Jisc’s pilot reflections favour formative uses of AI feedback

Updated Sep 07, 2026

Jisc’s 20 May 2026 pilot reflections recommend formative feedback as a starting point for experimenting with AI in assessment. They describe participants’ experiences, rather than a controlled comparison establishing the safest or most effective use. Formative First.

The pilot was scheduled for September 2025 to August 2026, across colleges and universities. It included purpose-built assessment tools and general-purpose assistants. The May posts were interim reflections from community sessions, forms and individual discussions; they should not be presented as its final evaluation. Pilot launch and design.

What participants reported

Some institutions described creating additional formative opportunities and greater student acceptance of AI as a learning companion. Jisc also reports concerns about marking consistency. Its recommendation does not make formative assessment risk-free or remove the need to check feedback.

A companion post distinguishes institutions’ legal bases for processing from choices to seek students’ agreement. It describes opt-in and opt-out arrangements and engagement through panels and student representatives. These are reported approaches, not a ruling that one consent model is legally suitable everywhere. Student consent reflections.

The oversight post describes risks of staff accepting outputs too readily or being influenced by them. One workflow has educators record their judgement before seeing AI feedback. Jisc presents this as a practical approach to meaningful review, without demonstrating that it will work best in every setting. Human oversight reflections.

Define what a local trial needs to show

Our practical suggestion is to specify the task, participants, review step and stopping criteria before collecting evidence. State what happens when an output is misleading and who can resolve a student’s concern.

Separate speed from usefulness. A shorter turnaround does not by itself show that students understand feedback or can apply it. Equally, favourable student comments cannot establish marking accuracy.

A comment review can explore perceived clarity, fairness and trust if the questions and participant group are suitable. Keep that evidence alongside the trial’s checks on outputs and educational outcomes, with each source’s limits visible.

The governance checklist can help scope the survey comment component. It is not a substitute for assessment validation or institutional decisions about the lawful handling of student work.

Review note, 7 September 2026: identified the posts as interim participant reflections, updated the elapsed pilot period, qualified consent and oversight examples, and removed unsupported claims of universally lowest risk or highest learning value.