In the rapidly evolving landscape of higher education, artificial intelligence (AI) tools like ChatGPT have become ubiquitous among UK university students. Recent surveys indicate that 95% of undergraduates use generative AI in some capacity, with 94% incorporating it into assessed work. Yet, a pivotal new study from Wonkhe reveals a critical insight: it's not institutional AI policies that dictate whether students harness these tools for genuine learning or mere output generation, but the very design of assessments themselves.
This research, drawn from focus groups with over 100 students across 20 UK universities, paints a concerning picture. Nearly half worry their grades reflect production rather than knowledge, and 38% admit submitting work they couldn't fully explain. As one student reflected in the study, assessments often feel like 'production lines' rather than opportunities for intellectual growth. With cheating cases surging—nearly 7,000 proven instances in 2023-24 alone, up threefold from the prior year—UK higher education faces an urgent call to rethink how it evaluates student learning.
🔬 The Surge in AI Adoption Among UK Students
The Higher Education Policy Institute (HEPI) Student Generative AI Survey 2026 underscores the scale of integration. Conducted with 1,054 full-time undergraduates in late 2025, it shows AI use for assessments climbing to 94%, a stark rise from 53% just two years prior. Notably, 12% now include AI-generated text directly in submissions, up from 8% in 2025.
Students cite time-saving (49% report improved experiences) and enhanced understanding as benefits, yet 59% fear skill erosion, particularly critical thinking. Disciplines vary: computing students embrace AI for coding support, while arts and humanities lag in institutional encouragement. This near-universal adoption, per HEPI, demands assessments that evolve beyond traditional essays and exams, which AI excels at mimicking.
- 95% overall AI use
- 94% for assessed work
- 65% note assessment changes due to AI
- Gender gap: men 20+ points more likely to use heavily
Cheating Cases on the Rise: The Detection Dilemma
Freedom of Information requests to 131 UK universities reveal 6,968 proven AI misuse cases in 2023-24—5.1 per 1,000 students, projected to hit 7.5 this year. Traditional plagiarism dipped, but AI's subtlety evades tools: a University of Reading study found 94% of AI-generated exam papers went undetected. Experts like Dr. Peter Scarfe warn this is 'the tip of the iceberg,' as proving misuse risks false accusations.
Over 27% of universities lacked separate AI misconduct tracking last year, complicating data. Students report anxiety over detectors, with 75% in a 2026 wellbeing report citing 'AI detector anxiety.' This underscores policy pitfalls: bans or declarations punish honest users while savvy cheaters edit outputs seamlessly.
Unpacking the Wonkhe Research: Assessment Design as the Decider
The Wonkhe report 'Trained to Stop Learning,' published March 2026, shifts focus from prohibition to design. Through thematic analysis of student voices, it identifies six AI modes—from idea generation to full production acceleration. Crucially, visible accountability (e.g., viva-style defenses) prompts learning-oriented use, like self-testing, while absent oversight enables substitution.
Only 21% feel courses reward thinking; students crave assessments emphasizing application, feedback, and personal insight. Poor design—unclear briefs, late feedback, time poverty—drives autopilot AI reliance, hitting disabled students hardest (e.g., ADHD support gaps) and widening gender divides. Download the full Wonkhe report for detailed findings.
Photo by Marija Zaric on Unsplash
UK Institutional Policies: From Bans to Balanced Guidance
Most UK universities now have AI policies, yet Wonkhe deems them 'incoherent.' Declarations deter ethical users without curbing heavy misuse. The Quality Assurance Agency (QAA) advocates redesign over detection, promoting 'authentic assessment' via real-world tasks. Government guidance echoes this, urging malpractice prevention without outright bans. Visit QAA's AI resources hub for toolkits.
HEPI's survey shows 80% find policies clear, but only 36% feel encouraged to use AI productively. Russell Group institutions improved slightly, yet arts students report least support.
Real-World Case Studies: Pioneering Redesigns in UK Universities
Durham University shares generative AI case studies in teaching, including AI for diagnostic skills in medicine. Leeds experiments with AI in student education, fostering ethical integration. UCL proposes radical shifts: AI-augmented portfolios with vivas. King's College London offers a 'menu' of resilient approaches, from verbal exams to project defenses.
Scottish universities adapted English for Academic Purposes (EAP) assessments post-GenAI, blending orals and portfolios effectively. These cases show tailored designs boosting integrity and learning, per Advance HE's PGTA support research.
Challenges: Equity, Wellbeing, and Skill Erosion Fears
AI exacerbates divides: non-users (mostly women) fear disadvantage (74%), while disabled students use it for unmet needs. Time-poor learners (working/caring) prioritize survival over depth. 59% worry about reduced critical thinking; late feedback turns assessments summative.
- Digital access gaps persist
- Gender usage disparity >20%
- 15% use AI for wellbeing support
- Institutional gaps (libraries, briefs) fuel reliance
Solutions: Prioritizing Thinking Over Output
Wonkhe urges 'accountability moments' like developmental vivas, peer explanations, and application tasks. HEPI recommends AI inductions, curriculum revamps, staff training. QAA's Hallmarks playbook aids authentic redesign: scenarios, reflections, collaborations.
Experts advocate discipline-specific strategies—computing integrates coding aids, humanities emphasizes critique. Peer learning builds belonging, curbing substitution.
Photo by Annie Spratt on Unsplash
Future Outlook: Employability and Beyond
68% see AI skills as vital; understanding-focused assessments boost career confidence, especially orals. As AI evolves, UK universities must align with graduate needs: ethical judgement, communication—irreplaceable human skills. Proactive redesign positions institutions as innovators, per HEPI.
Read HEPI's full 2026 survey for more data.
UK higher education stands at a crossroads. By centering assessment design on understanding and accountability, universities can transform AI from threat to ally, ensuring students emerge as thinkers, not just producers. This shift, long overdue, promises deeper learning and equitable outcomes.