Home / Insights / Higher education news

AI marking: scores up to 33 points apart

On 9 November 2026 Times Higher Education Campus published a review of AI marking assistants by Aditi Jhaveri and Eunice Tang of the Hong Kong University of Science and Technology.

The review says the most persistent finding in the research is also the most inconvenient: AI tools consistently award higher marks than human evaluators. In one study the same undergraduate assignment, submitted repeatedly to ChatGPT, scored up to 24 percentage points apart; Claude produced ranges of up to 33 points. Another study found a measurable bias against non-native English speakers, which would matter in internationally diverse cohorts. The authors also report gains in organisation, structure and writing quality for students who received AI feedback through rubric-aligned tools.

Why it matters: marks carry consequences. A team that wants AI help with feedback should first compare it with moderated human marks, check results across student groups, and keep a person responsible for the grade. That sits within the assessment and moderation training MTIS Education offers.

Source: Times Higher Education Campus, 9 November 2026, https://www.timeshighereducation.com/campus/ai-marking-assistants-future-assessment-or-not-fit-purpose

MTIS Education training and advisory