Most university discussions about generative AI in assessment have focused on students. Policies explain what students may use AI for, what counts as inappropriate assistance and how AI use should be acknowledged. Yet staff are also beginning to use generative AI across the assessment lifecycle. They may use it to draft assessment briefs, suggest rubric wording, identify themes across a cohort or help prepare feedback. (QAA, 2023; Russell Group, 2023).

That creates a parallel policy question. What are staff actually allowed to do, where must human judgement remain decisive, and how should universities account for data protection, technical change and inclusion? Our analysis of publicly available staff-facing generative AI guidance from 16 UK universities suggests that this side of the policy landscape is still much less developed.

Staff use is often governed indirectly

All 16 documents in our study connected generative AI with assessment in some way, but only a minority gave concrete direction on decisions such as using AI in marking, moderation or feedback. Much of the guidance remained at the level of broad principles about integrity, fairness and responsibility.

This matters because general statements do not necessarily answer the questions staff face in practice. Can a lecturer paste student work into an approved AI system to help identify common feedback themes? Can AI suggest feedback wording that the marker then edits? Can it help interpret a rubric? Can an AI-generated suggestion influence a mark? What happens if the student’s work contains personal or sensitive information?

Without clear answers, staff may reach different conclusions in different modules or departments. One person may avoid useful forms of support because they are unsure whether they are permitted, while another may use AI in ways that create avoidable risks. Policy ambiguity can therefore produce inconsistency even when everyone is trying to act responsibly.

Four broad policy positions are emerging

We identified four recurring positions in the staff guidance.

One institution took a prohibitionist approach in which AI was effectively kept outside marking or feedback. Four used conditional enabling with oversight, allowing some AI support under defined conditions such as human review, prior approval or restrictions on uploading identifiable student work. Eight were design-led or principle-only, concentrating on responsible assessment design without giving detailed rules for staff AI use in marking and feedback. Three were silent or ambiguous on these decisions.

These categories are useful because they show that the policy question is not simply whether an institution is positive or negative about AI. What matters is the configuration of work between people and systems. A university can support innovation while still requiring a human marker to retain final authority, review every AI-assisted output and remain accountable for the result.

Putting the right human in the loop

We used the Right-Human-in-the-Loop model, or R-HiTL, to examine these patterns. The central idea is straightforward. Responsible AI use is not achieved merely by saying that a human is involved somewhere. The right person needs to retain authority at the right decision point. (Mosqueira-Rey et al., 2023; Alfredo et al., 2024).

In assessment design, AI may help draft wording or generate possibilities, but an academic should approve the final task and rubric. In marking, AI might be used for limited pre-processing or pattern identification, but the final academic judgement should not be outsourced. In feedback, AI may help generate a draft or alternative wording, while the member of staff remains responsible for accuracy, appropriateness and the message sent to the student. (Henderson et al., 2025; King’s College London, 2025).

This way of thinking moves policy away from broad phrases such as human oversight and towards specific workflows. It asks who makes the decision, what the AI is allowed to contribute, who checks the output and who is accountable if something goes wrong.

Technical guidance can age quickly

A further concern is that generative AI changes faster than most university policy cycles. Some guidance makes claims about what AI systems can or cannot do, refers to particular commercial products, or assumes that certain capabilities are fixed. Those statements can become outdated quickly as tools become multimodal, add accessibility functions or change their data-handling arrangements.

For staff, outdated guidance is not simply an inconvenience. It can shape assessment practice. A policy based on an old technical limitation may unnecessarily rule out a useful accessibility function. A document that names a particular tool without explaining the underlying principle may become confusing when that product changes or a different approved system is introduced.

Staff-facing guidance therefore needs a review mechanism. Institutions should state how frequently technical claims will be revisited and should write policy at the level of educational and governance principles wherever possible, rather than tying rules too closely to the current feature set of one platform. (Department for Education, 2025; Jisc, 2024).

EDI cannot be an afterthought

The policies in our sample tended to foreground integrity and institutional risk more strongly than inclusion. Explicit references to disabled, neurodivergent or otherwise marginalised students were comparatively rare.

Yet decisions about staff AI use can affect inclusion in practical ways. AI-supported feedback might help generate alternative formats, simpler explanations or multilingual versions where appropriate. It may also help staff manage heavy feedback workloads in ways that create more timely support. Conversely, poorly governed use can introduce bias, inconsistency or privacy risks. (Nicholson, 2025).

A blanket prohibition can appear safe while also closing off some potentially useful forms of support. An unrestricted approach creates different risks. The stronger option is a defined workflow that identifies where AI assistance is appropriate, where it is not, and what checks are required. Those decisions should be reviewed for their effects on different student groups rather than treated as purely technical choices.

What staff-facing policy needs to provide

Our findings suggest that staff guidance should become more operational.

Institutions need to map the main assessment decision points and state what AI may do at each one. A policy might permit AI to suggest feedback sentences provided they are checked and edited before release, while prohibiting AI from assigning grades. It might allow staff to identify common cohort themes using an approved environment while forbidding the upload of identifiable student work to external public tools.

The policy should also state what staff must disclose to students, what records should be kept and who remains accountable for the final decision. Clear data-handling rules are essential. Regular technical review and explicit EDI checks should be built into the policy process rather than left to individual lecturers.

This does not require a lengthy rulebook. In many cases, a short decision tree or workflow can communicate more effectively than several pages of principles.

Staff guidance is part of assessment infrastructure

Universities are beginning to recognise that staff use of generative AI cannot remain an informal matter. If AI is used in assessment design, marking or feedback, the rules around that use become part of the assessment infrastructure itself.

The strongest policies will not ask only whether AI is allowed. They will specify the division of labour between staff and systems, protect human judgement at consequential points, set clear data boundaries and recognise the inclusion effects of different choices.

Student-facing AI guidance has become increasingly visible. Staff-facing guidance now needs the same level of attention. Without it, institutions risk asking staff to exercise responsible judgement without giving them a clear structure in which to do so.

Sources and further reading

The sources below are those used in, or directly relevant to, the original article and the claims retained in this blog adaptation.

Alfredo, R., Echeverria, V., Jin, Y. et al. (2024). Human-centred learning analytics and AI in education. A systematic literature review. Computers and Education: Artificial Intelligence, 6, 100215. https://doi.org/10.1016/j.caeai.2024.100215

Department for Education (2025). Generative artificial intelligence in education. GOV.UK guidance. https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education

Henderson, M., Bearman, M., Chung, J. et al. (2025). Comparing generative AI and teacher feedback. Student perceptions of usefulness and trustworthiness. Assessment & Evaluation in Higher Education. https://doi.org/10.1080/02602938.2025.2502582

Jisc National Centre for AI in Tertiary Education (2024). Generative AI. A primer. https://nationalcentreforai.jiscinvolve.org/wp/2024/08/14/generative-ai-primer/

King’s College London (2025). Guidance for AI-assisted marking and feedback. https://www.kcl.ac.uk/about/strategy/learning-and-teaching/ai-guidance/ai-assisted-marking-and-feedback

Mosqueira-Rey, E., Hernández-Pereira, E., Alonso-Ríos, D. et al. (2023). Human-in-the-loop machine learning. A state of the art. Artificial Intelligence Review, 56, 3005–3054. https://doi.org/10.1007/s10462-022-10246-w

Nicholson, R. (2025). Navigating the intersection of AI, accessibility and education in 2025. Jisc National Centre for AI in Tertiary Education. https://nationalcentreforai.jiscinvolve.org/wp/2025/02/03/navigating-the-intersection-of-ai-accessibility-and-education-in-2025/

Quality Assurance Agency for Higher Education (2023). Reconsidering assessment for the ChatGPT era. QAA advice on developing sustainable assessment strategies. https://www.qaa.ac.uk/docs/qaa/members/reconsidering-assessment-for-the-chat-gpt-era.pdf

Russell Group (2023). Russell Group principles on the use of generative AI tools in education. https://www.russellgroup.ac.uk/media/6137/rg_ai_principles-final.pdf