All insights
Method21 Nov 20267 min read

How to audit your share of answer across the major engines

A repeatable way to capture what AI says about your brand, so improvement can be measured against something real instead of a screenshot someone took once.

The first AI visibility audit most teams run is a person typing their brand name into ChatGPT and reading the answer. It is a useful shock, and a useless baseline. A single answer tells you nothing about frequency, consistency or which prompts you lose.

Start with prompts, not brands

Build a fixed set of 20 to 40 prompts drawn from the questions your buyers actually ask. Three families are usually enough:

  • Category prompts. "Best B2B providers of X." These show who owns the category in the model's view.
  • Comparison prompts. "X versus Y" and "alternatives to X." These show whether you appear at the moment of shortlisting.
  • Problem prompts. "How do we solve Z." These show whether you are cited as the source of an approach.

Keep the wording exactly as a buyer would phrase it, and freeze the set. Changing prompts mid-programme destroys comparability.

Capture verbatim, with dates

For each prompt, record the full answer, the sources cited, the brands named and the order they appear in. Date every capture and note the model and interface, because answers differ between a logged-in session, an API call and a search-grounded mode.

Two captures per prompt, a week apart, is enough to see whether an answer is stable or noise. Unstable answers are not a result; they are a warning that the prompt is too generic to measure.

Score three things

Presence. Are you named at all, for each prompt family?

Accuracy. When you are named, is the description right? A wrong but flattering summary is a liability, not a win.

Source share. Which domains are being cited? This is the most actionable number in the whole audit, because it tells you who to out-publish rather than who to out-bid.

Report movement, not snapshots

A baseline is only worth having if you re-run it. Monthly is enough for most B2B categories, quarterly is defensible for slow-moving ones. Report the change against your own baseline and name the work that shipped between captures, so the number has a cause attached.

Be honest about variance. Models change, prompts behave differently by region, and a competitor's press coverage can move an answer without you doing anything. Presenting a single flattering capture as proof of success is the fastest way to lose the room.