Create intermediate Works with: Any AI

Summary Faithfulness Evaluation

Table output

Check summaries against their sources.

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

Evaluate whether a summary is faithful to its source. The source text: [PASTE_SOURCE]. The summary: [PASTE_SUMMARY].

Process:
1. Break the summary into atomic claims (one fact per line) and number them.
2. For each claim, find the supporting passage in the source (quote it with its location) or mark it Unsupported, Contradicted, or Overstated.
3. List important points in the source that the summary omitted, ranked by importance.
4. Check numbers, names, dates, and qualifiers (may, some, up to) for exactness.
5. Check balance: does the summary favor the start of the source or one side of an argument?
6. Score: faithfulness (percentage of claims supported), coverage (important points included), concision, and clarity, with the arithmetic.
7. Provide a corrected summary using only supported claims.

Return a table: Claim # | Claim | Evidence quote and location | Verdict | Fix. Then a short verdict (accept, revise, reject) with the reasons. Add a checklist for human reviewers and a warning that automated checks by the same kind of model can miss errors, so spot-check manually. Never add information that is not in the source.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.

• faithfulness