Students treat AI as a helper, not a replacement, in academic work

The rapid spread of genAI across universities has triggered intense debate over cheating, authorship, assessment, creativity and the future of academic work. Tools such as ChatGPT, Gemini, Claude, Perplexity and DALL-E can generate text, summarise sources, suggest ideas, simulate feedback and automate repetitive tasks.

A new study suggests the real disruption in higher education is not cheating alone, but the transfer of academic labour from writing to oversight. The study, Romanticising the algorithm: power dynamics and ideological implications in human–AI collaboration for higher education, published in AI & Society, is authored by Giulia Banfi and Marco Luca Pedroni of the University of Ferrara. Based on 40 reflexive diaries from 186 undergraduate journalism students, it shows that AI is being used as a helper and creative prompt, not as a trusted replacement for student authorship

AI enters higher education as both tool and power structure

The study argues that genAI should not be treated as a neutral teaching aid. Its use in universities is tied to wider questions about power, knowledge, performance and responsibility. The real concern is not only whether AI produces accurate or inaccurate content, but how the technology changes what students understand as learning, authorship and academic competence.

The research places AI use within a university environment already shaped by performance pressure, output-based assessment and technological management. In such a setting, AI can appear attractive because it saves time, produces polished drafts and helps students complete tasks faster. But the same qualities can encourage cognitive delegation, standardised writing and a shift away from deeper reflection.

The study challenges the common framing of AI as either a major opportunity or a major threat. Instead, it shows a more complex picture. Students used AI to improve efficiency and stimulate creativity, but they also monitored its weaknesses, debated its legitimacy and resisted full substitution of human judgment. Their engagement was marked by negotiation rather than surrender.

The evidence comes from an undergraduate Digital Media and Journalism course at the University of Ferrara during the 2024/2025 academic year. Students worked in 40 teams on AI-assisted editorial projects across a 12-week module. They developed outputs such as digital magazines, podcasts and social media strategies on socially relevant topics, including climate crisis, digital rights and disinformation.

The course was designed as both a production exercise and a critical inquiry into AI. Students attended preparatory workshops on AI in journalism, including automated writing, fact-checking, podcast production and ethical analysis. They then used AI tools during brainstorming, research, drafting, rewriting, visual design and layout generation. Each team maintained a reflexive diary documenting how AI was used, what worked, what failed and how the group judged the ethical and intellectual implications.

The study found that students did not use AI passively. They treated it as a resource that required correction, interpretation and human control. This is significant because higher education policy often focuses on detection, prohibition or compliance, while students’ actual practices show a more nuanced struggle over learning, responsibility and agency.

Four student profiles reveal negotiated AI use

The analysis identified four main profiles of AI use among the student groups: selective optimisers, critical experimenters, exploratory users and tech functional users. These profiles show how students integrated AI into their work in different ways while keeping varying degrees of control.

Selective optimisers were the largest group. These students used AI in a targeted, medium-intensity way. They turned to AI to speed up specific tasks, refine writing, generate titles, produce summaries or improve the formal quality of outputs. Their approach reflected a strong interest in efficiency, but they did not give AI central control over the project. They used it as a practical assistant while retaining responsibility for checking and correcting the final work.

Critical experimenters used AI more intensively but with active supervision. They tested the technology’s limits, explored its usefulness and reflected on its risks. They were more willing to experiment, yet they also insisted on human oversight. Their use of AI showed that high engagement does not necessarily mean uncritical acceptance. For these students, AI became a space for testing, comparison and reflection.

Exploratory users treated AI more as a creative resource. They used it to generate ideas, test stylistic alternatives and open new expressive directions. Their approach was less focused on efficiency and more focused on inspiration. However, this creative use also brought uncertainty, especially around originality, consistency and the risk of algorithmic standardisation.

The tech functional profile appeared only once and represented minimal use. This group used AI for narrow technical functions, such as basic text correction or headline drafting. It did not treat AI as a creative partner or intellectual collaborator. This limited use reflected caution and a stronger preference for human responsibility.

Across these profiles, the dominant stance was neither enthusiasm nor rejection. Most groups adopted a negotiating position: they adjusted AI use according to task, phase of work and perceived risk. They were more comfortable using AI for brainstorming, writing support and research assistance than for final revision and validation.

The distinction is important because students treated review and verification as human responsibilities. They often used AI to generate drafts or suggestions, but they did not trust it to determine final quality. Revision became the point at which students reasserted academic authority. They checked facts, revised style, corrected structure and evaluated whether the output matched the group’s intent.

The study also found that students tended to assign AI two main roles: subordinate assistant and cognitive partner. As a subordinate assistant, AI carried out commands under human direction. As a cognitive partner, it helped generate ideas and shape content without replacing human decision-making. Only a minority treated AI as a teaching agent or substitutive partner.

This pattern shows that students saw AI as more than a simple tool, but less than an independent author. They recognised its capacity to influence ideas, language and production, yet remained wary of giving it too much power. Their use of AI was therefore both practical and political: it reflected the pressures of modern higher education while also resisting the idea that automation should define learning.

Efficiency gains collide with questions of authorship and trust

Students repeatedly identified two major benefits of AI: speed and creativity. AI helped them move past blank-page moments, generate drafts, suggest angles, create summaries and provide starting points for further work. It reduced the time required for routine tasks and helped teams organise production under deadline pressure.

These benefits aligned with the performance-driven culture of higher education. Students valued AI because it helped them produce outputs faster and meet project demands. In that sense, AI reinforced a system that rewards speed, polish and measurable deliverables.

However, the diaries also showed persistent unease. Students reported that AI-generated text could be repetitive, shallow, banal, inaccurate or lacking in an authentic voice. They found that prompts had to be carefully designed and that weak prompts often produced generic responses. They also observed that AI could generate false or unreliable information, requiring human verification.

The result? AI saved time at the start of a task but created new work later. Students had to verify sources, rewrite text, restore originality, deepen analysis and ensure coherence. The technology did not remove labour. It redistributed labour from creation to oversight. This is one of the study’s most important insights. In AI-assisted education, students may become less like sole authors and more like editors, curators and supervisors of machine-generated material. Responsibility does not disappear. It moves. Students become accountable for deciding what to accept, reject, revise and authenticate.

This raises deeper questions about academic authorship. If AI generates a first draft, suggests an argument or shapes language, where does human authorship begin and end? The students did not resolve that question in a simple way. Instead, they treated authorship as a negotiated process. Human creativity was preserved through selection, revision, interpretation and final judgment.

The study also links these practices to broader concerns about neoliberal education. When universities prioritise efficiency, measurable outputs and performance, AI can easily become a tool for producing faster deliverables. But students’ discomfort with shallow, standardised or voice-less text shows resistance to that model. They valued productivity, but they also defended depth, originality and human expression.

This tension was especially visible in journalism education, where writing, interpretation, ethics and authorship are central. AI could support editorial work, but it also threatened to flatten voice and weaken critical engagement. Students used AI to brainstorm, draft and refine, but they remained aware that journalism depends on judgment, context and responsibility.

The findings suggest that the main educational challenge is not whether students should use AI. They already do. The challenge is how universities can build pedagogical frameworks that teach students to use AI critically, ethically and reflectively.

Such frameworks would move beyond detection-based responses to AI. Rather than treating AI mainly as a cheating threat, universities would need to teach verification, prompt awareness, source evaluation, authorship reflection and critical analysis of algorithmic power. Students must learn not only how to use AI but also how to question what it produces and what kind of academic subject it encourages them to become.

The study warns against the corporate and institutional rhetoric of responsible AI when it remains superficial. Policies that focus only on acceptable use or risk management may fail to address the deeper cultural shift underway. AI is not just entering classrooms as software. It is reshaping the meaning of effort, competence and originality.

The students’ diaries show that young users are neither naive believers nor automatic cheaters. They are negotiating with AI inside a system that already pushes them toward efficiency and output. Their uncertainty, caution and insistence on human verification point to a critical capacity that universities can strengthen.