evidoria

← Back to browse

Good practice Imported

Cambridge Study Finds AI Not Yet Reliable for Grading University Essays

United Kingdom · Cambridge · See the United Kingdom profile · See the Cambridge profile

Evidence: Quasi-experimental Top 70% 40/100 · Ask Evidence Copilot about this practice

Cambridge psychologists and AI researchers tested Claude and ChatGPT grading 750 real essays from three UK universities. The systems matched human examiners' grade bands only 35–65% of the time, rewarding essay length and vocabulary over academic substance.

750+
Essay corpus size
35-65 %
AI-human grade-band match rate
3-8x longer
AI feedback length vs human
Cambridge Study Finds AI Not Yet Reliable for Grading University Essays

Details

Promoter
University of Cambridge
Period
2026
Keywords
higher education, psychology, assessment research

Context

A University of Cambridge-led team of psychologists and AI researchers evaluated three frontier LLMs, including recent Claude and ChatGPT versions (as of April 2026), on grading real undergraduate psychology essays.

Activities

The team assembled a corpus of 750+ psychology essays from three UK universities and compared AI-assigned grade bands against grades originally awarded by human examiners.

Results

AI systems matched human examiners' grade band only 35-65% of the time depending on model and essay set; all three systems were oversensitive to essay length and vocabulary range regardless of argument strength, and produced feedback 3-8 times longer than human assessors.

Conclusions

The Cambridge team's conclusion was explicitly cautionary: AI is 'not yet good enough' to grade open-ended university essays without significant human oversight.

Implementation
Indicative cost
Low (< €50k)
Time to results
Short (< 1 year)

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful