AI SEARCH (GEO)

Audit whether a page answers its question directly

Use when a page targets a question but you suspect the answer is buried.

does-page-answer-question.md

Download .md

You are auditing whether a page answers its target question directly enough to be used by an AI assistant.

Target question: {{QUESTION}}
Page content: {{PAGE_TEXT}}

Do this in order:
1. Quote the single passage that comes closest to a direct answer, and give its approximate position (percentage through the page).
2. Score the page 0 to 5 on each of: directness, completeness, self-containment, evidence, freshness signals. Give one sentence of justification per score, and quote evidence.
3. State plainly whether an assistant asked this question could answer it using only this page. Yes or no, then why.
4. List every sub-question a reader would still have, ranked by how likely it is to be asked.
5. Write a replacement answer block of 60 to 90 words to place directly under the H1.

Constraints:
- Do not pad scores. A 3 means average, not polite.
- Quote real text from the page for every claim you make about it. If you cannot find supporting text, say "not present on page".
- The replacement block must not use em dashes and must not restate the question before answering it.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{QUESTION}}
{{PAGE_TEXT}}

Getting a better result

If the closest answer sits past 40 percent of the page, that is usually the whole problem.
Run the audit again after editing to check the score moved for the right reason.
Keep the replacement block as plain paragraph text; assistants lift prose more reliably than styled callouts.

Questions about this prompt

When should I run this rather than extracting citable claims?

When a page was built for one specific question and you suspect the answer arrives late. Claim extraction tells you what is quotable anywhere on the page. This is narrower: it asks whether {{QUESTION}} is answered at all, and how far down a reader has to go before anything resembling an answer appears.

What do I need in front of me?

The target question worded as a reader would type it, and the full page text including the preamble you are tempted to trim. The position figure in step one is a percentage through what you pasted, so cutting the introduction beforehand hides the exact delay the audit exists to measure.

What does it give me back?

The closest passage quoted with its position on the page, five scores out of five, a plain yes or no, a ranked list of remaining sub-questions and a replacement block of 60 to 90 words. The position figure is the finding. Anything past forty per cent usually explains the whole result.

What is the mistake that costs me?

Placing the replacement block under the H1 and leaving the original buried answer where it sits. Two answers to one question compete, and the lower one often carries older wording. Move or delete the old passage, then rerun and check the directness score moved for that reason rather than by luck.