AI Hallucination Report
Last updated June 2026
A look at how and where AI models fabricate information, and what it means for trusting their answers.
Key takeaways
- Hallucinations cluster around citations, exact numbers, and recent events.
- Asking for sources is a common trigger for fabricated references.
- Search grounding lowers but does not remove hallucination risk.
Where hallucinations concentrate
Hallucinations spike around citations, exact numbers, recent events, and obscure entities. Requests for sources are an especially common trigger for fabricated references.
Search grounding reduces hallucination rates but shifts the risk toward source misreading rather than eliminating errors.
Don't just trust — verify
Run your question through ChatVerify and compare answers across leading AI systems.
How to spot a hallucination
Be suspicious of overly specific details delivered with high confidence, citations you can't independently find, and answers that change when you re-ask the question slightly differently.
When two strong models disagree on a fact, treat both answers as unverified until a credible source settles it.
Frequently asked questions
What is an AI hallucination?
It's when a model states something false as if it were fact — a fabricated statistic, a made-up citation, or a confident but incorrect claim. It happens because models predict plausible text, not verified truth.
How can I reduce hallucinations?
Compare answers across multiple models, ask for sources and check them, and be extra cautious with recent events and exact numbers. ChatVerify surfaces disagreement and sources to flag likely hallucinations.