Field survey · about
What's collected here, the analyses built on it, the bar each entry has to clear, and who built it.
Every entry across all three catalogs describes something that was actually built, run, or trialled — a named person, course, institution, or research team, with a real source link. Proposals, frameworks, and guidance documents don't qualify anywhere on this site, no matter how sensible they sound.
A named person, course, or institution describing exactly what they did with generative AI in a higher-ed teaching context — collected at universities and sector bodies worldwide. No frameworks; only specific examples.
A real research project where AI was used at some stage of the actual research process — literature search, data collection, coding, analysis, hypothesis generation, drafting. Research about AI itself (benchmarking a model, studying attitudes toward AI) doesn't qualify — the study's subject has to be something else.
A strategy an educator or institution actually ran to reduce student AI use — not a proposal, and not just an opinion that it probably helped. Entries are tagged by whether the source reports concrete measured evidence it worked, so evidenced results and documented attempts stay distinguishable rather than the attempts being excluded outright. After-the-fact AI-detection tools don't qualify as strategies, evidenced or not — the strategy has to change the task or the incentive. The only detection items listed are studies that test the detectors themselves; they're marked † and tagged "Detector study" rather than given an evidence tag.
Alongside the catalogs, the Analyses tab collects four pieces of work built on them or beside them:
Cross-tabulations of disciplines, tools, activity themes, years and regions across the teaching case studies, plus a small model that answers “given X, what’s likely?” Research applications trends does the same for the research catalog, without a predictor because there are too few entries per category.
What UK and Irish universities tell students about generative AI: what it can be used for, what must be acknowledged, what help is offered, and how that differs. Every statement was coded from the institution's own document, and every count traces to a quote. It is a draft and still in progress.
A literature briefing on how well teachers and other readers can tell AI-written text from student writing: accuracy, false accusations, confidence, and what helps. It is a working briefing, not a systematic review, and sources are labelled by how closely they were read.
A separate AI perceptions explorer presents findings from three linked studies of how students and staff perceive generative AI. It is not part of the case-study catalogs and isn't held to their inclusion bar.
Anyone can suggest a case study through the submission form — one form for all three catalogs, with a field to pick which type it is. Every submission is screened against the same criteria above before it goes live; nothing is auto-published.
This catalog and its content were built by Juliana Gerard, a lecturer in Linguistics at Ulster University. The application code was written with Claude Code (Anthropic), which was also used to source candidate case studies, screen them against the inclusion criteria, write entries, and draft the analyses; the human work is setting those criteria, directing the research, and reviewing what actually qualifies.
The summaries, quotes-in-context, and metadata written for these catalogs are available under CC BY 4.0 — reuse them anywhere, including commercially, as long as you credit this project and link back. That license covers only what was written here: this site claims no copyright or ownership over the case studies themselves, which remain the work of their original authors and institutions — each entry simply links out to the source. The application code is MIT-licensed and open source on GitHub.