence/absence × in class/outside), with degrees of freedom = 1 and a single threshold for the entire study. We adopted α = 0.01 (cut-off point χ² ≥ 6.63) and designated lemmas with χ² ≥ 10.83 (α = 0.001) as strong associations. The differences in the magnitude of χ² between classes are not due to different criteria but reflect the actual strength of each lemma’s association: the greater the concentration of a term within a class, the higher the χ². Therefore, some classes show several lemmas with χ² ≥ 10.83, while in others, the lemmas only surpass the inclusion threshold (χ² ≥ 6.63). Notably, a minimal set of pedagogically relevant multiwords was retained. These actions enable each class to be interpreted as a specific semantic-pragmatic field. The application of this methodology was facilitated by the IRAMUTEQ software 0.7 alpha 2 2020 (Ratinaud & Marchand, 2012), an open-access program developed in the R environment using Python (Camargo & Justo, 2013).

3.4. Detection of Typologies of Competencies

Based on the classes identified in the stage or moment described above, a discursive analysis was conducted to identify recurring patterns of meaning and different ways of expressing students’ experiences with artificial intelligence. This process was carried out using positioning theory linked to discourse (Davies & Harré, 1990; Harré, 1991, 2015; Harré et al., 2009). Operationally, from the created classes, the subcorpora were identified and then studied to determine the following aspects: (i) The positions students take toward artificial intelligence. For this, the emotional tone (enthusiasm, distrust, curiosity), perception of usefulness (necessary, secondary, dispensable), and level of involvement (constant use, occasional use, rejection) were considered. (ii) The identities revealed in the discursive interaction between students and their representation of AI. The forms of self-presentation and evaluations expressed were also analyzed. (iii) How students constructed their experiences with AI. In this case, functional narration, operational justification, and comparison with other traditional methods were examined. (iv) An evaluation of the student experience based on emotions and perceptions of AI’s usefulness in their everyday life.

3.5. Discursive Integration and Interpretation of Typologies

This process was developed within each class identified through lexicometric analysis. The goal was to uncover the recurring and specific ways of describing, legitimizing, and valuing the use of AI as an extension of human capabilities, which, for this study, are considered to be typologies of competence expression.

3.6. Relationship Between Competencies, Risks, and MIL-AI

Building on the previously identified categories, the following process was used: first, all text segments that referenced negative effects of using AI were coded. To do this, the key terms considered included: I lose, I depend on, risk of, I am worried, I do not distinguish, could, impersonation, and others. Second, categorization was initiated based on the codes from the previous analysis stage. Third, a convergence relationship was established among the categories to identify emerging skills necessary for media and information literacy with AI. These skills were grouped to highlight their main themes. Each part of the corpus was analyzed separately by two researchers, resulting in a high agreement level, with a kappa index of 0.90.

4. Results

4.1. Perceived Competencies

The lexicometric analysis revealed four clusters representing the competencies or superpowers students believe they possess. The first is cognitive-discursive and was called “AI-assisted writing.” Students experienced a shift in how they construct academic writing. They saw themselves as textual co-producers alongside AI, understanding that this technology enhances their ability to accelerate, refine, and classify discourse according to their interests and preferences.

The overrepresentation of the terms “question” (χ² = 43.15) and “information” (χ² = 21.88) shows that the connection with AI is built around a questioning interaction focused on content management (see

Table 1. Competence: AI-Assisted Writing.

Lemma	χ2	Standardized Error	p
Question	43.15	20	0.000
Access	22.51	9	0.000
Information	21.88	66	0.000
Formulate	20.77	6	0.000
Writing	19.36	24	0.000
Way	17.38	43	0.000
Synthesize	17.27	5	0.000
Professional	16.14	6	0.000
Fast	13.79	27	0.000

). The student formulates queries, gathers and restructures materials, and cycles back to the system for further refinement. In this process, AI is valued as a booster of communicative independence, providing levers of discursive adjustment: “access” (χ² = 22.51), “manner” (χ² = 17.38), and “professional” (χ² = 16.14); these allow changes in tone, style, and structure based on the text’s communicative goal. This competence does not develop in linear stages but instead as a set of three modes of use that can occur simultaneously, depending on the task and context. Each mode is identified by its functional co-text (adjacent verbs and objects) and stable lexicometric anchors in typical segments. The modes that were identified are as follows:

(i) Dialogic exploration: manifested through iterative dialogue between students and AI. The student adopts a tone of active curiosity and an attitude of an intellectual explorer (“search, filter, synthesize”) to refine their questions. He assessed his experience with AI from a cognitive agility perspective. Through question-answer interactions, he seeks to improve the conceptual precision of his drafts. Regarding lexicometric anchoring, this pattern is supported by the terms: “question” (χ² = 43.15) and “information” (χ² = 21.88), with verbs of inquiry and conceptual refinement (formulate, contrast, synthesize), and markers of immediacy (rapidness). Evidence of this is observed in the following statements: I can get instant answers to specific questions...” (score 143.01) and “...I can search, filter, and synthesize information from multiple sources simultaneously; this allows me to answer complex questions...” (score 126.65).

(ii) Strategic synthesis: The aim of this typology is to optimize information tactically. Using a practical and straightforward tone, students are guided toward cognitive efficiency. They view themselves as knowledge managers with the skills to reduce their efforts and boost productivity. The pattern is lexicometrically anchored in the overrepresentation of the words “rapid” (χ² = 13.79), “synthesize” (χ² = 17.27), “access” (χ² = 22.51), and “manner” (χ² = 17.38), which co-occur with elaboration operations (e.g., research, create, draft, structure, format). The functional co-text (rapid + synthesize + production verbs) describes textual transformation and product assembly rather than simple planning. The excerpts illustrate the transformative (not merely organizational) function: “AI allows us to adapt educational content to our specific needs; I can quickly conduct research, create papers, and write essays…” (score 131.70) and, “…it helps me quickly clarify doubts and structure my ideas in a single step…” (score 123.13).

(iii) Executive Polish: This competency pertains to excellence and formality. The tone of its representation is confident and formal, with a critical stance that defines the identity of emerging experts. Here, the use of AI is visualized for a thorough review of writing processes, keeping human authorship at the forefront. The assessment is oriented toward raising the textual standards. The pattern is lexicometrically anchored in the terms “professional” (χ² = 16.14) and in the logical polishing verbs “formulate” (χ² = 20.77) and “synthesize” (χ² = 17.27), which co-occur with fine-tuning actions (review, rewrite, paraphrase, adjust, and version). The extracts demonstrate this quality function: “…it helps me greatly improve the way I formulate sentences… my ability to retain information and learn has improved exponentially…” (score 123.13) and “…I can look up information more quickly and accurately, but I always maintain a critical filter…” (score 131.70).

The second competency was autonomous and participatory learning management. In this context, the student is an active participant in their learning process: they organize resources, obtain explanatory scaffolding in real time, and reformulate content to better understand and retain it. The lexical core that supports it (see

Table 2. Competence: Autonomous and participatory learning management.

Lemma	χ2	Standardized Error	p
Explanation	48.28	24	0.000
Organize	35.82	34	0.000
Time	32.31	47	0.000
Summary	29.26	16	0.000
Study	53.04	38	0.000
Doubts	13.56	24	0.000
Resolve	14.8	21	0.000
Personalized	20.33	9	0.000
Efficient	22.63	28	0.000

) shows the recognition of AI as an instant tutor capable of answering questions and clarifying uncertainties to navigate, step by step, through the understanding of the topic under study through the creation of contextualized explanations for each question posed. Another relevant feature of this competency is cognitive synthesis, which is the ability of AI to condense information into digestible formats, enabling students to reframe knowledge in clear frameworks that promote retention and a deep understanding of the subject. From a lexicometric perspective, the codification of learning management was possible because the lemmas were predominantly adjacent to study/organization verbs (understand, study, review, clarify, organize, schedule, prioritize) and not to product development operations (write, format, present, compile).

In terms of its materialization, this competency is embodied in three complementary modes of use: (i) Developing one’s own voice: the student re-expresses what they have learned in their own words to consolidate meanings, using AI as a scaffolding that facilitates paraphrasing and adjusting their own style for cognitive, not delivery, purposes. Lexicometric anchoring occurs in the interaction of the terms: “explanations” (χ² = 48.28) + “personalized” (χ² = 20.33) with “summary” (χ² = 29.22) and the evidence of what was said is observed in the following extracts: “…to be able to write it in my own words, I sometimes use artificial intelligence for school…” (31.90); “…to do extra work, I can write my work better and it helps me create more exercises for my self-study…” (25.79).

(ii) Instant tutoring and uncertainty resolution: The student recognizes that AI guarantees the immediate resolution of uncertainty by constructing contextual explanations that favor the understanding of the content. The dominant lexical support is made up of the following expressions: “doubts” (χ² = 13.56) “solve” (χ² = 14.80) “explanations” (χ² = 48.28) + “personalized” (χ² = 20.33) and the evidence of its operation is found in the following textual sample: “…as well as… I can quickly clear up doubts I have, make summaries of a topic…” (29.03); “with artificial intelligence I can study faster, understand the topics better and do work with greater clarity and organization” (28.73); (iii) Organization of information and time: the competence is expressed as planning and structuring: it focuses on the generation of summaries and conceptual maps that allow to assimilate and organize the information in a clear and efficient way. At this point, cognitive efficiency and mental clarity were assessed. The lexical support is concentrated in the terms: “summary” (χ² = 29.22) + “organize” (χ² = 35.82) + “time” (χ² = 32.31) and the discursive evidence is in the following textual sample: “…make summaries of a topic that is going to be studied in class, conceptual maps…” (29.03) “…research is a necessary aid for the contribution of the study because doing reading exercises and critical analysis I can study faster…” (29.01).

Continuing with the findings, in the third competence, a saturated lexical core (χ²) is observed that combines product names (presentations, projects, summaries), scaffolding markers (guide, serves, tool), and course tasks (homework, activities), which indicates a use oriented towards assembling and delivering academic artifacts with less friction and a better fit (

Table 3. Competence: Facilitating AI-Assisted Scholarly Production.

Lemma	χ2	Standardized Error	p
Presentations	37.28	20	0.000
Guide	17.08	8	0.000
Serves	31.11	8	0.000
Homework	17.08	8	0.000
Summary	13.09	4	0.000
Tool	15.74	17	0.000
Projects	11.72	10	0.000
Learning	12.03	11	0.000
Activities	7.88	15	0.000

). The overrepresentation of “guide” (χ² = 17.08) and “serves” (χ² = 31.11) profiles AI as a structuring agent that proposes logical sequences and interactive options for the presentation and personalized appropriation of content. Similarly, homework (χ² = 17.08), doing (χ² = 18.25), and presentations (χ² = 37.28) show the delegation of routine activities (quoting, outlining, formatting) to achieve cognitive load relief, freeing up resources for creativity and critical deepening. To shape this competency, the co-occurrence of the lemmas, elaboration/delivery verbs (do, assemble, present, compile, cite, format), and product objects (slides, reports, summaries) was considered. With this competency, AI assumes itself as the orchestrator of academic workflow.

Three clearly differentiated usage trends have been identified: (i) Production acceleration: in this usage, the student presents himself as an efficient manager of his deliverables; he shortens deadlines, increases productivity, and prioritizes action and cognitive economy. The pattern is supported by a lexical core saturated by “duties” (χ² = 17.08) and “activities” (χ² = 7.88), co-occurring with elaboration verbs (to do, to assemble, to present), which indicates unblocking routine tasks to move forward with less friction. The discursive evidence is clear: “…the way to write and elaborate paragraphs and texts in a better way… serves as a guide, lessening the burden of homework that can become saturated” (score 21.90). (ii) Guided structuring of the deliverable: here, the student acts as an assisted learner: they assemble their products following the scaffolding proposed by the AI to ensure internal coherence, sequencing, and formal completion. (iii) Alleviation of low-value cognitive load: Students delegate routine microtasks (summaries, reviews, formatting, citation verification) to the AI to free up attentional resources and focus them on activities with higher cognitive value (creativity and critical analysis). The pattern is supported by a lexical core saturated by “summary” (χ² = 13.09) and “learning” (χ² = 12.03), in co-occurrence with verbs of preparation/study (review, prepare, study) and with the functional marker “serves” (χ² = 31.11), which shows its instrumental nature: Discursive evidence: “…it has been of great help to me in resolving doubts more clearly… it also helps me to review before a class” (score 21.01).

The fourth and final competency is research meta-reflection, characterized by the adoption of AI as a mentor for methodologies aimed at resolving conflicts in task completion. The way in which the terms “instructions” (χ² = 14.56) and “guide me” (χ² = 14.56) are represented reveals how students use AI to learn to design methodological sequences ranging from defining the “topic” (χ² = 16.75) to choosing “research” techniques (χ² = 18.70), thereby enhancing metacognition of the research process itself (see

Table 4. Competence: Research meta-reflection.

Lemma	χ2	Standardized Error	p
Use	23.84	22	0.000
Time	20.92	15	0.000
Quickly	20.92	15	0.000
Help	20.89	10	0.000
Power	19.90	8	0.000
Capacity	19.41	6	0.000
Research	18.70	35	0.000
Topic	16.75	11	0.000
Question	15.36	127	0.000
Process	14.56	5	0.019
Instructions	14.56	5	0.019
Guide me	14.56	5	0.019
Analyze	13.78	10	0.019

). This competency also promotes dynamic self-regulation, as students can evaluate and adjust their approaches in real time. Terms such as “analysis” (χ² = 13.78) and “process” (χ² = 14.56) show that knowledge production follows a design-execution-evaluation cycle in which each phase feeds into the next to reduce methodological biases and errors. As a notable feature, this competency is outlined when the vocabulary appears anchored to verbs of methodological design and auditing (define, delimit, justify, select, evaluate, adjust, contrast) and procedural nouns (techniques, sample, instruments, criteria).

This competency materializes in the following typologies: (i) Mastery of prompt formulation (assisted methodological design): students use AI as a metacognitive mentor to design effective commands and derive instruments and methodological sequences (define problem, delimit variables, choose techniques). This use expresses growing cognitive autonomy: students learn how to research and what to research in the process. (ii) 24/7 uninterrupted consultation: AI enables permanent tutoring that sustains investigative continuity; when in doubt, the student consults, adjusts, and verifies procedural decisions in real time. This mode consolidates the self-regulation of the process (evaluating and correcting the methodological direction when necessary) and is linked to the lexicon of “process” (χ² = 14.56) and “analysis” (χ² = 13.78). In this sense, they consider that: “AI supports me, whenever I have any questions” (score = 15.77). (iii) Agile thematic exploration: AI can be used to quickly explore the thematic field, identify relevant avenues of inquiry, and define the object of study, thereby reducing the downtime between design and execution. This use is articulated with “theme” (χ² = 16.759) and the vocabulary of process and analysis, which inscribes the search in a cycle of design, execution, and evaluation. Example: “I can research any topic quickly, which allows me to advance my project without interruptions.” (score 14.95).

4.2. System of Relationships Between Competencies and IML Dimensions

As stated in this study, the modern informational landscape, shaped by artificial intelligence, calls for a redefinition of the skills that define the subject’s epistemic agency (Braidotti, 2019; Kleinman & Barad, 2012; Barad, 2003). Given the rapid speed with which information is produced, distributed, and consumed, it is necessary to strengthen the critical capacity of subjects so that they can maintain their cognitive autonomy, be ethically responsible in their interaction with information, and question a technological system that responds without question, completes without understanding, and recommends without context. In this sense, four competencies are identified (critical discernment, academic integrity, cognitive autonomy, qualitative management) that are considered integrated configurations of knowledge, skills, dispositions, and performance criteria that allow the subject to sustain a reflective, situated, and responsible informational practice (

Table 5. System of relationships between competencies and MIL dimensions.

Competences	Risks Without MIL-AI	MIL-IA Competencies	Indicators
AI-assisted writing	Loss of own voice, sophisticated plagiarism, and expressive dependence	Critical discernment	Hybridization awareness: recognizing what each agent contributes.
		Academic Integrity	Evaluating the authenticity of information to mitigate impersonation. Ethics of information transparency. Preservation of discursive identities.
Autonomous and participatory learning management	Fragmentation of attention, illusion of learning, intolerance of uncertainty.	Critical discernment	Augmented metacognition distinguishes between deep and shallow learning.
		Cognitive autonomy	Dependency assessment to recognize when too much is being delegated to the AI. Preservation of capabilities by de-intermediating all skills.
		Qualitative management	Critical Healing: Do not accept every AI response as an irrefutable truth.
Facilitating AI-assisted scholarly production	Quantity over quality as the norm, performance anxiety, and devaluation of effort.	Critical discernment	Temporal discernment to recognize when speed is the enemy of quality.
		Cognitive autonomy	Resisting productivity pressures. Valuing the process: Recognizing the value of effort and cognitive struggle.
		Qualitative management	Evaluating the depth of texts beyond their quantity.
Investigative metareflection	Availability bias, delegation of judgment, loss of serendipity, and loss of exhaustiveness.	Critical discernment	Understanding algorithmic biases in information search.
		Academic Integrity	Methodological validation through the verification of the processes suggested by AI.
		Cognitive autonomy	Lateral thinking to find what AI does not suggest
		Qualitative management	Diversification of sources to avoid depending on AI syntheses

).

Delving deeper into the findings, critical discernment competency is presented as the ability to interrogate the origin, internal logic, and consequences generated by AI-mediated cognitive products. It empowers the subject to resist the opacity of generative systems by establishing a reflective and deliberate distance from the content to assess its legitimacy and potential biases. The competent user appeals to methodical doubt to demand explanations, deactivating the uncritical delegation of thought while distinguishing between collaboration and substitution and between expansion and distortion of its capabilities. The second competency identified was academic integrity, oriented toward recognizing hybrid attribution to separate human and algorithmic contributions in discursive production as an expression of intellectual and epistemic honesty. The vision of integrity identified goes beyond the classic approach to plagiarism by recognizing the potential impersonation of epistemic subjectivity experienced in interactions with AI. The novelty of this study lies in the work toward conscious authorship and the maintenance of ethical coherence in the production of knowledge.

The third competency is cognitive autonomy, which is conceived as the subject’s ability to direct, regulate, and sustain their own thought processes in algorithmically mediated environments. It allows for the restoration of the subject’s decision-making centrality in the face of cognitive shortcuts, personalized recommendations, and invisible structures that guide the understanding of the world. In this sense, epistemological sovereignty is created that defends the right to think with others but not in place of oneself. Finally, the fourth competency identified was qualitative information management, which involves the ability to organize, assess, and reconfigure information, knowledge, and cognitive practices based on their relevance, depth, and meaning. Qualitative management has made data interpretative frameworks available to prevent the erosion of the formative meaning of knowledge.

As can be seen, the four identified competencies function as intertwined structures whose central effect is the conscious confrontation of predatory information through the critical interpretation of the informational environment to attack the illusion of veracity and the uncritical delegation of judgment (critical discernment); ethical and responsible production of knowledge while preserving authorship and productive transparency in hybrid environments (academic integrity); regulation and protection of thought in interaction with intelligent systems without falling into the logic of platforms (cognitive autonomy); and organization based on criteria based on the abundance of existing data (qualitative management). From this perspective, citizenship education must integrate these competencies as structural cores. It is assumed that beyond adapting to this AI-mediated informational ecosystem, we must dare to redefine what it means to learn, produce, validate, and share knowledge in a space where humans are not the only producers of meaning but are the only ones who can assume an ethics of thought that justifies educational updating as a civilizing task.

5. Discussion of the Results

It has been shown that the identified competencies do not operate as watertight compartments, but rather as a functional framework that reconfigures what it means to be a student in informational ecologies influenced by AI. Academic agency ceases to be an exclusively individual attribute to become relational and situated: knowing, being, and acting are intertwined in practices where students no longer merely consume content but design, negotiate, and refine meanings with algorithmic support in a logic of human-machine co-determination. This shift is consistent with embodied and relational learning approaches (Juelskjær & Schwennesen, 2012) and the traits of Generation Z, which predominates in university classrooms today (Chan & Lee, 2023). Therefore, it is no coincidence that students recognize and value AI in their educational journeys because they perceive improvements in performance and results (Yusuf et al., 2024; Selim, 2024). The novelty lies not in adding labels but in showing how these practices are integrated into everyday experience to produce unprecedented academic autonomy and, at the same time, new vulnerabilities.

In operational terms, writing, self-managing the study, and producing and designing research were co-involved. While writing with AI support, students clarify doubts, reorganize information and time, assemble deliverables with less friction, and audit their method decisions. First-person evidence (“I can search, filter, and synthesize information from multiple sources simultaneously,” score 126.65) captures extended cognition: part of the processing is conducted in networks of tools and corpora, and the subject reallocates effort to interpret and decide, configuring a distributed academic subjectivity (Kim et al., 2025). In this context, posthumanism was used as an analytical lens (not as the participants’ self-identity) to describe these distributed agencies and their pragmatic assemblages (Haraway, 2022; Braidotti, 2016, 2017, 2019). In this context, student identity becomes performative and multiple: the same subject alternates roles (textual refiner, director of cognitive processes, time manager, epistemic navigator, cautious critic, problem solver) depending on the task and context, in line with the notions of performativity and multiplicity of identity (Butler, 2015; Kenny, 2019).

However, the gain in fluidity entails tensions that must be precisely identified. In the field of writing, AI-assisted co-authorship democratizes expressiveness and raises rhetorical standards, but it also exposes the subject to the porosity between human thought and algorithmic assembly: there is a risk of navigating between multiple textual agencies with weakened theoretical understanding (Heil et al., 2025), displacing the work on one’s own voice, or enabling inappropriate shortcuts (Shata & Hartley, 2025; Krause et al., 2025). In the sphere of production and work organization, the acceleration of delivery reconfigures academic temporality; the mechanical is delegated to iterate ideas quickly, but the overabundance of information and permanent availability establish an attention economy that displaces the maturation of projects and transforms flexibility into attentional precariousness (Eriksen, 2001; Krause et al., 2025). The result is a plausible risk of more output with less knowledge (Heil et al., 2025) and technological codependency that is difficult for those who feel effective to recognize (Rosas-Meléndez et al., 2025). This is not about abandoning AI but rather conditioning its use with rules of integrity, verification, and explicit limits on delegation.

It is precisely here that the contribution of this research critically engages with established frameworks such as DigComp and the MIL/UNESCO guidelines. The aim is not to contrast new categories with existing ones; rather, an operational re-specification is proposed for scenarios with generative models that shift the focus from what to how competence is exercised and verified. Thus, informational evaluation requires discernment with an awareness of the model: recognizing the probabilistic nature of outputs, detecting hallucinations, verifying provenance and dating, and externally triangulating before drawing conclusions. Content creation demands orchestrated authorship: iterative prompt design, toolchaining, version traceability, and a declaration of collaboration with AI to maintain discursive accountability. Learning to learn requires strengthening cognitive autonomy: non-delegation thresholds to protect core practices (e.g., problem posing, disciplinary argumentation) and documentation of delegation when it occurs, so that the learner can explain their decisions, why, and on what criteria. Finally, problem solving translates into methodological auditing: justifying choices, comparing them with disciplinary criteria, and reviewing procedural biases in the design-execution-evaluation cycles. With this re-specification, the frameworks cease to be agnostic lists and become teachable and auditable practices of media and information literacy for AI (MIL-AI) tailored to the uses we actually observe.

This training framework aligns with recent proposals advocating for critically aware, ethical, and autonomous AI (Ndungu, 2024; Ranieri et al., 2024; Saliu, 2024; Ye & Mahizer, 2025) and with the idea of augmented cognitive citizenship that articulates individual agency and public responsibility (Marushchak et al., 2024; Sriprakash et al., 2025). If AI expands access to 24/7 mentoring, facilitates reformulation with one’s own voice, and offers methodological guidance, it is up to institutions to institutionalize these practices so that capacity expansion is transformed into situated, accountable, and verifiable knowledge. This implies curricula capable of establishing usage criteria, non-delegation thresholds, traceability of collaboration, and standards of qualitative judgment that prevail over speed. In the absence of these conditions, AI tends to confuse access with understanding, speed with deep learning, and delegation with autonomy. Under these conditions, the university can act as an ethical laboratory where one learns to inhabit technology judiciously, decide when not to delegate, and sustain the authorship of one’s thinking in increasingly algorithmic informational ecologies.

Therefore, the key is not to celebrate a supposed automatic emancipation or to denounce a homogeneous threat, but to accurately describe the intertwined behavior of competencies in educational interaction and translate it into pedagogical design. The data show students negotiating with AI: they use it to explore problems, organize information, assemble products, and justify decisions, but they also condition their recommendations and delimit areas of non-delegation. In this daily negotiation, the student’s role is redefined beyond what DigComp predicts, because competence ceases to be a label and becomes a verifiable practice regime: knowing what to do, how to do it with AI, when not to do it with AI, and how to be accountable for the process of learning. This is, ultimately, the argument we propose to the community: AI amplifies but does not replace. Where students have strong MIL-AI, the expansion of capabilities becomes deep learning, and where it is not, the same expansion results in empty efficiency or heteronomy disguised as autonomy. The pedagogical decision is no longer whether or not to incorporate AI, but rather how to train students to inhabit it with responsibility, integrity, and judgment, in line with the epistemological and ethical challenges of contemporary higher education.

6. Conclusions

This study demonstrated that students are developing new types of AI-supported academic interactions that, in their view, enhance performance and sense of control. These practices were grouped into four competencies: assisted authorship/writing, autonomous learning management, assisted academic production, and methodological meta-reflection. The formative value of these competencies lies not in their technical power but in how they are critically appropriated by the user. Consequently, the proposed MIL-AI approach (critical discernment, academic integrity, cognitive autonomy, and qualitative management) serves as a framework for transforming these practices into meaningful learning experiences rather than sophisticated forms of academic simulations.

It is further argued that AI transforms both the present action and the future of the academic subject: a distributed subjectivity emerges in which decisions, mediations, and technologies co-determine intellectual work. In curricular terms, critical discernment must translate into verifiable results and evidence: detection of hallucinations, verification of the provenance and dating of outputs, and triangulation with independent sources. Its implementation is straightforward: (i) controlled hallucination tests (designing prompts to force errors and documenting external verification); (ii) mandatory output sheets for each submission (model/version, relevant parameters, date/time, sources cited, verification steps); and (iii) comparative resolution of the same problem with at least two human/primary sources and one AI output, explaining the convergences, conflict, and time lags. Academic integrity has become traceable: declaration of collaboration with AI (what was delegated, at what stage, and with what justification) and version control of the text and references.

To avoid codependency and preserve core practices, cognitive autonomy requires explicit thresholds for non-use (e.g., problematization, disciplinary argumentation, reading primary sources), “no AI” tasks at key moments, and delegation contracts that document the reason, scope, and impact of algorithmic support. Finally, qualitative information management demands specific safeguards in AI-rich contexts: data chain of custody (origin, context, and date of each citation), assisted and audited coding (AI suggestions; the team compares with human coding, calculates agreement, and writes reflective memos about discrepancies), assisted but justified theoretical sampling (inclusion/exclusion criteria and potential biases), and triangulation maps that link citation → code → category → conclusion. These practices should be weighted in course rubrics (accuracy, traceability, conflict resolution, declaration of uncertainty) so that the assessment incentivizes the epistemic behavior we declare desirable.

Based on the above, it is assumed that universities should not “domesticate” technology but rather teach students how to navigate it with discernment. When the four student competencies align with teachable and auditable MIL-AI practices, the expansion of capabilities reported by students becomes situated, responsible, and verifiable knowledge. Without this grounding, the same expansion risks confusing access with understanding, speed with deep learning, and delegation with autonomy. Therefore, the institutional responsibility is to design curricula that make discernment, integrity, autonomy, and qualitative management visible and enforceable, ensuring that AI serves as a tool for better thinking rather than a shortcut that diminishes training.