ChatGPT and Academic Cheating: What the Data Actually Shows

TL;DR

Guardian FOI 2025: 7,000 UK students caught with AI = 5.1/1,000. Stanford 2023: 9–16% self-admit. Scarfe et al. 2024: 94% of AI submissions go undetected. AMI D2 scores 39 countries — China (68) and Iran (100) highest, Australia (44) and Canada (44) lower.

The confirmed case data

United Kingdom (Guardian FOI, June 2025)

The most detailed national dataset comes from The Guardian's Freedom of Information investigation. Requests to UK universities revealed:

• 6,971 formally confirmed AI misconduct cases in 2023–24
• 5.1 cases per 1,000 students — the highest confirmed rate for any country with comprehensive data
• Cases concentrated in business, law, and social science departments
• Significant variation between institutions

United States (Stanford Survey, 2023)

A Stanford University survey across multiple institutions found:

• 9–16% of students self-reported submitting AI-generated content as their own work
• Rates varied significantly by subject area
• Business and social science students reported higher rates than STEM

The detection problem (Scarfe et al., 2024)

Researchers at the University of Reading conducted controlled experiments submitting AI-generated work through standard university assessment processes. They found:

• Approximately 94% of AI submissions went undetected
• AI detection tools showed significant false positive and false negative rates
• Turnitin's AI detection was more effective than manual review but still missed substantial proportions

If 5.1 per 1,000 UK students are caught and 94% go undetected, the implied true rate is approximately 85 per 1,000 — or about 8.5% of students. This is consistent with the Stanford self-report findings.

The global picture

The AMI D2 (AI submissions) dimension scores all 39 countries based on available data. Sources include FOI disclosures, published survey data, and country-adjusted literature estimates.

Countries with highest D2 scores (estimated AI misconduct)

Countries with lowest D2 scores

Why high D2 in Iran and Colombia?

The very high D2 Trends signals for Iran and Colombia reflect high search volume for AI submission keywords relative to their student populations. In Iran, internet restrictions mean students use VPNs to access ChatGPT — creating a pattern of intensive, purposeful use that likely correlates with academic use. In Colombia and other Latin American countries, rapid ChatGPT adoption combined with limited institutional AI policies creates conditions for high use.

University policy responses

The AMI D2 data is partly a measure of institutional response as well as student behaviour. Countries with clearer policies, better detection, and stronger disclosure frameworks will appear to have higher confirmed misconduct rates even if actual rates are similar to countries that look the other way.

This is addressed in the AMI methodology by the enforcement-detection correction, which adjusts P-Scores for countries with strong detection infrastructure.

What this means going forward

D2 is the newest and fastest-evolving dimension in the AMI. The evidence base is less than three years old and estimates will be revised with each annual update. The combination of improving AI detection tools, evolving institutional policies, and changing student behaviour makes this the most uncertain dimension in the dataset.

View the full methodology | Download the dataset

Related

Explore the full dataset