How to read an AI answer sample

Keep search and the answer apart

An AI answer and a Google ranking are not the same lever. Write them in two columns and do not let a sales deck merge them. You might publish pages, sell something, or answer the same questions in support. None of that turns one chat screenshot into a measurement.

This page is the method for a baseline.

Write the questions people actually ask

• The pages, the name, or the offer you want checked.
• The questions customers actually ask, not a list written to flatter a model.
• The reporting you already have, such as Search Console, if you want a later change tracked there.

Record the sample

Write the sample size and the dates on the same page as the finding. Note where those answers mentioned you, cited you, or skipped you, and who they cited instead. A later check can use the same questions. One screenshot of one chat is an anecdote. Do not call it a measurement.

If the sample is too small to say anything steady, say that. Do not stretch it into a claim about the whole site.

Track the change with your own data

Movement after the baseline belongs in first-party data such as Search Console and referral tags. The sample tells you what those answers did on those dates. It does not tell you that the next sample will move.

A context file such as llms.txt can help a system that chooses to read it. Google has said that file does not help or hurt Search, AI Overviews, or AI Mode. If the file is present, that is a fact about the site. It is not a Search win. Showing up as a supporting link in an AI answer still depends on a page being indexed and eligible for a snippet. There is no extra markup that forces a citation.

Do not invent a GEO score or a competitor's private numbers. If you want the repeating job written down, that is the workflow guide. If you want a room of people working from their own material, that is the workshop guide. This page is the sample.