AI in Assessment and Feedback

The reasons are familiar. Marking is time-consuming, cognitively demanding and relentless. In many cases, educators are spending up to 10 hours per week on assessment-related tasks, with peaks significantly higher during key submission periods. When workloads are this heavy, something has to give. Too often, it is the quality, depth or timeliness of feedback that suffers.

At the same time, AI is already being used in assessment, often informally and without clear institutional guidance. The real risk is no longer whether AI is used. It is whether it is used in a consistent, governed and pedagogically sound way.

A sector at a turning point

Across schools, colleges and universities, AI adoption is accelerating, but guidance has not kept pace. In many cases, AI is being used under the radar, without shared expectations or clear boundaries. This creates inconsistency across teams and raises important questions around safeguarding, fairness and professional standards.

The challenge for institutions is not whether to engage with AI. It is whether adoption will be unstructured and reactive, or deliberate and strategically led.

A national pilot grounded in real practice

The Jisc AI Assessment Pilot brought together a group of colleges, universities and post-16 providers from across the UK to explore how AI could be used responsibly in assessment and feedback.

Crucially, this was not a controlled experiment. Educators embedded AI into live teaching and assessment workflows, testing its impact on feedback quality, consistency across cohorts, turnaround time, and staff workload.

What emerged was something the sector has been lacking: a shared, evidence-informed understanding of AI in assessment, grounded in real practice.

The feedback gap: what needs to change

Before looking at solutions, the pilot reinforced a long-standing issue. There is a persistent gap between what feedback should achieve and what it often delivers.

This is not simply an operational challenge. It is a pedagogical one. When feedback is delayed by weeks, students may no longer clearly remember their thinking, making it significantly harder to apply that feedback to future work. Timely feedback is not a convenience. It is essential to learning.

What the pilot revealed so far…

Tom Moule, Senior AI Specialist, Jisc

One of the most important insights from the pilot was that workload reduction alone is not the main driver.

This reframes the conversation. The value of AI is not simply that it saves time. It is that it enables more consistent marking, greater depth of feedback, and faster turnaround for students.

For leaders concerned with quality assurance and standardisation, this is significant.

Why formative assessment led the way

The pilot also showed where AI is currently most effective. Formative assessment naturally emerged as the primary use case. In this context, AI enabled faster feedback cycles, more detailed commentary, and increased opportunities for student improvement.

This is where the impact on learning becomes most visible.

Improving workflow without adding complexity

Peter Kilcoyne, Managing Director, TeacherMatic

A major barrier to adoption is not capability, but workflow friction. If AI tools add complexity, they will not be used.

The ability to process multiple submissions and generate structured feedback at scale transformed what had previously been a time-intensive process. This reinforces a key principle: AI must integrate into existing practice, not sit alongside it.

Why general AI alone is not enough

A key question raised during the pilot was simple: why not just use tools like ChatGPT or Copilot?

In practice, general AI assumes a level of expertise that is not always present. It relies on educators to construct effective prompts, apply pedagogical frameworks, and interpret and refine outputs. At scale, this leads to inconsistent outputs, uneven practice across teams, and challenges in governance and oversight.

What the pilot demonstrated is that success depends not just on the AI itself, but on how it is structured, guided and embedded into teaching practice.

Human judgement remains central

Across all institutions, there was clear alignment on one principle: AI should support professional judgement, not replace it.

Every successful implementation observed in the pilot shared one non-negotiable principle: the educator remains the final decision-maker.

Real impact in the classroom

Alessio Corso, Associate Professor

Working with large cohorts of first-year engineering students, Alessio Corso used AI-supported feedback within a structured, rubric-driven approach. He developed detailed criteria aligned to specific components of the assignment and used AI to support consistent, criteria-based feedback across the cohort.

AI-supported feedback provided structured comments aligned to assessment criteria, clear identification of strengths and areas for improvement, and greater consistency across a large cohort. This level of structured feedback would be difficult to replicate consistently at scale using traditional marking approaches alone.

Crucially, Alessio maintained full oversight, reviewing outputs to ensure alignment with academic standards and his own professional judgement.

Lisa Sayce, PCET Lecturer

From a Further Education perspective, Lisa Sayce’s experience highlighted both the practical efficiency gains and the pedagogical impact.

Working with trainee teachers submitting regular lesson reflections, Lisa used AI to support high-volume feedback tasks that would traditionally require significant time. The result was not just faster marking, but better feedback.

Feedback became more structured, more detailed, and more aligned to professional expectations. AI supported her in identifying patterns, linking theory to practice, and providing more targeted developmental feedback.

This is particularly significant. Feedback is often criticised for being impersonal, yet here it was perceived as more personalised, clearer and more actionable.

Stephen Clarke, Head of Digitally Enhanced Learning

If AI requires advanced technical expertise, adoption will remain limited. If it supports and scaffolds practice, it becomes accessible across teams.

Adoption: the real challenge

The pilot also highlighted a key reality. Most educators are not early adopters. They sit within the persuadable middle: open to innovation, but cautious, and requiring reassurance and evidence.

The challenge is not convincing innovators. It is supporting the majority of staff who need clarity, confidence and practical guidance.

Why the SAFE Framework was developed

Esam Baboukhan, Director, TeacherMatic

One of the clearest challenges identified through the pilot was not capability, but uncertainty. AI-supported feedback is evolving rapidly, yet guidance has not kept pace.

Educators are asking: What is acceptable? Where are the boundaries? How do we ensure fairness and consistency? This matters because assessment sits at the heart of academic integrity, professional judgement, and institutional trust.

This is why the SAFE Framework was developed. Not as a theoretical model, but as a practical, institution-ready starting point.

Introducing the SAFE Framework

Its purpose is simple: to help institutions move from uncertainty to confident, structured adoption of AI in assessment and feedback.