How Common Is Cheating in College? Global Data

TL;DR

Estimates of college cheating prevalence vary widely:

• Newton (2018) meta-analysis: 15.7% of post-secondary students paid for academic work
• McCabe surveys (2002–2015): 60–80% of students admitted some form of academic dishonesty
• Stanford and US surveys: 9–16% used AI for assignments
• Guardian FOI 2025: 5.1/1000 UK students formally caught using AI (with 94% miss rate per Scarfe 2024)

Substantial cross-country variance from Canada (lowest Prevalence) to China (highest).

The Newton 2018 meta-analysis

Newton, P. M. (2018). "How common is commercial contract cheating in higher education and is it increasing? A systematic review" [verify exact citation].

The systematic review pooled multiple self-report studies on contract cheating. Headline finding:

• 15.7% of post-secondary students had paid someone to complete academic work

The 15.7% reflects studies of varying quality, populations, and time periods. Some studies found rates below 5%; others above 30%. The 15.7% is the pooled aggregate.

Applied to the global higher education population of ~225 million students, the 15.7% implies 30+ million students have used contract cheating services.

The McCabe surveys

Donald McCabe (Rutgers) conducted the most extensive self-report academic integrity surveys ever undertaken. Key features:

• 70,000+ students surveyed
• 70+ institutions across multiple countries
• 2002–2015 active period (pre-ChatGPT)
• Anonymous self-report methodology
• Multiple misconduct categories covered

Key findings

• 60–80% of college students admitted some form of academic dishonesty during their studies
• Plagiarism rates of 30–50% in many samples
• Unauthorised collaboration (collusion) rates of 40–60%
• Cheating on exams rates varying by setting

The McCabe data established that academic dishonesty is common — not the rare aberration that institutional rhetoric sometimes implies.

Limitations

• Self-report bias (some students overreport, others underreport)
• Recall reliability for "during your college career" questions
• Pre-ChatGPT era — AI submission category did not exist
• Limited international representation (most samples in Anglophone countries)

Post-ChatGPT data (2023 onwards)

The AI submission category emerged in late 2022. Initial data:

Stanford and US surveys

Stanford has run confidential undergraduate surveys post-ChatGPT [verify specific Stanford study]. Estimates:

• 9–16% of undergraduates use AI for assignments
• Higher rates in some disciplines (computer science, business)
• Lower rates in disciplines with established academic writing conventions (English, philosophy)

The Guardian FOI (June 2025)

UK-wide FOI investigation: nearly 7,000 UK students formally caught using AI in 2023–24 (5.1 per 1,000 students).

Scarfe 2024 detection finding

University of Reading study: 94% of AI-generated submissions went undetected. Applied to the Guardian data:

• 5.1/1000 detected × (1/0.06) detection factor = ~85/1000 estimated true rate
• Implies ~8.5% true incidence in UK

The detection correction generally

Across countries, detection capability varies. Countries with strong detection report higher rates; countries with weak detection report lower rates. The actual underlying rates may be more uniform across countries than detected-case data suggests.

What the AMI data shows

The AMI Prevalence Score combines six dimensions. Across the 39-country v1.5 dataset:

Top 10 by Prevalence

Bottom 10 by Prevalence

The 20x range between the lowest (Canada P=4.90) and highest (China P=99.98) demonstrates that "how common is cheating" has very different answers in different countries.

By misconduct type

Plagiarism

• McCabe samples: typical 30–50% rates of self-reported plagiarism
• AMI D4 dimension: ranges from 32 (Norway) to 73 (China)
• Most common form across most populations

Contract cheating

• Newton 2018: 15.7% paid for academic work
• AMI D1 dimension: ranges from 33 (Anglophone ban countries) to 100 (multiple)
• Growing through the 2010s; possibly declining post-ChatGPT

AI submissions

• Stanford/US: 9–16% confidential survey
• UK FOI confirmed: 5.1/1000 (with ~94% detection miss rate)
• Estimated true: 8%+
• New category emerging since late 2022

Collusion

• McCabe samples: 40–60% rates of self-reported unauthorised collaboration
• AMI D5 dimension: ranges from 50 (Singapore) to 75 (Nigeria)
• Possibly the most common form depending on definition

Exam impersonation

• Substantially rarer than other categories
• AMI D3 ranges from 8 (Singapore, Australia, Ireland, NZ) to 28 (Nigeria)
• Concentrated around high-stakes entrance examinations

Data fabrication (research)

• Retraction Watch: 5,390 misconduct-linked retractions of 69,911 total
• AMI D6 ranges from 12 (NZ) to 100 (China)
• Most concentrated form by impact (fabricated research enters scientific literature)

What this means

Cheating is more common than rhetoric suggests

Both the McCabe data (60-80% admitting some form) and the Newton data (15.7% paying for work) indicate that academic misconduct is far more common than institutional rhetoric typically acknowledges.

Specific forms vary widely

Different misconduct types have very different prevalence patterns. Aggregate statements ("most students cheat" or "few students cheat") obscure the variation.

Country differences are substantial

The AMI shows roughly 20x variance between the lowest and highest country Prevalence scores. National context matters enormously.

Detection asymmetry distorts comparisons

Countries with strong detection report more cases — not because they have more underlying misconduct but because they catch more of what occurs. Cross-country comparison must account for this.

Underlying rates are likely higher than detected rates

The Scarfe 2024 finding (94% AI miss rate) applies to AI submissions but illustrates a general principle: detection catches a fraction of actual misconduct. True rates are likely higher than confirmed-case data suggests.

Sources

• Newton, P. M. (2018). "How common is commercial contract cheating in higher education and is it increasing? A systematic review", Frontiers in Education
• McCabe, D. L. (multiple ICAI publications)
• Scarfe, P., et al. (2024). University of Reading detection study
• The Guardian (June 2025) FOI investigation
• AMI v1.5 dataset and methodology

Full methodology | Download dataset