Integrity in assessments: How Assessio approaches faking and cheating
Download the evidence-based guide to faking, cheating, and AI use in online assessments.
Online assessment is now standard, but it raises a fair question: what stops candidates from cheating? This report answers that directly, drawing on four Assessio studies from 2025–2026 to show how often faking happens, how effective AI assistance really is, and the layered measures we use to prevent and detect both.
Key insights
The evidence is more reassuring than many assume. Here are three findings every hiring team should know.
10%
of candidates used a tool that wasn’t permitted
In a study of 500+ applicants, fewer than one in ten used unauthorised help — and only 2.1% would consider using AI in a real assessment.
44%
AI’s accuracy on figural questions
AI performs well on verbal and numerical items but struggles with images, scoring below the average human on important assessment criteria.
10,000+
human candidates benchmarked
Comparing AI against 10,000+ real candidates reveals the signatures of AI-assisted completion — which can be flagged for invesitgation.
Built on research,
not assumption
This paper is based on original Assessio studies from backed by peer-reviewed research.
Our approach to cheating and faking
No test is immune to faking. What matters is whether the measures against it are rigorous enough. Find out how ours operate.
Download the report now
Get your free copy and discover every prevention and detection measure in detail.
Frequently asked questions
Faking means misrepresenting yourself — it applies mainly to personality, motivation, and values assessments. Cheating means using unauthorised assistance, such as a calculator, search engine, or AI tool. It applies mainly to timed cognitive tests. The risks, detection methods, and countermeasures differ for each.
Research estimates range from around 8% to 40%, with higher figures usually coming from studies that reward high scores rather than normal recruitment conditions. In our study of 500+ practising candidates, fewer than 10% used a tool that wasn’t permitted, and only 2.1% would consider using AI in a real assessment.
Less effectively than most assume. AI performs well on verbal and numerical questions but scores just 44% on figural ones — below the average human. Strict time limits make switching applications impractical, and AI tools often recognise an assessment portal from a screenshot and refuse to help.
Through layered prevention: clear warnings about what counts as fraud; cognitive tests with multiple equivalent versions and reordered answers; complex images that are hard for AI to read; reasoning-based questions that aren’t searchable; strict time limits; and platform restrictions such as disabled copy-paste and limited restarts.
No — and we’re upfront about that. Faking is primarily a threat to validity. The right response isn’t to claim immunity, but to build protections rigorous enough to make successful faking genuinely difficult, detect the most likely forms, and be transparent about how they work.
