# Jisc pilot finds AI helps feedback more than final marking

> A year-long trial across UK colleges and universities found the strongest evidence for AI drafting feedback. Time savings depended on the task and the case for AI in final marks was weaker.

The Yoink Report (CodeYoink) | UK | AI in learning | News of 3 November 2026 | Published 6 November 2026

Jisc, the technology body for UK colleges and universities, published findings on 3 November 2026 from a year-long pilot of AI in marking and feedback.

[The report](https://nationalcentreforai.jiscinvolve.org/wp/2026/11/03/when-does-ai-improve-marking-and-feedback-evidence-from-the-ai-in-marking-and-feedback-pilot/) says 37 institutions joined in September 2025 and 28 stayed active to the main evaluation. They used TeacherMatic and Inspera Graide. A third product, KEATH.ai, withdrew. The report follows [interim posts in May](https://nationalcentreforai.jiscinvolve.org/wp/2026/05/20/insights-from-the-ai-in-marking-and-feedback-pilot/).

Jisc says AI was most useful for repeatable tasks where staff kept responsibility for academic judgement, such as drafting and structuring feedback. Workload savings were possible but conditional. They were clearest with digital submissions, explicit criteria and large batches of similar work. In long or interpretive tasks, staff could lose the saving to checking and rewriting.

Participants reported that AI could be too generous to weak work and did not always recognise exceptional work. Some tools struggled to read handwritten, visual or very long work in full.

For students, Jisc calls its evidence tentative because it gathered less direct feedback from them. Some accepted AI feedback if it was faster or clearer. Others raised concerns about fairness and trust. Views were more positive where lecturers explained why AI was being tested and how students could challenge the process.

In one effective approach for lecturers, the marker reads the work and makes notes before seeing any AI output, so the tool does not steer the judgement. Most activity in the pilot was formative, meaning work used for development and not for high-stakes results. Jisc says use in final marks needs a higher level of evidence and oversight.

Jisc calls the findings indicative, because this was a practice-based pilot and not a controlled comparison. The report gives no overall figure for hours saved. Case studies on custom AI assistants are due later in 2026.

## Sources

- Jisc National Centre for AI: [When does AI improve marking and feedback? Evidence from the AI in Marking and Feedback Pilot](https://nationalcentreforai.jiscinvolve.org/wp/2026/11/03/when-does-ai-improve-marking-and-feedback-evidence-from-the-ai-in-marking-and-feedback-pilot/)
- Jisc National Centre for AI: [Insights from the AI in Marking and Feedback Pilot](https://nationalcentreforai.jiscinvolve.org/wp/2026/05/20/insights-from-the-ai-in-marking-and-feedback-pilot/)

Picture: Anthony O'Neil, CC BY-SA 2.0, https://commons.wikimedia.org/wiki/File:Joint_Information_Systems_Committee._-_geograph.org.uk_-_7309953.jpg

Web page: https://codeyoink.com/news/jisc-pilot-ai-marking-feedback