# Compare assistant answers against your page

> Use when an assistant answers a question about you and you want to know where it got it wrong.

## Fill in before running

- `{{QUESTION}}`
- `{{ASSISTANT_ANSWER}}`
- `{{PAGE_TEXT}}`

## Prompt

```
You are comparing an AI assistant's answer against the source page it should have used.

The question asked: {{QUESTION}}
The assistant's answer: {{ASSISTANT_ANSWER}}
Our page content: {{PAGE_TEXT}}

Produce:
1. A table: | Statement in the answer | Present on our page? (verbatim quote or "absent") | Verdict (accurate / distorted / outdated / fabricated / from elsewhere) | Likely cause |
2. A list of things our page says that the answer omitted, ranked by how much the omission changes the meaning.
3. For each distorted or fabricated statement, the exact sentence to add or change on our page to make the correct version easier to lift than the wrong one.
4. A one-paragraph summary of the single structural reason the answer went wrong, if there is one.

Constraints:
- Break the assistant answer into individual statements before judging. Do not assess it as a whole.
- Quote our page verbatim in column two, or write "absent". No paraphrasing.
- Do not assume malice or model failure where an ambiguous sentence on our page explains it.
- Do not recommend contacting the model provider.
```

## Getting a better result

- Most distortions trace back to one ambiguous sentence, so look for the shared cause before fixing individually.
- Save the assistant answer with a date; you will want the before and after.
- "From elsewhere" verdicts tell you which third-party page is outranking yours as a source.

---

From the QuQi prompt library - https://www.quqi.io/prompts/assistant-answer-diff