Running an AI visibility audit

Write down the questions a buyer would actually ask on the way to choosing something like you. Include the awkward ones — the objections your sales team hears — because those sit closest to a decision and are usually the weakest spot.

Thirty to fifty questions spanning discovery, comparison, objection and intent is enough to be representative. Below twenty, normal run-to-run variance will look like signal.

The order matters, because each one explains the next.

• Access — can the retrieval crawlers reach your pages at all? Everything else is moot if not.
• Presence — on how many of those questions are you named, and in what position?
• Sources — which pages do the answers cite? This is where the leverage is.
• Accuracy — what do engines state about your pricing, features and availability, and is any of it out of date?

A low score on its own means little. What matters is your figure against the competitors named on the same questions, because a crowded category produces lower numbers for everyone in it.

Treat the first run as a baseline rather than a verdict. Answers are generated fresh each time and are not deterministic, so the number worth watching is the trend across scheduled runs.