Which is the most reliable early signal of a potential hallucination in an AI response?
Answer: The response contains a specific named reference — person, case, statistic — you cannot independently verify
Specific named references (people, cases, statistics) are where hallucinations most commonly hide. Vague claims are harder to fabricate convincingly.