Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 46 of the 47 references it cites were found in a registry or an index; 9 of the 27 sources its claims cite were checked for support.

Assessment-focused governance mode? Mapping Generative AI policies across 76 UK higher education institutions: Policy framework and recommendations

Open access
Tao, Yingnian, Ryan, Mark, Bialy, Filip, Zhang, Xingna · 2026 · Research Square (preprint)

Abstract

Higher education institutions are under pressure to develop Generative AI (GenAI) policies in teaching, learning, and assessment. How these policies are framed provides important insights into how the sector is responding to the opportunities and challenges posed by GenAI. This paper examines GenAI policy guidelines published by 76 UK universities, comprising 132 policy and guidance documents. Through qualitative in-depth thematic analysis, we identify five dominant policy themes: institutional positioning on GenAI; assessment design; marking and feedback; teaching and learning; and AI ethics. Our findings reveal an assessment-centred and integrity-first AI governance model. Institutional attention is directed primarily toward safeguarding assessment integrity, preventing academic misconduct and regulating student use of AI in assessed work. By contrast, considerably less attention is given to the pedagogical integration of GenAI into teaching and learning processes. We further find variations across institutions in key policy areas, including AI attribution, AI detection, and AI use in marking and feedback, reflecting broader epistemic, technological and governance uncertainties surrounding the technology in higher education. In response, we propose a policy framework and ten recommendations to support a more balanced approach to GenAI governance. As the first large-scale mapping of GenAI policy across UK higher education, this study makes an original contribution to scholarship on AI governance in education and provides evidence-based guidance for future policy review and development.

Overview

CoChat Summary

TL;DR

This preprint maps Generative AI policies across 76 UK higher education institutions, analysing 132 policy and guidance documents (199,738 words) through qualitative thematic analysis. It finds an assessment-centred, integrity-first governance model with much less attention to pedagogical integration, and proposes a five-dimension policy framework plus ten recommendations.

Key findings

  • Five dominant policy themes emerged: institutional positioning on GenAI; assessment design; marking and feedback; teaching and learning; and AI ethics.
  • Policy attention is heavily assessment-focused: 84% of universities cover institutional positioning, 84% cover GenAI in marking and feedback, and 64% cover GenAI in assessment design, while only 12% address GenAI in teaching and learning; 49% address AI ethics.
  • Institutional approaches diverge in key areas: 61% have a GenAI assessment use category, 47% address ethical GenAI use/academic misconduct/consequences, 30% mention GenAI detection tools, 16% address GenAI in marking and feedback, 9% address marking principles, and 17% address marking practices.
  • The authors propose a five-dimension GenAI policy framework and ten recommendations, including developing guidance for GenAI in teaching and learning processes, monitoring AI detection technology, considering AI citation/referencing, and co-creating and regularly reviewing policy.

Methods

  • The corpus was drawn from the QS World University Ranking 2025, filtered to UK universities, yielding 90 institutions; after applying selection criteria, the final sample was 76 universities with 132 documents totalling 199,738 words, averaging 1.7 documents per university.
  • Inclusion criteria focused on institution-level, high-level guidelines/principles or stance/position statements on GenAI in teaching, learning, and assessment; research-, marketing-, and administration-focused policies were excluded, as were news/blog formats.
  • Thematic analysis followed an iterative four-step process informed by Braun and Clarke (2006) and Chiu (2024b), conducted in ATLAS.ti version 25.0.1; two authors coded documents, and frequencies are reported as numbers and percentages of universities rather than raw quotation counts.

Limitations

  • The paper states that it focused only on GenAI policy and guidelines in teaching, learning, and assessment, excluding research, operational use, and other areas such as marketing and student recruitment.
  • The paper states that it only examined publicly available documents and webpages that do not require institutional login, so more detailed internal guidance may have been inaccessible.
  • The paper states that it only analysed macro-institutional level policy and guidelines, excluding meso- or micro-level guidance at faculty, department, programme, or module level.
  • Evident from the text: this is a preprint and has not been peer reviewed; the authors also acknowledge that some degree of subjectivity is unavoidable in determining coding boundaries, though they mitigate this by reporting university-level frequencies.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.21203/rs.3.rs-10029667/v1

Cites

46 of 47 references verified0 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

8 Valid · 9 Not valid · 30 Not checked · 103 Own results and interpretations

58 more claims

Create a free account to see all 150 claims and to verify the 2 references still open.

See all claims — free account
  • Probably not a substantial claim. 56
  • It may state two separate claims as one. 50
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

Show all 40 references

Research works printed but not linked: 4

  1. Bangor University (2025) Generative Artificial Intelligence Software and Academic Integrity Guidance Document. https://doi.org/https://www.bangor.ac.uk/regulations/documents/artificial-intelligence-software-assessment-en.docx
  2. Jisc (2023) AI in tertiary education: A summary of the current state of play – Third edition
  3. QAA (2023a) Generative Artificial Intelligence in Education—QAA’s response to DfE call for evidence. https://doi.org/https://www.qaa.ac.uk/docs/qaa/news/generative-artificial-intelligence-call-for-evidence.pdf
  4. The University of East Anglia. (n.d.). Generative AI Policy for Teaching and Learning. The University of East, Anglia https://doi.org/https://assets.uea.ac.uk/f/185167/x/5558a0f812/generative-ai-policy-for-teaching-and-learning.pdf

Metadata

Authors
Tao, Y., Ryan, M., Bialy, F., & Zhang, X.
Affiliations
  • Lancaster University (GB)
  • University of Manchester
  • University of Liverpool (GB)
Venue
Research Square (preprint)
Published
2026-06-17
OpenAlex
W7164993731
Cited by
0
References
40
Type
posted content
Language
en
Open access
Locations
  • Research Square (acceptedVersion, cc-by)
  • Lancaster EPrints (Lancaster University) (submittedVersion)
Topics
  • Social Sciences › Education › Artificial Intelligence in Education
  • Social Sciences › Education › Academic integrity and plagiarism
  • Social Sciences › Education › Digital Education and Society
Keywords
generative AI, AI governance, higher education institutions, assessment integrity, academic misconduct, AI in teaching and learning, AI ethics, AI attribution, AI detection, British universities, policy framework

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Assessment-focused governance mode? Mapping Generative AI policies across 76 UK higher education institutions: Policy framework and recommendations. https://app.cochat.ai/papers/10.21203/rs.3.rs-10029667/v1
The paper
Tao, Y., Ryan, M., Bialy, F., & Zhang, X. (2026). Assessment-focused governance mode? Mapping Generative AI policies across 76 UK higher education institutions: Policy framework and recommendations. Research Square (preprint). https://doi.org/10.21203/rs.3.rs-10029667/v1

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.