Evaluation & Quality
Faithfulness
How well an AI answer stays supported by the context or evidence it was given.
Example
Evaluators check whether every answer claim is supported by retrieved material.
Why it matters
It helps prevent an answer from adding claims its sources do not support.
You'll hear it
“The summary adds a conclusion the document never makes.”
Can you control it?
Not directly
No, not directly. Faithfulness 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
- Can an answer be faithful to a source that is wrong?
- Yes. Following the source accurately does not establish that the source itself is true.
- Does copying a sentence guarantee faithfulness?
- No. Removing surrounding qualifications can change the meaning.
- Can an answer use only supported claims yet omit important details?
- Yes. Avoiding invention does not guarantee a complete or balanced account.