How to Spot AI Hallucinations
Last updated June 2026
Hallucinations are designed to look correct, but they leave clues. Here are the warning signs and a quick process to catch them.
Key takeaways
- Be suspicious of overly specific details and exact figures.
- Always open cited sources — don't trust the citation itself.
- Re-ask the question; unstable answers are a red flag.
- Cross-check with a second model.
Warning signs
Suspiciously precise numbers or dates with no source; citations you can't find or that don't load; claims about very recent events; overly confident answers to ambiguous questions; and details about obscure people, products, or cases.
Don't just trust — verify
Run your question through ChatVerify and compare answers across leading AI systems.
A 60-second verification process
1) Identify the specific claims (numbers, names, sources). 2) Open any cited source and confirm it supports the claim. 3) Ask a second model the same question. 4) Look for consensus and read the disagreements. 5) For high-stakes topics, confirm with an authoritative source or professional.
ChatVerify runs steps 2–4 for you automatically.