Jisc's AI marking and feedback pilot says formative feedback is the right place to start
Updated May 22, 2026
Universities experimenting with AI in assessment feedback now have a clearer steer on where the risk is lowest and the learning value is highest. On 20 May 2026, Jisc published Formative First: insights from the AI in Marking and Feedback Pilot, arguing that formative assessment is the best place to start when institutions test AI-supported marking and feedback. For Student Experience teams, PVCs, and quality professionals, that matters because the question is no longer only whether AI can speed up feedback. It is whether universities can introduce it in ways that protect trust, keep staff judgement visible, and generate evidence strong enough to act on.
What has changed in Jisc's AI marking and feedback pilot
The immediate development is not a new regulation but a new set of sector-facing early findings. Jisc's year-long pilot was launched in 2025 and runs from September 2025 to August 2026, bringing together colleges and universities across two strands: purpose-built marking tools, Graide, KEATH, and TeacherMatic, and general-purpose AI assistants using tools such as ChatGPT, Gemini, and Copilot. On 20 May 2026, Jisc published a series of reflections based on community sessions, feedback forms, and one-to-one interactions with participants. The clearest message is practical: start with formative feedback, not high-stakes grading.
"formative assessment is the best place to start"
Jisc's reasoning is specific. In the formative context, faster, more structured feedback can still be used while learning is in progress, and some institutions have used those efficiency gains to create more formative opportunities before summative assessment. Jisc also says students were often more comfortable with AI when it was framed as a learning companion rather than an automated assessor. The same series is more cautious about summative use, saying concerns about accuracy, transparency, and fairness remain sharper where final marks are at stake.
The supporting blogs add two further points that matter for implementation. First, Jisc says institutions in the pilot took different approaches to student consent, with some using opt-in or opt-out models because questions of trust, fairness, and comfort mattered alongside the legal basis for processing student work. Second, Jisc says human oversight is harder than it looks. The pilot found that institutions need to design review workflows deliberately, otherwise staff risk either being steered by AI output or checking it too lightly for the oversight to mean much in practice.
What this means for institutions
The first implication is about scope. If a university wants to test AI in marking and feedback, Jisc's early reflections suggest it should begin where the benefit to students is clearest and the downside is easier to contain: developmental, low-stakes feedback that students can use before final submission. That does not remove the need for moderation, but it gives teams a safer way to test rubrics, workflows, and staff confidence before discussing wider use.
The second implication is that AI pilots need a student engagement plan, not just a technical plan. Jisc says some institutions used student advisory panels, town-hall style sessions, or meetings with students' unions to surface concerns early and refine their approach. That matters because institutions are not only testing a tool. They are testing whether students see the process as fair, transparent, and respectful. It also echoes a wider finding on this site that students use Generative AI for feedback, but trust teachers more, especially when the stakes rise. A pilot is more likely to hold up if students can see where human judgement still sits and why the institution chose this use case.
The third implication is about workflow design. Jisc's human-in-the-loop findings warn against treating oversight as a vague reassurance. The more useful lesson is operational: define who reads the work first, what the AI is allowed to do, how staff challenge or rewrite output, and what counts as adequate review. Jisc highlights dual marking as one workable pattern, where staff note their own judgement before seeing AI feedback. Universities considering similar pilots should pair that with a student comment analysis governance checklist so that consent, review, communication, and evaluation decisions are recorded clearly from the start.
How student feedback analysis connects
These early findings are mainly about how universities produce feedback, but the next question is how they interpret what students then say about it. If AI-supported feedback is rolled out in a module or school, student comments are likely to mix together several issues: speed, clarity, tone, usefulness, fairness, personalisation, and confidence that a real academic judgement sits behind the final comments. Without a structured way to separate those themes, institutions can end up with a general sense of approval or discomfort but little clarity about what to change.
That is where open-text analysis becomes more useful. A consistent method such as our NSS open-text analysis methodology helps teams compare what students say before and after an AI pilot, while Student Voice Analytics can help institutions track those themes across modules, surveys, and representative channels with a reproducible method. The point is practical rather than promotional: if AI is going to reshape assessment feedback, universities need an evidence trail that can distinguish faster feedback from better feedback.
FAQ
Q: What should institutions do now if they are considering an AI marking and feedback pilot?
A: Start with a bounded formative use case, define the human review step in detail, and involve students early. Before the pilot begins, decide how you will collect evidence on usefulness, fairness, and trust, not only staff time saved.
Q: What is the timeline and scope of Jisc's latest pilot update?
A: Jisc published the early reflections on 20 May 2026. The wider pilot was launched in 2025 and is running from September 2025 to August 2026 across colleges and universities, covering both purpose-built marking tools and general-purpose AI assistants. This is UK sector guidance from a live pilot, not a new regulatory rule.
Q: What is the broader implication for student voice?
A: The broader implication is that universities will need to listen more precisely once AI enters assessment feedback. Student voice work will have to distinguish whether students are reacting to turnaround time, feedback quality, fairness, or the visibility of human judgement, rather than treating AI feedback as one issue.
References
[Jisc / National Centre for AI in Tertiary Education]: "Formative First: insights from the AI in Marking and Feedback Pilot" Published: 2026-05-20
[Jisc / National Centre for AI in Tertiary Education]: "The Value of Student Consent: insights from the AI in Marking and Feedback Pilot" Published: 2026-05-20
[Jisc / National Centre for AI in Tertiary Education]: "The Practicalities of Keeping the Human in the Loop: insights from the AI in Marking and Feedback Pilot" Published: 2026-05-20
[Jisc / National Centre for AI in Tertiary Education]: "AI in Assessment Pilot" Published: 2025-05-14
Request a walkthrough
Book a free Student Voice Analytics demo
See all-comment coverage, sector benchmarks, and reporting designed for OfS quality and NSS requirements.
- All-comment coverage with HE-tuned taxonomy and sentiment.
- Versioned outputs with TEF-ready reporting.
- Benchmarks and BI-ready exports for boards and Senate.
UK-hosted · No public LLM APIs · Same-day turnaround