Universities are confronting a rapid shift in how students study and submit work, as generative artificial intelligence becomes a routine part of undergraduate life. The change is prompting institutions to revisit assessment design while students and staff debate where legitimate academic support ends and misconduct begins.
A 2025 survey of 1,041 full-time undergraduates, conducted by Savanta for the Higher Education Policy Institute and Kortext, found that 92% of respondents had used artificial intelligence in some form, up from 66% the previous year. The share reporting use of generative AI for assessments rose from 53% to 88%.
The most common uses were relatively limited: explaining concepts, summarising articles and suggesting research ideas. But the survey also found that 18% of students had included AI-generated text directly in assessed work. That distinction matters to universities, which must separate tools that support learning from systems that effectively produce a student’s submission.
From prohibition to redesign
The findings suggest that blanket bans are becoming harder to enforce. Four in five respondents said their institution had a clear AI policy, while 76% believed their university could identify AI use in assessed work. Yet the same survey found that 59% agreed the way they were assessed had changed substantially in response to generative AI.
That combination points to a broader institutional challenge. Detection software may help identify suspicious submissions, but assessment policies must also explain which uses are permitted, how students should acknowledge AI assistance and what evidence of independent thinking is expected. Universities are increasingly considering supervised examinations, oral questioning, staged assignments and tasks tied to a student’s own research process.
Such measures can make learning more visible, but they bring costs. More live assessments require staff time, rooms and consistent marking. Oral examinations may disadvantage students with anxiety, disabilities or limited confidence speaking in an additional language unless suitable adjustments are provided. Replacing essays with invigilated tests may also narrow the skills that courses can assess.
Students see benefits—and risks
Students’ use of AI is not necessarily driven by an intention to evade learning. Generative tools can provide explanations at different levels, help users organise ideas and offer rapid feedback. For students balancing paid work, caring responsibilities or long commutes, that accessibility can be valuable.
However, convenience can become dependence. A polished answer may conceal inaccurate claims, fabricated references or reasoning that the student cannot reproduce independently. Overuse could also weaken writing, research and problem-solving skills—the capabilities employers expect graduates to carry into workplaces where AI tools are increasingly available.
“The main uses of GenAI are explaining concepts, summarising articles and suggesting research ideas,” the HEPI survey reports, while also recording that 18% of students had inserted AI-generated text directly into their work.
Students are also uncertain about AI-led marking. In the survey, 34% said they would put in more effort if exams were assessed by AI, compared with 29% who would put in less and 27% who said their effort would not change. The narrow spread reflects an unresolved question: whether automated assessment would make feedback faster or make academic judgment less transparent.
Preparing graduates for an AI-shaped workplace
The debate extends beyond academic integrity. Employers are likely to value graduates who can use AI productively while checking its output, protecting confidential information and taking responsibility for decisions. Universities therefore face pressure to teach AI literacy across disciplines rather than treating it as a specialist concern for computer-science courses.
That does not mean outsourcing education to software. Students still need subject knowledge to recognise weak answers and ask useful questions. They also need opportunities to develop communication, ethical judgment and collaboration—skills that cannot be demonstrated simply by producing fluent text.
The most durable response may be a combination of transparent rules and assessments that test process as well as product. Requiring drafts, research logs, source checks or brief discussions about submitted work can give students legitimate ways to use AI while keeping responsibility for the final work with them.
The transition will remain contested. Some academics fear that normalising AI will erode trust in degrees; some students argue that refusing widely available tools leaves them unprepared for employment. Universities will need to address both concerns by making expectations clear, investing in staff training and ensuring that changes do not shift hidden costs onto students.
Sources: Higher Education Policy Institute and Kortext, Student Generative AI Survey 2025; HEPI Policy Note 61; ERIC record for the survey.