‘Now There Is Somebody I Can Go to, Although It’s an AI’: Evaluating Acceptance and Use of Obruche, a Pilot Chatbot to Prevent Power Asymmetries in Cross-Border Journalism Teams

Abstract

This exploratory study examines how journalists in/coordinating investigations use a chatbot designed to reduce power asymmetries during remote work. Twelve freelancers across Africa, Europe, and India tested Obruche, a chatbot advisor covering risk mitigation, pay equality, tension de-escalation, and intellectual property protection. Drawing on the Unified Theory of Acceptance and Use of Technology, semi-structured interviews were coded for Performance Expectancy, Effort Expectancy, Facilitating Conditions, and Social Influence. Results show journalists gravitate towards chatbots that are cognisant of their location-specific challenges and able to provide information that facilitates access to media outlets or peers for future collaborations. Next-best-action responses that expanded user queries or offered role-play scenarios also left journalists feeling supported, less lonely, and not judged. However, the chatbot’s female persona, scepticism of artificial intelligence, and chatbot novelty may reduce user acceptance. Obruche’s potential areas of intervention are linked to eight types of organisational power. The chatbot mainly assisted journalists to confront or rebalance Control of Knowledge and Information, and Control of Scarce Resources, aiding users’ Ability to Cope with Uncertainty. This research contributes to recent qualitative studies on journalists’ well-being by demonstrating how chatbots can mitigate power imbalances between dispersed teams of journalists. The benefits and concerns presented may inform future designs of similar team-mediation chatbots.

1. Introduction

Journalists in cross-border collaborations typically work in remote teams across countries, regions, or continents (Cândea, 2021; Ng, 2021). However, more journalists now work remotely as longstanding technology-driven shifts towards digital workspaces collide with job cuts, newsroom shrinkages, dire pay, and the COVID-19 pandemic (Örnebring, 2018; Casero-Ripollés, 2021). These developments have brought both advantageous autonomy and disempowering isolation to journalists (Bunce et al., 2018; Henderson et al., 2022), even as Generative Artificial Intelligence (hereafter GenAI) disrupts journalists’ production, publication, and audience engagement workflows (Cools & Diakopoulos, 2024; Thäsler-Kordonouri & Koliska, 2025; Bradshaw, 2026). This study extends research on AI-supported workflows by exploring how GenAI can support the flow of work between journalists. It assesses how investigative journalists perceive, use, and react to a chatbot that provides advice on risk reduction, pay equality, tension de-escalation, and intellectual property protection. Through Venkatesh et al.’s (2003) Unified Theory of Acceptance and Use of Technology (UTAUT), this article aims to determine the facilitators and barriers for chatbots that tackle the decision-based friction and loss of empathy that accompany remote journalistic work (Ng, 2021; Ivask & Lon, 2023).

Remote teams have received attention as scholarship advances towards journalists’ working conditions and the associated psychosocial consequences (Bélair-Gagnon et al., 2024). Nevertheless, Graham et al. (2024, p. 4) argue that most research “assumes journalists work in and from physical newsrooms where colleagues can support each other under increasingly stressful conditions.” The reality is that many investigative newsrooms are spaces convened by Zoom meetings, built on Slack or Telegram channels, and run on asynchronous project management tools (Bunce et al., 2018; Cândea, 2021). This trajectory has reduced logistical costs, expanded coverage, and prevented physical harm to investigators (Hoxha et al., 2024; Mitra et al., 2025). Ironically, remote journalistic work also brings heightened risks of digital and professional isolation, increased monitoring, loss of work–life balance, stress, and mental discomfort (Örnebring, 2018; Graham et al., 2024; Cheng, 2025). Borrowing a metaphor from Örnebring’s (2018, p. 120) respondents, the average cross-border journalist’s work–life is a “quilt” of different bosses, time zones, colleagues, and undulating salary levels. These factors predispose remote journalists to a decline in team communication and cohesion (Cherubini et al., 2021), with such gaps cited as reinforcing power hierarchies in remote workspaces (Morgan, 2017; Šimunjak, 2022). Elsewhere, these challenges that dispersed workers face are being surmounted using GenAI conversational robots, or chatbots. This study is inspired by various applications of chatbot technology that may find use in cross-border journalism.

Specifically, chatbots have moved beyond mere tools for information retrieval; in organisational studies, they are social actors used to regulate emotions, decentralise editorial meetings for academic journals, and ensure equitable participation or speedy decision-making in remote teams (Benke, 2020; Tavanapour et al., 2020; Do et al., 2023; Takhsha, 2024; Amiot et al., 2025). To extend these possibilities into cross-border journalism, Obruche, an “advisor” chatbot, was created and tested among 12 freelancers across Africa (n.8), Europe (n.3), and India (n.1). This article focuses on the user experience of the respondents, who are either reporters working in and/or coordinators of cross-border investigations. Through qualitative interviews, the main research question determined what factors influence journalists to use chatbots that give advice, which enhances workplace relations. Respondents’ varied AI use levels, working conditions, and locations brought rich data. By categorising their experiences using the UTAUT (Venkatesh et al., 2003) tenets as codes, novel insights emerged on how personal proclivities and societal factors affect investigative journalists’ use of work-related advisory chatbots.

Furthermore, during collaborative journalism processes, everyone does not have equal power (Ng, 2021), and these power asymmetries are rarely addressed (Waisbord, 2025). Therefore, the chatbot’s potential to resolve power asymmetries was also explored by analysing which sources of power (Morgan, 2017, p. 244) appear within the topics or issues that journalists input when using a GenAI advisor.

The following literature review discusses the place of chatbots and power in investigative journalism, the rationale behind using UTAUT for data analysis, and a brief rundown of the chatbot Obruche. Thereafter, the methods and detailed results are presented; both are followed by a discussion of the findings and study limitations.

1.1. GenAI, Chatbots, and Investigative Journalism

Defining Artificial Intelligence (AI) tools and their properties is a constantly evolving practice (Venkatesh, 2022). As such, this paper defines Generative Artificial Intelligence (GenAI) as any technology that relies on training data to create new content (audio, videos, text, images, or other media) in response to written prompts from a user (Cools & Diakopoulos, 2024, p. 1). Worldwide, GenAI has sparked a proliferation of tools that serve and disrupt various fields of human endeavour, including journalism (Lewis et al., 2025; Møller et al., 2025). Chatbots, which are GenAI-based agents that simulate natural language conversation with users (Kotenidis & Veglis, 2021), are a hallmark of journalists’ production, publication, and audience engagement workflows (Cifliku & Heuer, 2025; Thäsler-Kordonouri & Koliska, 2025). This study extends research on AI-supported workflows into GenAI supporting the flow of work, i.e., professional relations between journalists. It explores how investigative journalists perceive, use, and react to a chatbot that provides advice on risk reduction, pay equality, tension de-escalation, and intellectual property protection.

Two distinct phases of chatbot infusion into journalism exist. At first, “event-driven automated journalism” relied on Natural Language Generation to publish reports from vast caches of data (Caswell & Dörr, 2018, p. 477). These authors cited the Los Angeles Times newsroom’s “Quakebot”, which generated breaking-news posts on earthquakes. Such one-way distribution systems did not permit user dialogues (Graefe, 2016), a trend that changed with the second phase of “chatbot journalism” (Veglis & Maniou, 2019, p. 3). Here, the emergence of messaging apps led to the introduction of interactive newsbots by large-resource newsrooms like The Guardian; users could now “chat” by choosing topics of interest whilst in Facebook Messenger (Veglis & Maniou, 2019). A ten-year review by Ioscote et al. (2024, p. 876) sums up the trajectory: “automation of journalistic writing” started around the 2010s, diversified from 2018 into data-heavy tasks such as financial and sports writing, before tending towards GenAI in newsrooms in 2022, sparked by the launch of ChatGPT. Discourse has also taken a critical turn, based on systemic reviews of research on AI adoption in newsrooms. Sonni et al. (2024) found that most studies list training bias, lack of algorithmic transparency, and reduced levels of nuance and context in GenAI articles as major concerns.

Beyond newsroom-wide deployment, scholarly attention has also scrutinised journalists’ individual adaptations to and of AI (Gondwe, 2023; Dodds et al., 2025). Here, AI use is not about reporters from newsrooms using tools endorsed by management. Rather, it is about self-motivation; individual investigative journalists choosing to infuse AI into their unique workflows to the best of their personal and technological capabilities (Munoriyarwa, 2024; Thäsler-Kordonouri & Koliska, 2025; Kunert, 2025). Investigators use GenAI for querying massive data sets, comprehending legal documents, text-based coding, web-scraping, monitoring databases, filtering story angles, documenting their investigative process, and for verification, writing, or editing purposes (Cools & Diakopoulos, 2024; Reiter, 2024; Cifliku & Heuer, 2025; Fridman et al., 2025). Also noteworthy is Bradshaw (2026), who cites the same broad usage patterns, using metaphors that illustrate this study’s rationale. He lists three metaphors to describe the roles that GenAI chatbots currently play in the work lives of investigative journalists. First, he uses the metaphor “co-pilot” to conceptualise the relationship between the journalist and the chatbot, cites chatbots as a “technical bridge” to surmount disparity in computational knowledge, before declaring them a “brainstorming partner” in story sourcing and ideation (Bradshaw, 2026, pp. 210–216).

This study extrapolates these contexts to argue: if the chatbot can be a co-pilot steering the journalist’s relations with the journalistic product, could it be a co-pilot assisting the journalist’s relations with journalistic peers? If a chatbot is a bridge that helps journalists surmount technical data, could it help surmount tensions between colleagues, rather than tensions within datasets? And finally, if the chatbot can be a brainstorming partner for achieving the ideal story, could it help journalists achieve the ideal partnership? These reflections are motivated by the current context where communication studies show a “predominance of AI as a journalistic tool,” (Ioscote et al., 2024, p. 887), yet scholars need “to capture the totality of GenAI’s potential to complicate journalism and what it means to communicate as a human in the era of AI” (Lewis et al., 2025, p. 2). Heeding this call, this study captures GenAI’s possible usage in enhancing relations between cross-border journalists.

Although journalism studies focus on AI for external tasks like audience engagement, information gathering and data synthesis, a gap exists for using internal AI for managing team relations, as seen in Human–Computer Interaction (HCI) research. As digitalised virtual work by dispersed colleagues increases, conflicts are inevitable when social cues become imperceptible across screens, hierarchy-based decision bias occurs in silos, with workflow friction worsened by unequal participation in meetings (Benke, 2020; Tavanapour et al., 2020; da Silveira Colissi et al., 2021; Do et al., 2023). These HCI researchers have used chatbots to reverse such trends, and the present study attempts to replicate their solution for similar tensions within journalism work (Henderson et al., 2022; Bélair-Gagnon et al., 2024). My attempt is hinged on the rationale that unresolved conflicts boost power asymmetries in the workplace and new technology—in this case, chatbots—can alter the balance of power at work (Morgan, 2017).

1.2. Power and Cross-Border Investigative Journalism

Power exists in relationships of any sort and is exerted as control or influence, with good or bad consequences (Gray et al., 2022). In work-based relations, power influences who gets what, when, and how (Morgan, 2017, p. 243). Power is also a recurring theme in collaborative investigative journalism and AI. Although this study focuses on the former, it is worth noting that AI comes with a unique paradox of power. Investigative journalists investigate AI systems, holding those who deploy them accountable, yet they rely on these same systems to accomplish their watchdog role (Kunert, 2025). Simultaneously, these AI systems exert power on journalistic output, sparking caveats for human oversight (Lewis & Westlund, 2015; Thäsler-Kordonouri & Koliska, 2025). A third angle is the power wielded by the technology industry–owners of the AI systems–who are a new set of powerful actors that journalists and journalism have to contend with (Munoriyarwa & de-Lima-Santos, 2025).

Restating this paper’s focus, Ng (2021) notes that questions exist about the dynamics that facilitate successful journalistic collaborations and cites power among such queries. Since then, a few studies have directly examined power dynamics within cross-border journalism practice, revealing that structural inequalities, disparate risk distribution, resilience asymmetry and denied identities affect journalists’ psychological, financial, digital, and physical safety. To unpack these studies, power asymmetries appear between unlicensed journalists and state-licensed journalists, where the former depend on the latter for cultural and political authority (Yin & Zheng, 2024). Westwards, resilience asymmetry is identified within US–Mexico collaborations, where resource disparities mean journalists in the Global South require higher psychological endurance to manoeuvre the same collaborations as their Global North colleagues (Cheas, 2025). More widespread is disparity in risks, when local journalists are exposed to the bulk of legal and physical risks while their international colleagues are protected by distance and the might of their brand-name newsrooms (Mesquita et al., 2025). Finally, the overarching dynamic is structural inequalities that result in chequered access to journalism grants: colleagues who control funding dictate the editorial agenda and safety protocols of the collaboration (Waisbord, 2025). While scholars unearth these different iterations of power in collaborative journalism, it may be worth exploring the possibility of interventions via chatbots, particularly as chatbots are already gaining traction in journalism workspaces.

This study attempts the challenge, by determining if a chatbot can provide information that prevents journalists from encountering identified power asymmetries in collaborations, or guides journalists already experiencing the same. Specifically, this author’s prior work utilised Morgan’s (2017) Organisational Theory to document 15 sources of systemic and episodic power as recounted by journalists in cross-border collaborations (Meyer, 2025). These sources of power are: Control of Knowledge and Information; Control of Scarce Resources; Control of Boundaries; Ability to Cope with Uncertainty; Control of Decision Processes; Formal Authority; Control of Technology; Management of Gender Relations; Use of Organisational Structure; Control of Counter-organisations; Interpersonal Alliances and Control of Informal Organisation; Symbolism and Management of Meaning; Structural Factors That Define Action; The Power one Already Has; and Control of Cultural Norms.

Table 1. Sources of Power in Networked Investigative Cross-Border Collaborations.

Source of Power	Characteristics	Power Breach Scenarios
Control of Knowledge and Information	Ability to influence the content, timing, and direction of information flows	Withholding information about available training opportunities from some colleagues, not including reporters in pre-publication editing of their own investigations
Control of Scarce Resources	Ability to control funding, personnel, and sought-after goods or services relevant to the workplace or work tasks	Payment arrangements that leave some colleagues waiting longer for smaller amounts of money, compared to their peers working on the same investigation
Control of Boundaries	Ability to control access, integration, or isolation between actors within and external to the workplace	Preventing colleagues from accessing specific people and entities that may lead to further work opportunities
Ability to Cope with Uncertainty	Possessing enough influence to deal with or avoid unpredictable situations that are directly or indirectly related to accomplishing work tasks	Downplaying unique risks faced by colleagues because attending to such risks would affect production timelines; a work arrangement where colleagues cannot access definite solutions to navigate unique risks
Control of Decision Processes	Ability to influence where, when, and how decisions that bind all workers are made, including controlling the outcomes of decision-making agenda, processes, and issues.	One actor or a few actors setting meeting venues, times, and formats that repeatedly disadvantage or exclude most colleagues
Formal Authority	Title-based, role-dependent, and team-recognised power	Conflict over production processes and byline allocation done by the authority actor
Structural Factors that Define the Stage of Action	Historical, geopolitical, and other underlying factors embedded in wider society that determine who emerges as influential in the workplace and what roles they occupy	Pre-determined working conditions and budgets offered by coordinators are unfavourable for colleagues on the ground
The Power One Already Has	Ability to use existing influence to attract more power; enhanced when the actor is of a demographic, social, or professional group that is perceived or accepted by others as more influential	The appearance of influential colleagues in certain reporting or filming sites puts their less powerful colleagues in danger
Symbolism and the Management of Meaning	Ability to influence the structures that boost or retard colleagues’ professional visibility and reality (ceremonies, language, appearances)	Exclusion of colleagues from attending career-advancing events (conferences, award ceremonies); Veteran colleagues denouncing the skills and techniques of perceived younger colleagues on the sidelines of career-enhancing events
Control of Technology	The power to introduce, adjust, and control any technology used for work	Asking team members to communicate or produce investigations using an application that may be dangerous or too expensive for some colleagues to access.
Interpersonal Alliances, Networks and Control of Informal Organisation	The power that comes from having interlinked social circles within the profession, which brings advanced information and provides backing for personal interests	Influential actors and their social circle blacklisting colleagues from working or accessing career-enhancing opportunities
Gender and the Management of Gender Relations	The power to include or exclude colleagues based on gender, or to influence work relations based on gender-related issues	Consistently relegating female journalists to support tasks and tasks without bylines
Use of Organisational Structure, Rules and Regulations	The ability to proactively/retrospectively create, readjust, or use rules and regulations to advance the actors’ interests or the group’s interests	Influential colleagues maliciously invoke one country or continent’s journalism ethics to prevent context-relevant publishing for other countries during a cross-continental investigation
Control of Counter organisations	The ability to create and control an entity that is not part of the established power structure in the wider profession.	A regional association of journalists was established as a response to non-inclusion that ends up also excluding journalists in one form or another
Control of Cultural Norms	Power to promote or downplay ethnicity, social, or faith-based beliefs held by individual colleagues	Non-recognition or non-prioritisation of heritage-based mental health interventions for relevant colleagues at risk

provides details on each type of power and examples of what a breach of these powers looks like in collaborative investigative journalism.

These sources of power enable workers to advance their interests, with consequences that either perpetuate or resolve workplace conflict (Morgan, 2017, p. 244). In journalism, when a power source is breached, an imbalance results between colleagues, impacting the financial, physical, psychological, or digital safety of whoever holds less or no power. For example, a breach of Formal Authority occurs if an editor insists that a freelancer interview politicians who have threatened to harm the freelancer during previous investigations. This threatens the freelancer’s physical safety and possibly their financial safety if the collaboration is terminated by the editor. A rebalance of power may occur if the freelancer seeks the intervention of a higher intermediary (e.g., a network coordinator, funder, or project manager). This study aims to explore whether a chatbot could intervene by providing advice to prevent or resolve power breaches, resulting in the development of a chatbot named Obruche.

2. Study Background

2.1. AI Assistant Obruche

From 2018 to 2023, this author researched journalists’ concerns about remote investigative journalism production and power asymmetries between journalists in transcontinental teams (Meyer, 2019, 2025). This present study advances that work by taking an exploratory approach to ascertain if/how a chatbot could address these identified challenges to cross-border work. The chatbot for this study, Obruche, was conceived based on specific power imbalance incidents detailed by journalists (Meyer, 2025; pp. 807–815). Obruche was created using Jotform, a no-code AI Agent template provider, and the AI service used by Jotform is based on OpenAI’s GPT-4 (Jotform Inc., 2026). Chatbots based on GPT-4 (i.e., Generative Pre-Trained Transformer 4) access vast datasets and can deliver basic advisory services across various professions, though subject to hallucinations (Fieberg et al., 2023; Nori et al., 2023; Tang et al., 2024). Jotform was chosen because it allows for manual and informal testing of response validity via a web chat console (Gao et al., 2025) in training mode (see Supplementary Material for example screenshots). Manual augmentation of the existing GPT-4 database can also be achieved by uploading PDFs, weblinks, question-and-answer entries, and other types of written text. This author distilled the findings from the earlier-referenced research into written question-and-answer formats, quizzes, and role-play scenarios. These were then uploaded onto Obruche’s backend; some topics are shown as follows in Figure 1.

From pregnant women in Indonesia to young adults and nutritionists in the United Kingdom (Puspitasari et al., 2022; Ashton et al., 2023), individuals or experts with lived or professional experience are routinely qualitatively surveyed for information that forms the knowledge base of chatbots. Likewise, this study used prior qualitative data from journalists to achieve two types of technical validation for chatbots as defined by Hua et al. (2025, pp. 384–385). The first stage was “foundational bench testing,” where converting these expert insights into scripted tests and scenarios ensured Obruche met “baseline functional standards” for exploratory chatbot research. In addition, recruiting a second, new set of journalists to “assess usability and acceptability” of the chatbot brought this work into what Hua et al. (2025) call “pilot feasibility stage.” Just as experts generate questions for sector-relevant chatbots to rate the validity of such chatbots (Goodman et al., 2023), similarly, this recruitment of domain-expert users (i.e., working journalists) tends towards organic validation of the utility and role-fidelity of Obruche. Finally, for chatbots that are not expected to achieve clinical efficacy or handle mortally sensitive information, basic functional stability and user experience over short periods is acceptable (Hua et al., 2025), although limitations exist, as discussed in Section 6.

Obruche is an AI Assistant, a chatbot that provides direct help to users through conversational interfaces (Veglis & Maniou, 2019); screenshots of Obruche’s interface are in Figure 2. A warning was also included on Obruche’s landing page to prevent potential users from revealing identifying information. Furthermore, Obruche’s voice function was disabled to prevent (un)intentional voice training by any host companies.

This chatbot was completely self-funded by the author and developed in 2025, based on reports from journalists between 2018 and 2023. No journalists from the author’s previous work participated in this study. Aside from Jotform’s innate GPT-4 general database, the additional data on which the chatbot was trained is the author’s prior field notes, interview transcripts, and her findings from earlier-cited studies. Access to the author’s knowledge base is available upon reasonable and written request.

2.2. UTAUT as Analytical Framework

The UTAUT was used as a standardised guide to analyse journalists’ responses to an AI chatbot designed to prevent and reduce power imbalances in cross-border investigations.

From healthcare to education and business, the UTAUT is a framework for determining why users embrace or reject chatbots and which factors influence these reactions (Peng et al., 2022; Massa et al., 2023; Menon & Shilpa, 2023; Tian et al., 2024; Bianchini et al., 2025). Developed by Venkatesh et al. (2003), the UTAUT has four pillars, which researchers use as metrics: Performance Expectancy (PE), Effort Expectancy (EE), Social Influence (SI), and Facilitating Conditions (FC). The first pillar, PE, refers to the usefulness of the chatbot to the individual’s circumstances (professional or personal), while EE covers ease of use. Social Influence is used to gauge the effect of external networks on the individual’s intention to use the chatbot, and FC is the available infrastructure and support that influences chatbot use.

The main chatbot types being studied are multipurpose professional-to-personal assistants like ChatGPT (Menon & Shilpa, 2023; Tian et al., 2024) and voice-activated agents (e.g., Google Assistant, Alexa), which have been assessed for their ability to induce daily habits (García de Blanes Sebastián et al., 2022). Research trends also show a focus on specialised support chatbots that assist with mental health or symptom checks, HIV awareness and medication support, as well as financial advisory support (Peng et al., 2022; Massa et al., 2023; Wutz et al., 2023; Othayoth & Khanna, 2025). Chatbots that enable collaboration, assisting, or advising knowledge workers by facilitating meetings, smart scheduling routines, and project management have also been assessed (Tavanapour et al., 2020; Benke, 2020; Venkatesh, 2022; da Silveira Colissi et al., 2021; Bianchini et al., 2025).

Related to this is an emerging trend within UTAUT-led chatbot studies, where researchers extend the original UTAUT model by adding new variables. For example, in healthcare and insurance circles, Trust in a chatbot is cited as an influential factor when sensitive information breaches or hallucinations are perceived as likely (Wutz et al., 2023; Chen & Zhang, 2025; de Andrés-Sánchez & Gené-Albesa, 2025). A somewhat opposite variable is Personal Innovativeness (PI), where users with a high PI ignore information breach risks and use the chatbot purely for its novelty (Bianchini et al., 2025). To the best of this author’s knowledge, the UTAUT has not been used to assess investigative journalists’ use of advisory chatbots.

The UTAUT was created before mainstream GenAI chatbots launched and was used to assess technology adoption or non-adoption by employees in organisational settings (Venkatesh, 2022). Recent research trends depart from this path. Specific to chatbot studies, the UTAUT tenets have been used to code qualitative responses from participants who do not work in organisations—e.g., students, and self-employed individuals using ChatGPT (Menon & Shilpa, 2023; Alghalbie et al., 2026). Customer experiences whilst using a bank’s chatbot for personal reasons (Mogaji et al., 2021) have been coded with the UTAUT. Finally, it has served as pre-determined codes to assess responses to proposed, non-existent chatbots (Peng et al., 2022) and to code responses during beta-testing of exploratory, non-work-related chatbots (Cheah et al., 2024).

These research techniques, where the UTAUT becomes qualitative codes in chatbot studies, appear common in medical, financial, and educational sectors, but are yet to be explored in journalism chatbot studies. As summarised in Figure 3, the rationale for using the UTAUT for data analysis in this study is that it allows comparison of a chatbot’s adoption/non-adoption by journalists across locations and cultures because of its discrete categories and standardised vocabulary. Secondly, when its pillars are used as codes, the UTAUT prioritises chatbot design from a user perspective (Peng et al., 2022), which means the findings of this study can provide insights for anyone aiming to design similar GenAI assistants for journalists.

Applying the UTAUT framework, the first research question was:

RQ1. 

What factors influence journalists’ intentions to use and accept a chatbot that gives advice on cross-border collaborations?

As the broader rationale for Obruche was power, how it originates within collaborations and affects the flow of work between colleagues, the second research question was:

RQ2. 

On which sources of power could a GenAI chatbot advisor assist investigative journalists working in cross-border collaborations?

3. Methods

3.1. Participant Recruitment

Respondents were recruited through an advertised poster on social media. This open call approach was undertaken to ensure respondents are self-motivated; by initiating interest to participate, they perceive the chatbot as relevant to their circumstances. An open call also excluded journalists from the author’s previous research. The inclusion criteria were: (1) being a journalist who routinely works on cross-border investigations, (2) being a journalist who routinely coordinates investigative teams remotely, and (3) being comfortable with using a chatbot in English. All expressions of interest were replied to with an email containing details of the study. Written consent was first obtained before participants received a general link to access Obruche at their leisure. Each participant was allowed as much time as they required to use the chatbot, after which they communicated readiness for a semi-structured interview. No specific times or days were chosen for respondents to use this chatbot. Participants were in full control of deciding when or how long to use Obruche for, and what to use it for. Thereafter, participants could contact the author whenever they felt willing to proceed. The decision to proceed to interview was initiated by each participant and took days to weeks after chatbot use. The author did not conduct think-aloud sessions or focus group interviews to avoid researcher prompts influencing participants during testing (McDonald et al., 2020) and to avoid groupthink during interviews (MacDougall & Baum, 1997). These measures ensured respondents used Obruche at their own pace, for their own personal circumstances, and of their own volition. This is closer to a real-world user scenario, unlike studies that pin users to researcher-decided time or where researchers unintentionally influence users by assigning chatbot tasks (Cannanure et al., 2025). Interviews were conducted remotely, using Zoom video (n.11) and WhatsApp voicenotes (n.1) between November and December 2025, with each lasting between 35 min and 90 min.

Participant numbers were influenced by the need to achieve data saturation whilst minimising the environmental impact of chatbot use. Early saturation can be expected in studies where pre-determined codes guide the researcher and the interviewee population is homogenous (Buckley, 2022; Mwita, 2022; Hennink & Kaiser, 2022), which was the case here. This study used the UTAUT as pre-determined codes, and participants were all investigative journalists. Furthermore, one-on-one interviews were chosen rather than focus groups because the former yields more data (Schuster et al., 2022), and saturation was considered as achieved when new information or themes ceased to emerge (Guest et al., 2006). For theme-based data saturation, Wutich et al. (2024, p. 3) estimate a sample size from nine interviews as sufficient, while similar qualitative studies for beta-testing of chatbots interviewed between 6 and 14 users (Griffin et al., 2023; Cheah et al., 2024; Salyers et al., 2025). Malterud et al. (2016) advise that adequate qualitative studies are traditionally assessed on the depth and richness within a dataset, as opposed to size. Finally, qualitative analysis prioritises nuance and complexity, requiring in-depth engagement with data items; the result is that qualitative datasets are often smaller than quantitative norms (Braun & Clarke, 2016).

This study was approved by the Independent Research Ethics Committee, UK (ID-IREC023) and is anonymous. All interviews were recorded, transcribed, and stored on a secure computer, which only the researcher could access. Participants received a token of €15 after the interview. The ethics committee noted this sum as adequate for contributing towards participants’ internet costs because internet access was required to use the chatbot and participate in interviews. During virtual data collection, respondents provide the infrastructure for participation, unlike face-to-face studies, where the researcher arranges venues or travels to participants. This study heeds recommendations for researchers to cover direct costs of participation and costs incurred during participation (Nyangulu et al., 2019; Stellenbosch University, 2025). The sum was modest because, as Nyangulu et al. (2019, p. 2) note, any offered participation fee that is unjustified and significantly excessive tends towards inducement, which increases biased responses. The €15 fee was justified for internet costs and time spent, given that typical chatbot testing studies pay €30 to €100 upwards (Mainwaring & Wrapp, 2021; Jörke et al., 2025; Le Bonniec et al., 2025), with some increasing fees based on participants accomplishing specific tasks on the chatbot. None of these larger fees or task-based incentives were used in this study. Also noteworthy are two occurrences that indicate that the €15 was not sufficient to induce respondents’ responses: the fee was declined by two respondents, and the chatbot was freely criticised, as reported in the Results section.

3.2. Data Analysis

Transcripts from the semi-structured interviews were thematically coded (RQ1) according to the four tenets of the UTAUT as shown in Figure 2: Performance expectancy (PE); Effort Expectancy (EE), Facilitating Conditions (FC), and Social Influence (SI). Research Question 2 matched the specific uses described by respondents to the relevant power type. To conclude the interviews, respondents were asked whether the word “Power” resonated with their experience of the chatbot or not.

Each interview transcript was emailed to the relevant participant with the UTAUT codes specified therein, to check if their experiences were accurately interpreted (Vella, 2024, p. 665). This method—member checking—reduces researcher bias and enhances validity because respondents corroborate the researcher’s interpretations, and responses are not coded intuitively but into clear theory-based categories respectively (Buckley, 2022; Mwita, 2022). All participants endorsed the coded themes as categorised by the author.

Participants’ quotes are reported in the Results section, with additional quotes in the Supplementary Material file.

4. Results: Factors That Affect Journalists’ Use of a Chatbot Advisor

Participants were split between Coordinators managing cross-border teams and Journalists working in cross-border teams. Most had over 10 years of experience as investigative journalists, as shown in

Table 2. Participant Demographics.

	Age Range	Gender	Role	Location	AI User Level	Experience
P1	30s	Male	Journalist	Africa	Basic	4 years
P2	30s	Female	Coordinator	Africa	Advanced	8 years
P3	40s	Female	Journalist	Africa	Intermediate	10 years
P4	40s	Female	Coordinator	Africa	Basic	14 years
P5	20s	Female	Journalist	Africa	Intermediate	8 years
P6	40s	Female	Coordinator	Europe	AI Sceptic	25 years
P7	30s	Female	Journalist	Africa	Basic	7 years
P8	20s	Male	Journalist	Africa	Advanced	4 years
P9	40s	Female	Journalist	Africa	Basic	20 years
P10	40s	Female	Coordinator	Europe	Basic	15 years
P11	40s	Male	Journalist	Europe	Intermediate	16 years
P12	30s	Female	Coordinator	Asia	Basic	10 years

.

4.1. Performance Expectancy

Eleven of 12 users were positive about how the chatbot performed, reporting that the information provided by Obruche was helpful, reassuring, or novel. Participants felt that specific information on media outlets, funding resources, and peer contact details would enhance the chatbot’s appeal.

4.1.1. Contributors to Positive Performance Expectancy

The chatbot was found useful because the information shared was actionable and comprehensible. Several respondents stated they used it for ongoing and incoming tasks, expressing satisfaction with the information provided. A recurring query was information about potential collaboration locations. For example, a coordinator sought information on the geopolitical situation of a neighbouring country, while another saved tips on suggested options:

“Me being in [country A], I asked Obruche ‘I want to do an investigation in [country B]. And it gave me a lot of context. I live next to [country B] but I didn’t know. Maybe this is a personal problem, but … the way it was doing it, I don’t know what I expected, but it gave me more than what I expected. I was like, ‘Okay, this is actually actionable.’”

[P2]

“You know when the bot gives options ‘would you like me to do this or that’ I was copying and pasting everything. I wanted to make sure that this time I go for option 1 but the second time I go for option 2 because I want to see what the chatbot will give me in both cases. I copied and saved the tips. I felt I could definitely use a lot of things later on.”

[P12]

A young African freelancer in a similar situation said they no longer felt alone for their next assignment:

“Obruche is telling me I’m not alone in this collaborative investigative journalism work I want to do in [African country]. Now there is somebody I can always go to, although it’s an AI, that I can always reach out to, to have a conversation with.”

[P8]

Another freelancer-coordinator from India felt less judged and supported:

“I felt very seen that I could have all these questions without fear of judgment or without revealing the identity of people—my collaborators, my clients. I could just keep asking it questions. That felt quite relaxing.”

[P12]

Several participants also felt supported in tasks such as budgeting payment for upcoming investigations, creating safety plans, filling job applications, or making ethics-based decisions, and already infused the information received into their work–life:

“Some of us just want the work, experience the credentials, so you don’t think through negotiating or costing collaborations well. It gave me a better perspective. It’s actually something that’s going to live with me. The fundamentals and the basics it gave me, I wrote some of them down; I don’t know if I’m allowed to but henceforth, this is how I’m going to progress with collaborations.”

[P3]

“I asked it to design a safety planning template. I liked the emphasis on self-protection, and ethical clarity. It included things I could immediately do.”

[P1]

“I’m in the middle of applying for something. When I put in a prompt ‘prepare pay proposal and this is the role,’ I put in the role and the description, everything that’s in the job advertisement. That’s where it really gave me what I wanted.”

[P5]

Novelty was a major quality that participants ascribed to the information they received. The chatbot provided new information that users found favourable, with the equal pay calculator (see Figure 4 for the screenshot) reportedly one of the most useful features:

“The pay calculation formula is excellent. It also gave me a sense of ‘ah you’ve done a lot of work for free.’ But hey, it came really timely and it was very helpful.”

[P3]

“It told me ‘if the collaboration is coming from the US, use the US rate.’ This is something that I never thought about. I’ve just been a random freelancer. When I look back on some of the other jobs I’ve done, I think ‘oh no I was underpaid,’ because I was supposed to consider these things.”

[P9]

The follow-up responses embedded as next-best-actions into the chatbot also unearthed new information about copyrights, negotiation tactics, and sourcing during niche investigations:

“You mention you want to work with somebody in [African country], you need more clarification about payment and [the chatbot asks] ‘what about asking about the copyright issues for your photos?’ I was like, “wow.” So, the AI was even suggesting questions I could ask too. I feel the AI is, should I say, compassionate, to do that.”

[P8]

“I asked it to give me story ideas about climate. And it gave ideas about climate, on agriculture. It added sources for me to interview, sources where I can get specific information.”

[P7]

“I was surprised with some of the negotiation tactics and the email drafts or specific sentences that she proposed to use when communicating with publishers. So this was useful for me.”

[P6]

Regarding overall impressions, participants felt they could trust the information provided. While an intermediate user of AI in general [P8] liked “the fact that it doesn’t hallucinate,” a veteran coordinator [P6] reported that: “Obruche really offers serious and totally reliable information. Everything seems very thoroughly fact-checked and reliable,” and reiterated that their opinion was based on their knowledge of, and time spent working in cross-border journalism.

4.1.2. Playful Interactions

Another phenomenon recorded was “Playful Interaction” (Ronagh Nikghalb & Cheng, 2025), where user behaviour is motivated by curiosity, psychological needs, or solely for entertainment purposes. Imitating, Challenging, and Reflection were three types of Playful Interactions that surfaced. Imitating occurs when users ask the chatbot to impersonate someone or something; Challenging is when users input questions that test the chatbot’s limit, while Reflection is when users seek the chatbot’s opinion on a topic (Ronagh Nikghalb & Cheng, 2025, p. 13).

One respondent asked Obruche to imitate a coordinator and had several role-play scenarios:

“With these role-play sessions with the chatbot, now I know what to do, how to say certain things. I’d thought ‘how would it customize responses’ because it doesn’t really know everything about that project or my situation. But what was happening was, it was becoming a little more personal in a nice way. Every contract clause was being explained. Practicing made me feel more confident.” 

[P12]

For P6, a Europe-based coordinator with 25 years’ experience, playful interaction made her realise the chatbot “could offer more,” in terms of Reflection:

“I asked Obruche, ‘what do you think about the concept of equal wage and paying people equal wages no matter the location?’ and she had very thoughtful answers. It’s not just templates or instructions, but it can also give something like an opinion.”

[P6]

This participant also initiated a bias-based Challenge to Obruche:

“I even tested for chauvinism and racism. I said ‘okay there is this stupid guy from eastern Europe, maybe he’s so backward because he’s coming from a backward country called Moldova.’ Obviously, I don’t think that. But Obruche didn’t react like being pissed, but very calmly explained what the problem was, and how to change, and how to approach this differently. So this was also good.”

[P6]

Challenging also occurred based on language, as revealed by a second participant based in Africa, who is a new user of chatbots in general:

“Deliberately, after getting an answer, I said ‘translate to [African language]’ and it gave me a very good translation, honestly. When I got the [African language] version I said ‘translate to English.’ It translated the draft back to English.”

[P4]

When asked to elaborate on the motivation for this behaviour, the participant cited past experiences:

“Using ChatGPT, it is not that I’m condemning another person’s work, but let’s say I’m [names African tribe]… when you ask ChatGPT to translate a document or a paragraph, you don’t just copy and paste, you have to go through it. There are times you just discard some translations because they are not what you requested or what the document or draft is talking about. So I was surprised… The translation is very good. I was impressed with what the app translated, honestly. It switched between languages.”

4.1.3. Desirable Features to Enhance Performance Expectancy

Participants expressed that information on specific individuals, media partners, or laws could increase use of the chatbot. They specified this would mean inclusion of “databases of journalists,” including their “contact details” (P8), as well as funders, and reliable organisations “found to be safe and empathetic” to journalists (P5).

Some participants reported four instances of conversational breakdown. A conversational breakdown happens when “the user’s input is either misinterpreted or not interpreted at all,” resulting in an error response (Braggaar et al., 2023, p. 24). In this study, two types of error responses were described: Implicit Rejection and Out-of-vocabulary Explanation responses. With implicit rejection responses, “the chatbot ignores the error and moves on to a pre-programmed path,” while during out-of-vocabulary explanation responses, “the chatbot highlights the words of the user that caused the error so that the user can rephrase them or express them differently” (Braggaar et al., 2023, p. 26).

In two cases, Implicit rejection error responses were given by the chatbot. The first participant recalled that:

“The AI told me it could connect me with a specific media outlet in [names African country and organisation] but it didn’t provide any answer. I don’t know if that is a network error… I said the AI should connect me with the outlet. I didn’t mention that it should give me the contact details of people. So maybe my prompt was not very explanatory enough for it. Maybe that is why?”

[P8]

The second participant had just begun using the chatbot when the following occurred:

“There appeared to be a glitch and suddenly it addressed me as a freelancer, but I found it very strange because at the very top I had already said I’m a collaboration coordinator.”

[P10]

Two other journalists described how they received what appeared to be out-of-vocabulary explanation responses before achieving their desired response:

“It gave me one response like three times. It was frustrating. I didn’t like the frustration but it also taught me that now be specific about what exactly you want, what exact help you need.”

[P3]

You must be specific on the exact area of the topic you want it to give you information on. Telling Obruche ‘I want to investigate illicit financial flows,’ it said something like: ‘illicit financial flows is a wide topic about oil, human trafficking, drugs, so which part do you mean?’ So you must narrow it down.”

[P7]

4.2. Effort Expectancy

Participants had favourable impressions of Obruche’s name1, design and interface, with most finding it easy to use. Several of them described the responses as detailed and easy to understand, while a few voiced some concerns about the persona of the chatbot being a young woman.

4.2.1. Design Features

Notably, the name Obruche was described as “thoughtful,” and the reason why one participant [P2] shared the chatbot. The landing page was described as “clear and concise,” [P4] the overall tone cited as “conversational” [P5] and “balanced” [P6] while another participant remarked that “Obruche was like that perfect senior colleague that you can consult for guidance” [P3].

Four users switched between roles (reporter or coordinator) when chatting, while another used the chatbot across devices (phone and laptop). Furthermore, six users expressed they had not encountered a chatbot providing the assistance they received from Obruche, with at least one reporting that they see Obruche as an addition:

“When I saw Obruche I was surprised. I couldn’t believe that there will be a generative AI tool that can help journalists in regard to sorting out publishing times, copyrights, contract clauses, and what not when it comes to doing investigative journalism.”

[P8]

“It was a discovery, I felt ‘ah there’s something else that I can use.’”

[P9]

“I had never thought of it until I saw your LinkedIn post [recruiting participants] and I thought oh this is something. This is the first time I ever thought that such a thing existed or has been created.”

[P12]

4.2.2. Chatbot Persona

The majority (10) of the 12 participants found it favourable that the chatbot persona was a woman. One coordinator explained as follows:

“I’m not a gender biased person. But I like that she was a lady and I like that that lady looked very African. That was very important because I’m coming from this background where a lot of content on AI is still very built around what bias exists. Like when you tell Descript, ‘give me images of people in a boardroom,’ then it will be white men in suits. …You have to keep on prompting it, saying, ‘wait, I want a woman too,’ or ‘the woman should also be dressed as if she’s there for business’…coming from that context of using that type of AI, to having this one that actually looks like you. Wow.”

[P2]

Furthermore, a Europe-based participant did not see the chatbot’s gender as a problem because:

“I think in all cultures you have mostly women working for example as social workers or in different shelters giving you advice, so I think it’s quite intuitive that it’s a woman. I don’t think the gender is a problem.”

[P6]

However, these cultural occurrences were the same reason why one participant (also based in Europe) did not think Obruche should have a female persona:

“There are these stereotypes about who is better at managing relations in the collaboration: ‘oh, maybe a woman is better because a woman has a softer touch.’ And it [this chatbot with a female persona] also reminds me of how when technology companies built service robots, they started gendering the robots because they think people doing the mopping and the shopping are women, therefore they put a skirt on the robot.”

[P10]

The second participant, who disapproved of the chatbot persona being female, is based in Africa, and felt men there may not be favourably disposed towards using Obruche:

“What I didn’t like about that lady being there is the fact that because I know, being an African woman, I’m made to overthink everything. I’m made to be sensitive about everything around me. So for me I was like, can a [names own country] man or can an African man trust this chatbot? because there’s this thing of undermining ladies, and there’s this thing of ‘no, I can’t talk with a lady.’ People are now asking personal questions to chatbots. So I thought; can someone ask or talk about things that are personal, very personal to a lady chatbot?’

[P9]

This participant recommended that the female persona is replaced with a “neutral picture that is not a human,” adding that:

I’d choose something that is not gendered for the purpose that it should have wider usage. There should not be any stereotypes associated with the chatbot. If they’re given ideas, it would be like ‘after all it’s a woman who has told me this.’ So to be safe, it’s better to just be neutral.”

[P9]

4.2.3. User Menu

The texts on the landing page and user menu (see Figure 2) also resonated with some participants. According to P5:

“I love that before you start there’s the caveat that you don’t have to give personal identifying information. I don’t think a lot of tools have that. They let you figure it out until it’s too late. So I love that disclaimer; the chatbot was thinking of me, kind of.”

Nevertheless, there also appeared to be conflicting views on whether the introductory text should be removed or expanded, with two users perceiving the listed options as restrictive, in contrast to another who wanted more options:

“It’s good that the chatbot offers suggestions, but if you are offering templates right from the beginning then you may make the user pause, and instead of getting very directed specific information particular to the problem that the user is seeking answers for, then the person may just go to the template, which kind of defeats the purpose of a chatbot. Because a chatbot is supposed to give you answers specific to your problem.”

[P10]

“Right now I feel like I’m getting led by its suggestions; ‘I can help you do this, that’.” 

[P12]

“It can offer much more. So it would be good to put that [more information] immediately in the beginning menu because obviously it can do more.”

[P5]

4.2.4. Responses

Regarding the responses gotten from the chatbot, most users felt its quick and detailed responses increased their engagement time. Participants expressed satisfaction based on “the fact that the chatbot does not hallucinate” [P8], as well as its follow-up prompts being “direct” and “like a thinking partner” [P2].

Specifically, the regional focus of the chatbot’s responses was reported as useful. The swiftness of the chatbot’s responses was also a favourable factor that resonated with two participants:

“It was trying somehow to at least push me: ‘please tell me’ or I’d ask a question and it would say, ‘please tell me where you are based so that I can help you better.’ The other chatbots, some that I’ve used, don’t do that; it just responds.”

[P9]

“It’s a chatbot, but it was quick. Sometimes for some of the other AI platforms you’d have to wait, give some time for the response. This was quicker.”

[P3]

However, one participant (P10) felt the chatbot’s swift responses were overwhelming: “It took me a while to scroll down and I had to read because the information came quite fast and it was a lot.”

4.3. Facilitating Conditions

Participants revealed that increased awareness of what such a chatbot can assist journalists with, and delivering the chatbot in multiple formats, are conditions that could influence its usability.

The analysed data indicates that journalists find the concept of a cross-border journalism advisory GenAI chatbot assistant somewhat new. Although 11 of 12 respondents spontaneously used this chatbot to their expressed satisfaction, some said the starting process could be smoother if information about the chatbot’s capabilities is routinely provided. One participant (P9) felt “old-fashioned” for being unaware of chatbots like Obruche:

“I’ve just been like confined, maybe being old-fashioned. I’m like okay wow this is a kind of AI, are there others I don’t know about? I don’t know the minimum wage, I don’t know how certain things are done. If I didn’t use this AI to help in the process, then I might be ripped off. I might be underpaid.”

A second participant (P5) expressed that being an advanced user of AI did not equal immediate knowledge of this type of chatbot:

“I had to figure the chatbot out, because it’s the first time I‘m even introduced to an idea that something can help me negotiate pay, or collaborations better. So, I also had to think along the lines of how, or what kind of things could such a chatbot help me with?”

Desirable Features to Facilitate Further Use

Pricing, creating offline or data-light versions of the chatbot, and availability of the chatbot on other platforms are factors that participants feel would enhance usability. They specified that having Obruche as an app, on WhatsApp, or embedding it on the websites of existing journalism organisations could boost user access. One participant (P2) recounted how she momentarily forgot Obruche was only available as a hyperlink, not an app:

“You know the way you’ll go to your phone screen and you’re looking for the app Obruche so that you can just click then just pick up where you are? I did that. Then I realized I have to remember the website. But I’m on my phone, so the bookmarks are not even visible. I’m like ‘where is this Obruche again?’ I don’t know why I thought if I used it on my phone it will tell me ‘download app here.’ That’s really why I went to the phone. I’m lobbying for an app.”

Another respondent (P12) felt any new formats should fit in with existing user behaviour:

“People are already trained to talk to bots on WhatsApp. Everyone is working with different phones, some of them are very old, they can’t buy a new one right now. So instead of loading it as another app, the chatbot should just be on WhatsApp.”

Four participants explicitly expressed willingness to pay a reasonable price for the chatbot if that would make it perpetually available (see Supplementary Material).

4.4. Social Influence

Most participants stated Obruche would remain useful post-study and expressed their intention to recommend it to other colleagues. However, also cited were three sociocultural factors that may limit the usage of Obruche by journalists: the gendered persona of the chatbot, AI scepticism, and responses considered as generic.

As mentioned earlier concerning the chatbot’s persona (see Section 4.2.2), the female identity of the chatbot may reduce its appeal for two reasons. First, users with existing cultural bias against women may not want to be advised by a lady, according to earlier quotes from P9. Secondly, users may perceive the chatbot as perpetuating existing stereotypes, according to earlier quotes from P10.

Interestingly, the latter participant gave some insights on how prior knowledge affects usage of a chatbot; they “did not spend long on the chatbot” because the information “wasn’t new”:

“I wondered…whatever question I’m going to ask, is it just going to throw me this standard thing? I didn’t spend long on the chatbot.”

[P10]

Regarding AI scepticism, relevant responses from another participant indicate that previous bias towards chatbots is a notable influence in chatbot adoption, and this bias may never be disclosed, unless the user chooses to:

“Related to Obruche, I have changed my AI skepticism, but not [skepticism] to AI in general. I haven’t seen anything very super beneficial about chatbots, but then with Obruche I need to say I started with negative bias but I have changed my opinion.”

[P6]

In general, 11 of 12 participants reported they would continue using Obruche for advice on cross-border collaborations to estimate their daily rates, budgets, and communication with potential collaborators. Several participants expressed their intention to share the chatbot with others, with one informing the author at their interview that they had already done so:

“I had to forward it to the general manager of one of the networks that I work with, and I said, you need to share this on your platform for the seniors in the newsroom to try and go through this because it’s very insightful”

[P3]

Another participant’s motivation for usage was the chatbot’s context-specific focus:

“It acknowledges the challenges that we African journalists face when having negotiations with international media partners, which can make a person want to use it.”

[P1]

4.5. Sources of Power Where Obruche Can(Not) Intervene

Throughout Section 4.1.1, Section 4.1.2 and Section 4.1.3, participants report they used the chatbot to: fill collaboration job applications (P5); draft contract clauses and personal safety plans (P1, P2); gain insights on copyright issues (P8) and country-specific safety contexts (P2), which are examples of achieving Control of Knowledge and Information. Respondents also used Obruche to calculate pay and estimate daily rates for upcoming projects (P3, P9), indicating attempts to retain Control of Scarce Resources (money, in this case). One respondent tried using the chatbot to surmount Control of Boundaries by seeking names of outlets and individuals to directly work with (P8); they expressed dissatisfaction that Obruche did not provide this. Lastly, several user requests aligned with the power type Ability to Cope With Uncertainty, when respondents sought assistance on negotiation tactics for pay discussions (P6) and handling collaboration termination conversations. Also, through role-play conversations and the tips provided, other respondents felt less alone, less judged, and felt they could cope better, going forward (P8, P12).

Respondents were asked if the word “power” resonated with their experience when using Obruche. One answer suggests the chatbot could intervene by giving coordinators “nuance” when applying their Control of Decision Processes power. According to P2, interacting with Obruche means they will take a different approach to decisions on an investigation they are currently working on:

“Now, my conversations will no longer come from a place of ‘I do not know anything.’ And the editor there will not be like, ‘why are you asking those kinds of questions? In (country) you cannot do this here.’ So it gives me the power of nuance in even the way I frame my questions and how I communicate this to the different teams or editors.”

Specific to the Ability to Cope with Uncertainty, “role playing with the chatbot was powerful” for P12, a journalist who sometimes takes on unpaid informal coordinator roles in addition to their reporting role within the same collaboration. They added that:

“Power was practicing responses, because when you are speaking to a collaboration coordinator, you know that you have already said these words somewhere. You already asked these questions somewhere, so it is no longer unthinkable or undoable.”

The power of Control of Knowledge and Information intersected with Control of Scarce Resources when some journalists used the chatbot. Payments and payment inequality were one area where this occurred, as illustrated by P5, a freelancer with eight years’ experience:

“I was empowered, learning about the day rate in Europe and that I should apply using that day rate because the organization for a thing I’m applying for is based in Europe.”

An avid chatbot user (P3) reports that Obruche “exposed me to my shortfalls, like charging or billing,” adding that:

“I never knew that I had to consider how much people with my level of experience are paid in the region that I’m going to work for. It got me reflecting about the kind of risks I took years back as a young journalist for literally voluntary service. Knowledge is power. I feel more powerful.”

A coordinator (P4) used the chatbot to negotiate Formal Authority, seeking to manage tense relations with other coordinators in the network. They added that they felt ‘just as equal,’ after receiving advice:

“I asked how to make my network visible. About how to deal with a co-coordinator that is self-centered, that holds information from other network members. And I received some tips…those tips made me feel powerful, but not in a way to hurt or sideline members of my network. And through the chatbot, by knowing things I can do to make sure I am safe, I feel I am just as equal as other coordinators too.”

By seeking support on how to increase team members’ professional visibility, this network coordinator also used Obruche to assist in exercising their power of Symbolism and Management of Meaning.

Elsewhere, responses from a veteran (P9) suggest that journalists can rebalance the power of Control of Knowledge and Information by using a chatbot to gather “evidence” to “justify” fairer pay, while their Ability to Cope with Uncertainty may also be increased through tips on how to terminate unequal arrangements:

“When someone is commissioning you, they already have more power than you. Knowing how to justify my rates with evidence, that one makes me powerful, because that’s not something I knew before…This one empowered me with negotiating skills, with how to professionally drop-off from a contract that I feel I cannot do because maybe we are not reaching an agreement based on the rates they are offering.”

Two respondents reinforced these insights into the negative consequences of professional hierarchies within collaborations by citing the inability of chatbots to resolve challenges caused by structural inequality. Both revealed that the power that individual journalists gain through using a chatbot may be displaced by stronger systemic power “imbalances” in the wider world of investigative journalism. According to P6, a Europe-based coordinator:

“Chatbots can help in having different perspectives about collaborations and introduce important perspectives you were not thinking about, but I do think in collaborations where there are these systemic power imbalances, no chatbot can help with this.”

An African-based freelancer, P5 (who previously reported feeling empowered when she learnt about day rates in Europe), says initiators of collaborations need the chatbot more, because they hold more power:

“The more ideal thing is for us to get the information on Obruche to the people, the funders, to actually use this and work their way backwards. Instead of it coming from the point of the lesser powerful party being the one telling them: ‘hey, this is what you pay’ and stuff like that. So I hope it can reach more coordinators; the ones who are commissioning projects, they’re the ones who need to refer to this so that they do not leave people in a weird situation.”

These responses suggest that chatbots targeting freelancers alone cannot resolve two systemic sources of power: Structural Factors that Define Action and The Power One Already Has, as these powers are held by funders and initiators of collaborations.

5. Discussion

This study’s results indicate that journalists feel supported, less lonely, and not judged when using a chatbot that gives advice on cross-border collaborative work. This was demonstrated by how they used Obruche for upcoming projects, pasted job adverts into their chats, listed locations and budgets, as well as copied the chatbot responses for future use. Similar to conclusions from Bianchini et al. (2025, p. 14), those who used the chatbot for discrete tasks reported higher levels of interaction, in this case, spending more time. Users perceiving a chatbot as informational, compatible with their work conditions, and helpful towards accomplishing job-related tasks are developments that align with the concepts of performance expectancy (Venkatesh et al., 2003; Menon & Shilpa, 2023). However, some respondents also expected hyper-personalisation of information related to their peculiar circumstances. They listed helpful additional information, including names and contact details of potential collaborators or equity-focused investigative networks. Given that AI tools that inculcate more stakeholder data/information result in greater errors and bias (Venkatesh, 2022), adding these desired details requires anticipating and resolving the ethical challenges linked to data privacy and training bias.

The novelty of the chatbot appeared to spark two types of Playful Interaction (Ronagh Nikghalb & Cheng, 2025) behaviour in users: Imitation and Challenging. These interactions fall under later adaptations of the UTAUT, specifically Hedonic outcomes; a measure of acceptance where users derive fun or pleasure from technology (Chang, 2012, p. 111). By enacting several role-play scenarios with this chatbot, challenging it with intentionally racist queries or spontaneously asking it to translate back and forth into a second language, individual users deployed Obruche to their individual preferences. Whereas the use of technology for Hedonic outcomes is linked to younger users (Brown & Venkatesh, 2005, p. 415), there was no indication of that in this study. The three ‘hedonic’ respondents were in the oldest age brackets.

Although 11 of 12 users reported the chatbot as easy to use, service failures are inevitable in chatbots (Zhang et al., 2024). Of the four cases of error responses reported, one user (P10) stopped using the chatbot. The remaining three persisted in their tasks, arguably because knowledge workers with higher critical thinking are more likely to use chatbots even when the output is undesirable (Bianchini et al., 2025, p. 13). In two other cases, users expressed that they had never encountered this type of chatbot and therefore required time to understand what it could help them with. This indicates that Chatbot Novelty is a variable; training users on a chatbot’s underlying context appears necessary to increase its acceptance and usability. The error responses and user challenges due to low awareness of what this type of chatbot advisor can do are significant examples of facilitating conditions that can derail user behaviour even when users intend to use a chatbot (Menon & Shilpa, 2023).

While user training/awareness could have been achieved through a pre-interview walkthrough or by asking respondents to perform tasks on the chatbot using ‘think-aloud’ interviews (Peng et al., 2022; Cheah et al., 2024), this study opted to allow users to test Obruche in their own time. The rationale was that self-motivated use by journalists would be closest to real-world user scenarios for the chatbot.

The chatbot’s female persona was cited as a factor that can simultaneously enhance and reduce use. Although four users specifically gravitated towards the persona and used feminised pronouns when speaking about Obruche, two others felt the female persona perpetuated sociocultural stereotypes. A female respondent said such chatbots would not be accepted by men in her society. These divergent views re-echo the “ethical quandary” of AI designers: women are being objectified in AI, yet infusing female personas into AI chatbots makes them more human and preferable over male chatbots (Borau et al., 2021, p. 1052).

The results also suggest that though some users may show aversion to all chatbots (Venkatesh, 2022), the right chatbot could suspend pre-existing AI scepticism, as seen in one user (P6). Conversely, P10’s short-lived use of Obruche supports Zhang et al.’s (2024, p. 189) conclusion that immediate withdrawal may occur if a chatbot is perceived as giving generic information, or implicit error responses that offer no repair options for the user (Braggaar et al., 2023). However, there are indications from this study that out-of-vocabulary responses are better tolerated by users seeking information on specific story projects (i.e., respondents P3 and P7).

In terms of social influences, peer-to-peer and organisation-to-peer connections can increase usage and acceptance; respondents were already sharing the chatbot with their colleagues and managers or expressed intentions to do so. Results indicate that factors that would promote use and sharing of chatbots that are journalism collaboration advisors include: culturally and socially tailored responses (Peng et al., 2022, p. 11) adapted to user contexts, and delivery of the chatbot in formats users are already accustomed to (Terblanche et al., 2023), e.g., WhatsApp bots, with voice functions. As some journalists expressed willingness to pay for Obruche, perceived price value motivates acceptance of this type of chatbot, similar to studies on chatbots targeting health and business sectors (Liu et al., 2024; Usiña-Báscones et al., 2025).

Regarding this study’s underlying context, the chatbot’s interventions were linked to eight sources of power. Obruche mainly assisted journalists to confront/rebalance Control of Knowledge and Information and Control of Scarce Resources, as well as increased users’ Ability to Cope with Uncertainty. To a lesser degree, its interventions concerned challenges related to Control of Decision Processes; Control of Boundaries; Formal Authority; and Symbolism and Management of Meaning. Worth noting is that these results may be linked to the modest number of participants. Nevertheless, journalists’ perceptions after using this chatbot advisor indicate they felt empowered when Obruche provided unique information that enabled them to make decisions, perceive themselves as equal to their peers, minimise risks, and negotiate better working conditions. Interestingly, at least three journalists felt the burden was on them to provide “evidence” for fair pay and were satisfied they now could. The foregoing shows that chatbot advisors could contribute to enhancing journalists’ psychological, financial, digital, and physical safety.

However, this chatbot is noted as unlikely to succeed in resolving context-rooted or hierarchy-based sources of power, such as Structural Factors That Define Action and The Power One Already Has, respectively. Two respondents specified that systemic power imbalances in investigative journalism remain too formidable for a chatbot advising only freelancers to solve. Indeed, as Peng et al. (2022, p. 11) note, “upstream determinants” in any field can undermine the use of chatbots therein. However, one respondent [P5] disclosed that this same chatbot could be a solution to such ‘upstream’ forces, insisting that funders and coordinators need Obruche more, because they hold more power.

6. Limitations

This study has several limitations. It is not generalisable, due to its exploratory nature and the reality that more women (n.9) than men (n.3) responded to the open recruitment call, with most respondents based in African countries (n.8) and fewer insights from journalists in Europe (n.3) and Asia (n.1). Nevertheless, its findings are instructive because they reveal in-depth perspectives from a gender (i.e., women) and region (i.e., Africa) cited as vulnerable to power differentials in investigative journalism (Konow-Lund & Høiby, 2023; Munoriyarwa & Chiumbu, 2024). In addition, multi-country studies with over 140,000 participants show women are four times less likely to adopt AI than men, even with equal access (Otis et al., 2024). This research, therefore, makes a modest contribution towards revealing possible factors behind these trends, albeit from a journalism sector perspective. Future studies may cover non-binary and male journalists, as well as journalists, women journalists, and collaboration coordinators beyond Africa.

Secondly, although the use of the UTAUT as qualitative codes aligns with an emerging trend in chatbot studies (Peng et al., 2022; Menon & Shilpa, 2023), using pre-determined codes brings a risk of deductive bias, where the researcher’s assignment of responses to codes may be narrow-minded (Kennedy & Thornberg, 2018). However, to reduce deductive bias, member-checking of each coded transcript was done by the relevant interviewee, meaning the participants themselves confirmed the author’s interpretations of their user experience. Furthermore, the presence of rich descriptive data, the inclusion of broader related themes—e.g., Playful Interactions, Chatbot Persona, User Menu—and discussion of contradictory viewpoints, including the chatbot’s shortcomings (error messages), indicate attempts to reduce deductive bias during analysis (Alghalbie et al., 2026).

As this research was funded by an individual (the author), it is limited in scope because there was no third-party validation of the chatbot responses. However, given the exploratory nature of this work, it was necessary to prioritise expert-user validation rather than incur further costs for a low-risk chatbot. Lastly, by capturing a one-off snapshot of the chatbot using the UTAUT, this study is not longitudinal and does not account for how user perceptions may change, as every chatbot learns more over time and likely improves (Venkatesh, 2022, p. 644). Future longitudinal studies may assess changes in journalists’ user patterns of similar or new advisor chatbots.

7. Conclusions

This study applies the four tenets of the UTAUT to qualitatively depict the factors that enhance or hinder journalists’ use of a GenAI chatbot advisor for preventing and resolving power asymmetries in cross-border journalism work. Respondents indicated that chatbots could potentially assist cross-border journalists towards smoother working relations with their colleagues, provided they assist with discrete work tasks and workplace tension scenarios, whilst concurrently accommodating open-ended queries. In embracing user personalisation, chatbot design should prioritise training on responses that are cognisant of ethical boundaries yet anticipate hedonic usage. Furthermore, reducing conversation breakdowns and error responses increases usability. Novel chatbots may require targeted awareness training to increase usage, which falls under the general “AI literacy” training for journalists that Sonni et al. (2024, p. 1554) recommend. For single studies, think-aloud methods may be considered but planned with care, to prevent skewing data validity (Alhadreti & Mayhew, 2017; O’Brien & Wilson, 2023).

This study’s unique contribution is testing a chatbot designed to enhance equity in collaborative journalism, and assessing its feasibility using the UTAUT. It revealed discrete instances that buttress and broaden scholarly accounts of hedonic motivations, perceived risk, and sociocultural influences in chatbot use. The insights provided about the chatbot’s benefits and areas of concern may guide future remedial efforts or similar GenAI chatbot designs that enhance the safety of journalists in cross-border collaborations.