What is Hallucination?
When an AI states something false with confidence.
A hallucination is a confident but incorrect AI output, a wrong price, a feature you do not have, a claim that you were “discontinued.” For brands this is a reputation risk, because the false statement is delivered as fact to a buyer.
Hallucinations often trace back to a weak or outdated source the model leaned on. CiteTrack’s AI answer monitor runs your own model over the latest answer to flag likely-false claims about your brand and surface the source behind them, so you can go correct it.
Why — and when — does it matter?
A false confident statement can change purchase decisions or create legal and reputation risk. Material claims require evidence capture and human escalation.
How it works
Capture the exact answer, prompt, engine and date.
Separate verifiable claims from opinion.
Check owned facts and the cited source.
Classify severity and recurrence.
Correct source data and monitor the same prompt.
Worked example
An engine says a software product was discontinued because it retrieved an old migration notice. The team updates the notice, strengthens current product facts and monitors until the claim disappears.
How to track and report it
Reporting rule: State the engine, prompt set, market, time window and denominator. A number without scope cannot be reproduced or compared.
Practical playbook
- Capture the exact answer, prompt, engine and date.
- Separate verifiable claims from opinion.
- Check owned facts and the cited source.
- Classify severity and recurrence.
- Correct source data and monitor the same prompt.
Common mistakes
- Calling a subjective opinion a hallucination
- Correcting only the chatbot output
- Failing to preserve evidence
Questions people ask
What is Hallucination?
When an AI states something false with confidence. A false confident statement can change purchase decisions or create legal and reputation risk. Material claims require evidence capture and human escalation.
How does Hallucination work?
Capture the exact answer, prompt, engine and date. Separate verifiable claims from opinion. Check owned facts and the cited source. Classify severity and recurrence. Correct source data and monitor the same prompt.
How do you measure Hallucination?
Material false-claim count and recurrence. Affected prompts, engines and time to correction. Error rate = false verifiable claims ÷ verifiable claims reviewed × 100
What is a common mistake with Hallucination?
Calling a subjective opinion a hallucination. The scope and methodology should be stated before interpreting the result.