Evaluation & Quality
Hallucination rate
The frequency with which a model produces unsupported or fabricated information.
Example
A test counts answers containing claims unsupported by the available source.
Why it matters
It helps teams track a specific reliability problem across versions or uses.
You'll hear it
“Does that figure count wrong answers or individual unsupported claims?”
Can you control it?
Not directly
No, not directly. Hallucination rate describes a learned property, observed result, or broader concept rather than a normal user setting. You may influence it indirectly by changing the tool, data, prompt, or workflow.
Common questions
- Is there one universal hallucination rate for a model?
- No. The rate changes with the task, examples, evidence and rules used to judge the answers.
- Who decides whether an answer is a hallucination?
- People or automated judges apply stated rules and supporting evidence; either kind of judge can make mistakes.
- Can a zero measured rate prove the model never hallucinates?
- No. It only means the chosen test found none in its sample under those conditions.