The Efficiency–Verification Paradox of Generative AI Use: Exploring the Tension Between Productivity and Accuracy

Abstract

Artificial intelligence (AI), particularly generative artificial intelligence (GenAI), is influencing journalism and education by transforming professional practices, information production, and media literacy requirements. While generative AI improves efficiency in content creation and educational preparation, it also raises concerns regarding accuracy, credibility, verification, and ethical responsibility. This study examines how generative AI is experienced by journalists and teachers in primary schools in Croatia within the contemporary information ecosystem. Empirical data were collected through semi-structured interviews with ten participants, including five professional journalists and five teachers, and analysed using reflexive thematic analysis. The analysis identified a shared efficiency–verification paradox across both professional groups. Although generative AI reduces the time required for routine cognitive tasks, it simultaneously increases the need for verification, critical evaluation, and ethical judgement. Participants also emphasised the growing importance of AI literacy, critical thinking, and ethical competencies for navigating AI-generated content in professional and educational contexts. The findings suggest that generative AI does not necessarily replace professional expertise but instead redistributes professional responsibilities towards activities requiring human judgement and accountability. This study contributes to a broader understanding of how AI influences interconnected information professions and highlights the need to strengthen competencies required for responsible engagement with AI-generated information.

1. Introduction

Artificial intelligence (AI), particularly generative artificial intelligence (GenAI), has rapidly become embedded in professional and educational environments, transforming how information is produced, processed, evaluated, and communicated. Unlike earlier AI systems primarily designed for prediction, classification, recommendation, or automated decision-making, GenAI can produce text, images, audio, video, and other synthetic content based on patterns learned from large datasets (Banh & Strobel, 2023; Feuerriegel et al., 2024). Its growing use creates opportunities for efficiency while raising concerns about accuracy, credibility, transparency, professional responsibility, and human oversight. Recent scholarship suggests that GenAI is more likely to transform and redistribute professional work than eliminate human expertise, particularly in activities involving contextual interpretation, verification, ethical reasoning, and accountability (Pavlik, 2023; Guzman & Lewis, 2024; Dodds et al., 2025). Journalism provides an important context for examining these changes because journalistic work depends on the production, verification, contextualisation, and dissemination of information. AI is increasingly used for transcription, translation, information retrieval, summarisation, content adaptation, and other newsroom tasks (Thurman et al., 2017; Shi & Sun, 2024; Bartleman et al., 2026). At the same time, research identifies concerns about factual inaccuracies, fabricated information, transparency, editorial autonomy, accountability, and public trust (Diakopoulos & Koliska, 2017; Simon, 2022; Gutiérrez-Caneda et al., 2024; Porlezza, 2023). GenAI may therefore increase newsroom efficiency while simultaneously creating additional requirements for human evaluation and verification. Education represents a complementary context. Teachers increasingly use GenAI for lesson preparation, development and adaptation of teaching materials, assessment, and other routine activities, while also helping students navigate information environments shaped by AI-generated and algorithmically selected content. Research indicates that AI literacy, ethical awareness, confidence, and professional development influence teachers’ ability to integrate AI meaningfully into educational practice (Du et al., 2024). Recent research also identifies substantial variation in teachers’ preparedness and adoption of GenAI, with successful integration depending on individual, institutional, and pedagogical factors as well as technological capabilities (Song et al., 2025). Although journalism and education differ institutionally and professionally, they occupy complementary positions within the wider information environment. Journalists contribute to the production, verification, contextualisation, and dissemination of information, whereas teachers contribute to individuals’ ability to interpret and critically evaluate it. Examining these professions together therefore makes it possible to identify both profession-specific differences and shared patterns concerning AI use, professional judgement, verification, and responsibility. This connection is particularly relevant to media and AI literacy. Media literacy traditionally emphasises the ability to access, analyse, evaluate, and create media messages (Livingstone, 2004; Hobbs & Jensen, 2009; Bulger & Davison, 2018). GenAI expands these requirements because users must increasingly evaluate not only information itself but also the processes through which it is produced. AI literacy consequently involves understanding AI capabilities and limitations, recognising potential inaccuracies and biases, critically evaluating machine-generated outputs, and making informed and ethically responsible decisions about AI use (Vuorikari et al., 2022; Miao & Holmes, 2023). For journalists, these competencies relate closely to fact-checking, source evaluation, editorial judgement, and accountability; for teachers, they involve both evaluating AI-generated materials and developing students’ corresponding competencies. Despite the rapid growth of research on AI in journalism and education, these bodies of scholarship have largely developed separately. Journalism research has focused on automation, professional roles, editorial decision-making, autonomy, transparency, and accountability, while educational research has examined AI literacy, teacher preparedness, pedagogical adaptation, and responsible AI use. Comparatively less attention has been given to how professionals from different information-related occupations experience similar changes in responsibility and human oversight. In particular, qualitative comparative research examining journalists and school teachers in relation to the tension between AI-enabled efficiency and the need for human evaluation remains limited. This study addresses this gap through a comparative qualitative examination of journalists and informatics teachers in primary schools in Croatia. Rather than treating the two groups as professionally equivalent, it examines how distinct professional actors encounter GenAI at different stages of the information cycle and interpret its implications for professional practice. The study addresses four research questions:

RQ1. How do journalists integrate generative AI tools into their professional work, and what advantages and challenges do they identify?

RQ2. How do teachers evaluate the impact of AI on media literacy and their readiness to support students in critically evaluating AI-generated content?

RQ3. How do journalists and teachers, as distinct professional groups, interpret and negotiate the role of generative AI within their professional practices?

RQ4. Which ethical, professional, and educational challenges are identified as the most significant in the process of integrating AI into journalism and education?

The study is theoretically grounded in the Social Construction of Technology (SCOT) framework, which conceptualises technology as shaped through the interpretations, practices, and negotiations of relevant social groups rather than as producing predetermined social consequences (Pinch & Bijker, 1984; Bijker, 1995; Klein & Kleinman, 2002). Journalists and teachers are therefore understood as distinct relevant social groups whose professional norms, responsibilities, and institutional contexts shape how GenAI is interpreted and used. Human oversight serves as a complementary analytical concept for examining how professionals retain responsibility for evaluating AI-generated outputs and making context-sensitive decisions. On this basis, the study develops the concept of the efficiency–verification paradox: technological assistance can reduce the time required for routine cognitive tasks while simultaneously increasing the need for human evaluation, verification, and professional judgement. By bringing journalism and education into a common analytical framework, this study examines GenAI not as a technological innovation with predetermined consequences, but as a technology whose professional meaning and effects are negotiated within specific social and institutional contexts. It contributes to understanding how AI reshapes the distribution of professional tasks and responsibilities while maintaining the importance of human expertise in trustworthy information practices.

2. Literature Review

2.1. Generative AI and Journalism

The use of artificial intelligence in journalism predates generative AI and initially focused on automated news production, data processing, recommendation systems, and computational assistance, particularly in structured domains such as financial, sports, and election reporting (Thurman et al., 2017). GenAI has expanded these applications to information retrieval, transcription, translation, summarisation, language editing, headline generation, and content adaptation (Canavilhas, 2025; Shi & Sun, 2024; Pavlik, 2023). Recent research suggests that GenAI is transforming rather than simply replacing journalistic work. Studies identify applications across news gathering, production, and distribution, alongside concerns about accuracy, credibility, bias, transparency, autonomy, and ethics (Cools & Diakopoulos, 2024; Gutiérrez-Caneda et al., 2024). Broader reviews similarly identify automation, professional adaptation, misinformation, transparency, and trust as central issues in AI-supported journalism (Ioscote et al., 2024; Bartleman et al., 2026). Public acceptance also appears greater for supportive applications such as translation and grammar editing than for fully AI-generated news, particularly when human oversight remains visible (Simon et al., 2025). These findings indicate that AI integration in journalism is better understood as a redistribution of professional tasks than as straightforward substitution. Routine production activities may be accelerated, while source evaluation, editorial judgement, accountability, and verification remain central to professional practice (Diakopoulos & Koliska, 2017; Simon, 2022; Porlezza, 2023).

2.2. Generative AI, Teachers and AI Literacy

GenAI is increasingly used for lesson preparation, teaching-material development and adaptation, assessment, and idea generation. Although these applications may increase efficiency, AI-generated materials require professional evaluation because they may contain inaccuracies, biases, or contextually inappropriate information. AI literacy therefore extends beyond technical proficiency. It includes understanding AI capabilities and limitations, critically evaluating generated outputs, and recognising their ethical and societal implications. Research indicates that the field remains fragmented, with practical and ethical dimensions receiving less attention than technical knowledge (Sperling et al., 2024). AI literacy is also associated with teachers’ self-efficacy, perceptions of AI’s social benefits, and awareness of AI ethics (Du et al., 2024). Teacher preparedness and adoption remain uneven, reflecting individual, technological, institutional, and environmental factors (Cheah et al., 2025; Song et al., 2025). This suggests that meaningful integration depends not only on access to AI tools but also on professional judgement and continuing development. AI literacy also has a broader educational and societal function. Teachers must evaluate AI-generated materials while helping students recognise unreliable, biased, or manipulated information. This extends traditional media literacy, which emphasises accessing, analysing, evaluating, and creating media messages (Livingstone, 2004; Hobbs & Jensen, 2009; Bulger & Davison, 2018), towards competencies involving synthetic content, algorithmic processes, and the limitations of generative systems (Vuorikari et al., 2022; Miao & Holmes, 2023).

2.3. Comparative Research on Professional AI Use

Research on AI in journalism and education has largely developed along separate disciplinary lines. Journalism studies have focused on newsroom automation, professional autonomy, ethics, transparency, and accountability, while educational research has examined AI literacy, teacher preparedness, professional development, and pedagogical adaptation (Cools & Diakopoulos, 2024; Gutiérrez-Caneda et al., 2024; Du et al., 2024; Song et al., 2025). Less attention has been given to direct comparisons between professionals working at different stages of the information cycle. In particular, limited qualitative research examines whether journalists and teachers experience similar changes in professional responsibility, verification, and human oversight when GenAI is introduced into routine cognitive work. The comparison is not based on professional equivalence. Rather, journalism and education provide distinct contexts for examining a common issue: how professionals respond when AI can accelerate information-related tasks while requiring continued evaluation of its outputs.

2.4. Verification, Human Oversight and Professional Responsibility

Verification is fundamental to journalism and increasingly important in AI-mediated education. Journalists remain cautious about using GenAI for fact-checking because of concerns about accuracy and credibility (Cools & Diakopoulos, 2024; Gutiérrez-Caneda et al., 2024). Teachers similarly need to evaluate the accuracy and pedagogical suitability of AI-generated materials and help students critically assess machine-generated information (Vuorikari et al., 2022; Miao & Holmes, 2023). Human oversight consequently represents a shared principle across the two professions, although its immediate function differs. For journalists, it concerns editorial responsibility, source evaluation, credibility, and public trust; for teachers, it concerns pedagogical judgement, student learning, and responsible digital participation. In both cases, AI may support professional work without assuming responsibility for its consequences. This creates a potential tension between efficiency and verification. AI may reduce the time required to generate or process information, while professionals may subsequently spend additional time checking, correcting, contextualising, and evaluating the output. Efficiency may therefore involve a redistribution of effort from production towards verification and judgement rather than a simple reduction in total professional work.

2.5. Journalism and Education in the Contemporary Information Environment

Journalism and education have different institutional purposes but both contribute to trustworthy information environments. Journalism focuses on the production, verification, contextualisation, and dissemination of information, while education develops individuals’ capacity to interpret and critically evaluate it (Livingstone, 2004; Hobbs & Jensen, 2009; Bulger & Davison, 2018). GenAI strengthens the relationship between these functions by enabling information to be generated and transformed more rapidly while potentially complicating the assessment of its reliability, origin, and context. Journalists therefore operate primarily upstream in the information cycle, while teachers help individuals evaluate information encountered in increasingly AI-mediated environments. The comparison of these professions consequently addresses a shared societal challenge rather than institutional equivalence: maintaining trustworthy relationships between information, technology, and human judgement. This perspective also connects media literacy with AI literacy, particularly the ability to recognise synthetic content, understand AI limitations, and evaluate information critically (Vuorikari et al., 2022; Miao & Holmes, 2023).

2.6. Research Gap and Theoretical Positioning

Existing research provides substantial evidence concerning AI adoption, opportunities, risks, and professional transformation in journalism and education, but these research traditions remain largely separate. Moreover, much of the literature focuses on technological capabilities and adoption rather than on how professional groups construct the meaning and boundaries of AI use within particular institutional contexts. The Social Construction of Technology framework provides a useful basis for addressing this gap. SCOT conceptualises technology as shaped through the interpretations and negotiations of relevant social groups rather than as producing predetermined social consequences (Pinch & Bijker, 1984; Bijker, 1995). Its concept of interpretive flexibility is particularly relevant because the same technological system may acquire different meanings and uses across professional contexts. Later developments also emphasise the importance of organisational structures, resources, institutional conditions, and power relations (Klein & Kleinman, 2002), while closure and stabilisation help explain how particular interpretations become temporarily established within social groups (Humphreys, 2005). Applied to GenAI, this perspective allows journalists and teachers to be examined as distinct relevant social groups whose professional norms and institutional environments shape how AI is interpreted and used. Human oversight provides a complementary analytical concept for examining how professionals retain responsibility for evaluating AI-generated outputs and making context-sensitive and ethical decisions. Within this framework, the efficiency–verification paradox is understood not as an inherent technical property of GenAI but as a professionally situated outcome of its integration. AI may accelerate routine cognitive tasks while professional norms concerning accuracy, accountability, contextual judgement, and ethical responsibility continue to require human evaluation. The resulting redistribution of effort provides the central analytical link between technological capability and professional practice examined in this study.

3. Theoretical Framework

3.1. Social Construction of Technology as the Main Theoretical Perspective

Pinch and Bijker (1984) introduced the concept of interpretive flexibility, emphasising that the same technological artefact may acquire different meanings and functions among different social groups. Bijker (1995) further developed this perspective by emphasising processes of social negotiation and interaction. SCOT is particularly appropriate for GenAI because its significance depends not only on technical capabilities but also on how users incorporate those capabilities into existing practices. Journalists and teachers can therefore be understood as distinct but interconnected relevant social groups: journalists encounter GenAI primarily through information production, efficiency, accuracy, verification, and editorial responsibility, while teachers encounter it through lesson preparation, pedagogical practice, student learning, and AI literacy. Technological adoption also occurs within institutional conditions. Organisational arrangements, resources, professional norms, and power relations influence how technologies are implemented (Klein & Kleinman, 2002). The concepts of closure and stabilisation further suggest that understandings of appropriate technological use may become established within social groups while remaining open to later change (Humphreys, 2005). This is particularly relevant to GenAI because its capabilities and professional norms are still evolving.

3.2. Human Oversight as a Complementary Analytical Concept

Human oversight is used as a complementary concept to capture the continued role of professional judgement and responsibility in AI-assisted practice. It refers to professionals’ involvement in evaluating AI-generated outputs, making context-sensitive decisions, exercising ethical judgement, and retaining responsibility for consequences. This understanding is consistent with meaningful human control, which links technological behaviour to human reasons, intentions, and responsibility (Santoni de Sio & van den Hoven, 2018), as well as more recent approaches emphasising continuous human–AI interaction rather than a single final approval (Tsamados et al., 2025). In journalism, oversight involves fact-checking, source verification, contextual interpretation, editorial judgement, and accountability (Diakopoulos & Koliska, 2017; Porlezza, 2023). In education, it involves evaluating AI-generated materials, assessing pedagogical suitability, and guiding responsible AI use (Miao & Holmes, 2023). From a SCOT perspective, the boundaries between automated assistance and human decision-making are themselves socially negotiated. Human oversight therefore provides the conceptual link between the social construction of GenAI and the study’s central efficiency–verification paradox.

3.3. From Socially Constructed Uses of AI to the Efficiency–Verification Paradox

The efficiency–verification paradox follows from the interaction between GenAI capabilities and professional expectations. AI can accelerate routine cognitive tasks while simultaneously increasing the need for human evaluation, verification, contextualisation, and judgement. This is not treated as an inherent technical contradiction but as an outcome of how professional groups incorporate and constrain AI use. In journalism, AI may accelerate transcription, translation, summarisation, and headline generation, while journalists remain responsible for accuracy and credibility. In education, it may support lesson plans, worksheets, quizzes, and differentiated materials, while teachers must assess factual accuracy, pedagogical suitability, and curricular relevance. The paradox therefore represents a redistribution rather than a simple reduction of professional work. Routine activities may become faster or partially delegated, while human expertise becomes more concentrated on verification, contextual interpretation, ethical judgement, and accountability. The theoretical model guiding the study is consequently: GenAI creates technological possibilities, professional groups interpret and negotiate those possibilities, professional tasks are redistributed and efficiency gains may increase the relative importance of human oversight and verification. SCOT provides the principal theoretical lens for explaining this process, while human oversight captures the continuing role of professional judgement and accountability.

4. Materials and Methods

4.1. Research Design

This study employed a qualitative comparative research design to examine how generative AI is integrated into journalism and education, with particular attention to professional experiences, perceived benefits and limitations, and ethical considerations. A qualitative approach was selected because the study aimed to explore participants’ interpretations, experiences, and professional practices in depth rather than measure predefined attitudes or behaviours through standardised instruments. The study compared two professional groups within contemporary information environments: journalists involved in the production, verification, and dissemination of public information, and teachers who support students in critically interpreting and evaluating digital and AI-generated information. The study was theoretically informed by the Social Construction of Technology perspective (Pinch & Bijker, 1984; Bijker et al., 1987; Bijker, 1995). SCOT was used to examine how journalists and teachers assign meanings to GenAI and incorporate it into professional practice, while human oversight served as a complementary analytical concept for examining verification, professional judgement, contextual interpretation, and accountability. GenAI was therefore not treated as an autonomous force determining professional change, but as a technology whose role is negotiated within specific professional and institutional contexts. Data were collected through semi-structured interviews based on the research questions and theoretical framework. The protocol addressed GenAI use in everyday professional practice, perceived benefits and limitations, verification and evaluation of AI-generated outputs, professional and ethical responsibility, media and AI literacy, and anticipated future challenges. The comparative analysis focused particularly on the relationship between technological efficiency and the need for human evaluation, which informed the development of the study’s central concept, the efficiency–verification paradox.

4.2. Participants and Sampling

Participants were selected using purposive sampling based on their professional experience and direct engagement with the research topic. The final sample consisted of ten participants: five professional journalists and five informatics teachers employed in Croatian primary schools, as presented in Table 1.

Table 1. Characteristics of the study participants.

Group	n	Professional Background
Journalists	5	Online media (3), television (1), print media (1)
Teachers	5	Informatics teachers in Croatian primary schools

The journalists represented different media sectors to capture diverse perspectives on AI use in journalism, including online media (n = 3), television (n = 1), and print media (n = 1). The teachers were informatics teachers in Croatian primary schools, selected because their professional responsibilities include developing students’ digital competencies and engaging with technology, digital literacy, and responsible AI use. The teacher group included three women and two men, while the journalist group included two women and three men. The sample was intentionally balanced (n = 5 per group) to enable qualitative comparison while retaining attention to profession-specific experiences. Given the exploratory and interpretative purpose of the study, the sample was not intended to provide statistical generalisability but contextual and in-depth insight into how professionals interpret and negotiate GenAI. Participants were anonymised using the codes Journalists (J1–J5) and Teachers (T1–T5), and no personally identifying information was reported.

4.3. Data Collection

Data were collected between September and October 2025 through semi-structured interviews conducted in Croatian. All participants received the same core interview protocol addressing GenAI use, perceived advantages and limitations, verification practices, ethical considerations, professional responsibility, media and AI literacy, and expectations concerning AI’s future role in their professions. Seven participants participated in oral interviews, while three provided written responses to the same protocol: one journalist (J1) and two teachers (T1 and T2). Written responses were accepted because of professional and scheduling constraints and enabled the inclusion of relevant participants unable to participate synchronously. The two formats were treated as comparable in thematic scope but not identical in interactional depth: oral interviews allowed follow-up questions and clarification, whereas written responses were generally more concise. This difference is acknowledged as a methodological limitation. The seven oral interviews were audio-recorded with consent and manually transcribed, after which transcripts were checked against the recordings and corrected where necessary. Written responses were incorporated in their original form and analysed using the same thematic procedure. Before participation, all participants were informed about the study’s purpose, voluntary participation, confidentiality, anonymisation, and intended use of the data. Consent for participation and audio recording was obtained verbally before data collection.

4.4. Data Analysis

The data were analysed using reflexive thematic analysis (RTA), following Braun and Clarke (2006, 2021). It was selected because it supports an interpretative examination of participants’ experiences while recognising the active role of the researcher in constructing themes. Rather than treating themes as objectively emerging from data, RTA understands them as developed through the researcher’s engagement with the dataset, theoretical assumptions, and interpretative decisions (Braun & Clarke, 2021). The analysis followed Braun and Clarke’s six recursive phases: (1) familiarisation with the dataset, (2) generation of initial codes, (3) construction of candidate themes, (4) review and refinement of themes, (5) definition and naming of themes, and (6) production of the analytical report (Braun & Clarke, 2006, 2021). These phases were treated recursively rather than as a strictly linear sequence. The researcher first repeatedly reviewed the complete dataset, focusing on GenAI use, efficiency, verification, professional responsibility, human judgement, ethical concerns, and AI and media literacy. Meaningful segments were then coded primarily inductively, while the research questions and SCOT perspective provided a broader interpretative orientation. Related codes were organised into candidate themes and repeatedly compared across journalists’ and teachers’ accounts to identify convergent and divergent patterns. Four broader analytical patterns were identified: (1) GenAI as a professional or pedagogical support; (2) efficiency gains associated with AI use; (3) the growing importance of verification and human judgement; and (4) AI literacy, critical thinking, and responsible engagement with AI-mediated information environments. These patterns supported the development of the efficiency–verification paradox, which emerged from participants’ recurring accounts that GenAI reduced the time required for routine tasks while simultaneously creating additional demands for checking, evaluation, contextualisation, and professional judgement. Human oversight was used as a complementary analytical concept, referring to professionals’ continued involvement in evaluating AI-generated outputs, making contextual and ethical judgements, and retaining responsibility for outcomes. This concept connected participants’ concrete verification practices with the broader SCOT perspective by highlighting how professional actors incorporate, constrain, evaluate, and negotiate GenAI within existing practices. The final themes were compared across the two professional groups to distinguish shared patterns from profession-specific manifestations of AI integration. Examples of the development from initial codes to candidate and final themes are presented in Table 2.

Table 2. Development of themes through reflexive thematic analysis.

Initial Codes	Candidate Theme	Final Theme
Translation, transcription, grammar correction, headline writing, idea generation	AI supports routine work	GenAI as a professional assistant and pedagogical support tool
Faster workflow, time saving, productivity, reduced preparation time	Efficiency	Efficiency gains and the growing importance of verification
Fact-checking, source verification, human review, editorial and pedagogical responsibility	Verification and human oversight	Human judgement, and professional responsibility
AI errors, hallucinations, unreliable information, source comparison	Critical evaluation	AI literacy, critical thinking, and responsible digital participation

4.5. Researcher Reflexivity

Researcher reflexivity was integrated throughout the analytical process. Consistent with reflexive thematic analysis, the researcher recognised that qualitative findings are developed through the interpretative relationship between researcher, participants, data, and theoretical framework (Braun & Clarke, 2021). Attention was therefore paid to how prior disciplinary knowledge and assumptions about journalism, education, and AI might influence coding and theme development. Reflexivity was supported through repeated reading of the dataset, comparison of participants’ accounts, iterative examination of coding decisions, and revision of candidate themes. Contrasting and negative cases were retained rather than treated as deviations from dominant patterns, and premature generalisations from individual statements were avoided given the small sample and differing levels of AI adoption. The researcher also considered the implications of the two data-collection modes. The brevity of written responses was not interpreted as weaker engagement, while the absence of interviewer follow-up was recognised as potentially limiting the depth of some accounts. Finally, SCOT was used as an interpretative framework rather than a rigid coding template, allowing themes to remain grounded in participants’ accounts while examining how professional groups construct the meaning and role of GenAI within their professional and institutional contexts. The resulting themes are therefore understood as analytically constructed interpretations rather than objective categories independent of the research process.

5. Results

The reflexive thematic analysis identified four overarching themes concerning how participating journalists and informatics teachers in primary schools understood and experienced generative artificial intelligence (GenAI) in their professional contexts. Analysis was conducted across the full dataset and subsequently compared between the two groups. Among journalists, two closely related themes emerged: Generative AI as a Professional Assistant and Efficiency Gains and the Growing Importance of Verification. Among teachers, the corresponding themes were Generative AI as a Pedagogical Support Tool and AI Literacy, Critical Thinking, and Responsible Digital Citizenship. Across both groups, participants generally positioned GenAI as a supportive rather than autonomous technology, although its meaning differed by professional context. Journalists primarily associated it with accelerating routine newsroom tasks while emphasising verification and editorial responsibility. Teachers associated it mainly with lesson preparation and educational activities, while stressing students’ critical competencies and responsible engagement with AI-mediated information. Given the small sample of ten participants, these findings are presented as patterns within participants’ accounts rather than representative claims. Differences between participants, including the explicitly critical position of one journalist who did not use AI tools, are retained as analytically relevant.

5.1. Generative AI as a Professional Assistant

The interviews indicate that journalists primarily perceive generative AI as a supportive technology that enhances existing journalistic practices rather than replacing professional work. Participants described using AI tools for various routine cognitive tasks, including transcription, translation, language editing, summarisation, processing press releases, background research, headline generation, and preparation of interview questions. These applications were associated with increased efficiency, while responsibility for editorial decisions and verification remained with journalists. Themes identified among journalists are presented in Table 3.

Table 3. Themes identified among journalists.

Theme	Interpretation	Illustrative Quotation
Generative AI as a Professional Assistant	GenAI supported routine newsroom activities, while reporting, source evaluation, editorial judgement, and verification remained human responsibilities.	“I also use it to formulate questions for interviewees… AI helps me come up with questions.” (J5)
Efficiency gains and the growing importance of verification	GenAI accelerated selected tasks but also required additional checking and evaluation of AI-assisted outputs.	“Unfortunately, I double-check everything, which sometimes makes my work take longer.” (J3)

One participant described AI as a practical component of everyday newsroom work:

“I most often use AI tools to process press releases, police reports, rewrite news from other domestic and international news portals, and generate headlines.”

(J1)

Similarly, participants emphasised the usefulness of generative AI for multilingual tasks and interview preparation:

“It’s very useful for translating texts, especially from languages we’re not familiar with.”

(J3)

“I also use it to formulate questions for interviewees… AI helps me come up with questions.”

(J5)

Participants generally distinguished these uses from activities requiring professional judgement. Field reporting, interviewing sources, assessing credibility, contextual interpretation, and editorial decisions remained human responsibilities. As one journalist explained:

“I don’t think it will become the central part of the job because it can’t go into the field and report on an event or call a relevant source.”

(J2)

Although the level of AI adoption differed among participants, the dominant view was that generative AI functions as a professional assistant rather than an autonomous producer of journalistic content. One participant expressed a more critical position and avoided using AI-generated text because of concerns regarding style and authenticity:

“I don’t use AI tools… The sentences they produce are usually sterile and full of clichés.”

(J1)

This account demonstrates that professional engagement with GenAI could result not only in adoption but also in deliberate non-use. Overall, journalists evaluated GenAI primarily according to task-specific usefulness, distinguishing activities suitable for AI assistance from those requiring human professional judgement.

5.2. Efficiency Gains and the Growing Importance of Verification

The second major theme identified among journalists concerned the relationship between increased efficiency through generative AI and the continued need for human verification. Although participants recognised that AI tools can accelerate various newsroom tasks, they consistently emphasised that AI-generated outputs require careful evaluation before publication.

Journalists described generative AI as useful for supporting initial stages of content production, particularly drafting, translation, transcription, summarisation, and information processing. However, participants stressed that these benefits do not eliminate the need for professional review. Instead, AI-assisted work requires additional attention to fact-checking, source verification, contextual accuracy, and editorial judgement.

Several participants compared the use of generative AI to working with an inexperienced assistant whose outputs require supervision. As one journalist explained:

“My experience as a journalist tells me that everything has to be checked 150 times. People see using AI as if you had a junior intern, and you have to constantly check their work.”

(J3)

Concerns regarding factual inaccuracies, fabricated references, contextual misunderstandings, and AI hallucinations influenced journalists’ cautious approach towards AI-generated content. Participants emphasised that responsibility for published information remains with journalists and media organisations, regardless of whether AI tools were involved in the production process.

This responsibility was reflected in the following statement:

“I don’t have any ethical concerns because I use it only as a support tool. In the end, it’s still my name on the byline… and I verified the information afterward.”

(J2)

Participants also highlighted the importance of transparency and institutional guidelines for responsible AI use. They suggested that news organisations should establish clearer procedures regarding acceptable applications of AI, verification practices, and disclosure of significant AI involvement in content production.

At the same time, participants described situations in which AI increased rather than reduced the time required for journalistic work. Although routine tasks could be completed more quickly, the need to evaluate and verify AI-generated outputs introduced additional responsibilities:

“Unfortunately, I double-check everything, which sometimes makes my work take longer.”

(J3)

Overall, the findings indicate that journalists experience generative AI as a tool that improves workflow efficiency while simultaneously requiring continued human supervision. The relationship between efficiency gains and increased verification demands represents a central pattern in journalists’ experiences with AI-supported journalism.

5.3. Generative AI as a Pedagogical Support Tool

The teachers’ accounts positioned GenAI primarily as a support technology for lesson preparation and related educational activities. Table 4 presents the identified themes and representative quotations.

Table 4. Themes identified among teachers.

Theme	Interpretation	Illustrative Quotation
Generative AI as a Pedagogical Support Tool	GenAI supported lesson preparation and educational activities while teachers retained responsibility for pedagogical decisions and learning outcomes.	“Artificial intelligence has become my everyday digital assistant.” (T1)
AI Literacy, Critical Thinking, and Responsible Digital Citizenship	Teachers emphasised critical evaluation, verification, ethical awareness, and responsible participation in AI-mediated information environments.	“If information is not verified, it can cause fear, uncertainty, or even conflicts among people… Inaccurate information can threaten democracy and freedom of thought.” (T3)

Participants also described using AI to generate ideas, organise teaching content, and support communication-related tasks.

One teacher described AI as an integrated part of everyday professional practice: “Artificial intelligence has become my everyday digital assistant.” (T1) Participants consistently emphasised that the main advantage of generative AI lies in reducing the time required for routine preparation tasks. This allowed teachers to devote more attention to lesson planning, classroom interaction, and responding to students’ individual learning needs. However, teachers did not regard GenAI as an autonomous educational actor and retained responsibility for determining whether generated materials were appropriate for particular students, subjects, and learning objectives.

One participant explicitly connected efficiency with increased professional responsibility:

“The use of AI significantly reduces the time required… I realised that it is extremely important to educate myself about this topic, especially as a computer science teacher, which is why I feel a responsibility toward my students and my profession.”

(T2)

This account illustrates that effective AI use was understood as requiring continuous professional learning rather than familiarity with tools alone. Teachers needed to recognise limitations in generated outputs and determine when materials required checking or modification. Thus, GenAI was useful insofar as it complemented existing professional knowledge, while teachers retained responsibility for pedagogical decisions and the quality and appropriateness of educational materials.

5.4. AI Literacy, Critical Thinking, and Responsible Digital Citizenship

The second major theme concerned preparing students for information environments increasingly shaped by AI-generated and algorithmically mediated content. Teachers did not reduce AI literacy to technical proficiency; instead, they associated it with questioning information, evaluating and comparing sources, recognising limitations, and considering the social and ethical consequences of inaccurate or manipulated content. One participant emphasised the need for continuous professional development:

“I don’t think I will ever consider myself 100% confident, because tools and skills are advancing so rapidly… Through continuous professional development, I can always guide my students properly, at least in most cases.”

(T2)

Participants also expressed concern about students’ exposure to information through social media and other digital environments:

“It is not easy because students often believe everything they see on social media.”

(T3)

Critical thinking and verification were therefore presented as central to responsible digital participation. Teachers emphasised that students should question information rather than accept AI-generated or digitally circulated content simply because it appears plausible or authoritative.

The ethical dimension was particularly evident in concerns about inaccurate or manipulated information:

“If information is not verified, it can cause fear, uncertainty, or even conflicts among people… Inaccurate information can threaten democracy and freedom of thought.”

(T3)

Participants further noted that AI-related topics are most effectively addressed through discussion, practical examples, and activities that encourage students to analyse real-world cases. Examples such as deepfakes, personalised advertising, and AI-generated media were identified as useful teaching materials for developing students’ awareness of manipulation and information reliability. Overall, the findings indicate that teachers perceive AI literacy as an important extension of existing media literacy practices. For participants, preparing students for AI-related challenges involves developing critical thinking, ethical awareness, and the ability to evaluate information produced or influenced by artificial intelligence.

5.5. Cross-Professional Comparison of Journalists’ and Teachers’ Experiences

The comparative analysis identified both similarities and differences between the two professional groups. Despite their distinct institutional contexts, participants converged on the importance of human evaluation and responsibility in AI-assisted work. Table 5 summarises the main similarities and differences identified across the two professional groups.

Table 5. Cross-professional comparison of participants’ experiences with generative AI.

Dimension	Journalists	Teachers
Primary reported use of GenAI	Routine newsroom and preparatory tasks	Lesson preparation and educational activities
Main perceived benefit	Increased efficiency in selected routine tasks	Reduced preparation time and support for educational planning
Main concern	Accuracy, source verification, and editorial responsibility	Accuracy, critical evaluation, and responsible student use
Human responsibility	Editorial judgement, source evaluation, and verification	Pedagogical judgement, material evaluation, and educational guidance
Key competence	Fact-checking and source evaluation	AI literacy, media literacy, and critical thinking
Professional function	Supporting the credibility of public information	Supporting students’ critical interpretation of information
Shared principle	Human oversight of AI-assisted outputs	Human oversight of AI-assisted outputs

Journalists primarily understood GenAI in relation to information production and newsroom efficiency, whereas teachers focused on educational preparation and students’ competencies. Nevertheless, both groups distinguished between tasks that could be supported or accelerated by AI and responsibilities requiring professional judgement. For journalists, these included reporting, source evaluation, contextual interpretation, and editorial decisions; for teachers, they included evaluating educational materials, pedagogical decisions, and guiding students’ engagement with information. Verification represented a second major point of convergence. Journalists described it as fact-checking, source validation, and editorial review, while teachers emphasised critical evaluation, source comparison, recognition of unreliable content, and questioning AI-generated outputs. Although the practices differed, both groups positioned verification as a necessary human intervention between AI-generated content and its professional or educational use. The principal difference concerned the object of responsibility. For journalists, responsibility centred on the credibility of published information and accountability for the final product. For teachers, it extended from evaluating AI-generated materials to preparing students to navigate increasingly AI-mediated information environments. Overall, the findings indicate a shared pattern of human oversight combined with context-specific professional judgement. GenAI was not described as eliminating professional expertise; rather, it altered the distribution of professional attention. Routine production and preparation could be accelerated, while verification, evaluation, contextualisation, and responsibility remained central. GenAI was associated with greater efficiency in selected routine activities but did not remove the need for human intervention. Participants instead described continued attention to verification, evaluation, contextual understanding, ethical judgement, and professional responsibility. The findings should not, however, be interpreted as indicating uniform experiences. J1’s deliberate non-adoption demonstrates that professional responses ranged from practical use to critical avoidance, while teachers differed in their emphasis on professional learning, student guidance, and verification. The comparative analysis therefore identifies a shared analytical pattern rather than uniformity. Taken together, the results indicate that, within this exploratory sample, GenAI was experienced primarily as a technology that can redistribute rather than eliminate professional tasks. As routine activities became easier or faster, human attention remained necessary to determine whether AI-assisted outputs were accurate, appropriate, credible, and ethically acceptable.

6. Discussion

6.1. GenAI as a Socially Constructed Form of Professional Transformation

The findings support an understanding of GenAI as a technology that transforms professional practice rather than simply replacing human work. This interpretation is consistent with the Social Construction of Technology perspective, which emphasises that technological meanings and uses are shaped by relevant social groups, professional norms, and institutional contexts rather than determined by technical capabilities alone (Pinch & Bijker, 1984; Bijker, 1995). Although journalists and teachers engaged with the same broad technology, they interpreted its value through different professional responsibilities. Journalists emphasised accuracy, credibility, editorial autonomy, and accountability, whereas teachers focused more on pedagogical appropriateness, professional learning, and students’ ability to evaluate information. These findings support previous research suggesting that AI reorganises professional tasks rather than simply eliminating them (Pavlik, 2023; Guzman & Lewis, 2024; Dodds et al., 2025), while adding that this reorganisation is actively negotiated by professionals. The non-adoption reported by J1 further illustrates this interpretive flexibility. Adoption, selective use, and non-use may coexist within the same professional group depending on professional values and perceptions of technological quality. GenAI should therefore be understood as a socially situated form of professional transformation rather than as a technology with predetermined occupational consequences.

6.2. Human Oversight and Professional Responsibility

A central finding is the persistence of human responsibility despite increasing technological assistance. Participants in both professions treated AI-generated outputs as material requiring professional evaluation rather than as autonomous decisions. Although the specific forms of oversight differed, its underlying function was similar: professionals remained responsible for the consequences of using AI-assisted outputs. This extends existing discussions of transparency, accountability, and human agency in AI-supported journalism and education (Porlezza, 2023; Miao & Holmes, 2023; Guzman & Lewis, 2024) by showing that these principles operate as part of everyday professional practice. Human oversight should therefore not be understood merely as a final technical check. As AI-generated outputs become increasingly fluent and plausible, critical evaluation becomes an increasingly important component of professional competence. In this sense, GenAI does not eliminate professional judgement but changes where that judgement is exercised.

6.3. The Efficiency–Verification Paradox

The central conceptual contribution of the study is the efficiency–verification paradox: GenAI may accelerate routine cognitive work while simultaneously increasing the evaluative work required before an output can be used responsibly. This complicates the assumption that automation necessarily produces an equivalent reduction in professional workload. Previous research has identified efficiency benefits from AI alongside continuing concerns about editorial oversight, transparency, and professional autonomy (Simon, 2022; Pavlik, 2023; Shi & Sun, 2024; Canavilhas, 2025). The present findings extend this discussion by suggesting that efficiency should be considered across the complete professional workflow. A faster initial output is not necessarily a faster professionally usable output when verification, correction, contextualisation, or adaptation remain necessary. The same underlying relationship was observed in education, although the criteria for evaluation differed. The paradox therefore represents a redistribution of professional effort rather than a simple reduction. Routine cognitive tasks may become more efficient, while expertise becomes increasingly concentrated on assessment, contextualisation, critical evaluation, and responsibility. The concept should not be interpreted as evidence that GenAI fails to produce efficiency gains; rather, it indicates that such gains are conditional upon responsible verification.

6.4. Journalism and Education as Complementary Information Practices

The comparison further suggests that journalism and education represent complementary forms of information mediation. Journalism contributes to the production, verification, contextualisation, and dissemination of information, whereas education develops individuals’ capacity to interpret and critically evaluate it (Livingstone, 2004; Hobbs & Jensen, 2009; Bulger & Davison, 2018). In AI-mediated environments, both functions become increasingly important because the production of synthetic content can outpace the ability of users to assess its reliability and context. The findings therefore connect professional verification with AI literacy. Journalists primarily address reliability before information reaches wider publics, while teachers help students develop capacities for evaluating information encountered in digital environments. Although these are distinct practices, both contribute to maintaining trustworthy relationships between information, technology, and society (Vuorikari et al., 2022; Miao & Holmes, 2023).

6.5. Implications for Professional Practice and AI Governance

The findings indicate that responsible AI adoption requires more than access to technological tools. It also requires professional training, organisational guidelines, verification procedures, and clearly assigned responsibility. In journalism, this includes guidance on acceptable AI uses, verification, disclosure where appropriate, and accountability for published content (Diakopoulos & Koliska, 2017; Porlezza, 2023). In education, professional development should address evaluation of AI-generated content, recognition of inaccuracies and limitations, ethical reflection, and the development of students’ critical AI and media literacy (Miao & Holmes, 2023). Human oversight should also be institutionally supported. If verification is a condition of responsible AI use, professionals need sufficient time and resources to perform it. Otherwise, organisational pressure for efficiency may favour speed over accuracy and professional judgement. The SCOT perspective additionally suggests that governance should remain sensitive to professional context: journalists and teachers face different responsibilities, and relevant professional communities should therefore participate in defining acceptable uses and boundaries of GenAI.

6.6. Limitations and Future Research

The study has several limitations. The purposive sample of ten participants does not support statistical generalisation, and all teachers were informatics teachers in Croatian primary schools. Three participants provided written rather than oral responses, which allowed less opportunity for interactive probing. The study also relied on self-reported experiences rather than direct observation or analysis of AI-assisted outputs. Finally, the Croatian context and rapidly changing GenAI landscape limit the transferability and temporal stability of the findings. Future research should therefore employ larger and more diverse samples, cross-national and longitudinal designs, and methods that combine interviews with observation or analysis of AI-assisted outputs. The efficiency–verification paradox should be treated as an empirically derived concept requiring further examination rather than as a universal property of GenAI. Taken together, the findings suggest that the consequences of GenAI cannot be explained by technological capabilities alone. Professional groups interpret and negotiate its use according to their responsibilities and institutional contexts. SCOT helps explain this socially situated adoption, while human oversight captures the continuing role of professional judgement. The efficiency–verification paradox links these perspectives by showing how technological assistance can shift professional effort from routine production towards evaluation, verification, and accountability.

7. Conclusions

This study examined how GenAI is incorporated into journalism and education through a qualitative comparison of five journalists and five informatics teachers in Croatian primary schools. The findings show that GenAI is primarily experienced as a form of professional transformation rather than replacement. Its value lies mainly in supporting routine cognitive and preparatory tasks, while responsibility for evaluating and using its outputs remains with professionals. The main conceptual contribution is the efficiency–verification paradox. GenAI can make routine work faster without producing an equivalent reduction in overall professional effort because AI-assisted outputs still require verification, contextualisation, critical evaluation, and professional judgement. The significance of this finding is that it shifts attention from whether AI saves time at the point of generation to how professional work is redistributed across the complete process of production, evaluation, and responsible use. The findings also demonstrate the relevance of SCOT for understanding GenAI adoption. The technology acquired different practical meanings within journalism and education because professional groups interpreted it through different responsibilities and institutional contexts. Nevertheless, both groups retained human oversight as a central condition of responsible use. This suggests that the future significance of GenAI will depend not only on technological development but also on how professional communities establish boundaries, norms, and responsibilities around its use. The study therefore contributes to a broader understanding of professional expertise in AI-mediated information environments. As routine information production becomes increasingly accessible, the ability to verify, contextualise, critically evaluate, and take responsibility for information may become an increasingly important dimension of professional competence. The findings are contextually limited by the small Croatian sample, the focus on informatics teachers in primary schools, and the mixed interview formats. Further research should test the efficiency–verification paradox across larger, more diverse, and cross-national samples.