How to run an AI search visibility audit
A repeatable way to measure where your business shows up in AI answers, which sources get cited, and what’s worth fixing first.
By Dillon Jones · · 3 min read
Start with the buyer’s questions
A brand-name prompt is useful for catching incorrect facts, but it doesn’t tell you much about discovery. Start with questions people ask before they know your name. Ask sales for the comparisons they hear, read the language people use in support tickets, and look at the queries that already lead to your site.
Put together a small, reusable set: problem questions, comparisons, selection criteria, and a few brand checks. Group them by what the searcher is trying to decide. A recommendation for “CRM for a two-person sales team” is a very different thing from a definition of CRM.
Separate mentions from citations
For each run, log the prompt, surface, date, and any relevant settings. Record whether your business was mentioned, whether a page you own was cited, and what sources supported the answer. These are different signals. An assistant can name you without citing you, or cite your documentation while recommending someone else.
Run the same questions on a repeatable schedule, and don’t read too much into one response. Answers can change across sessions, locations, and product updates. Keep a few examples so you can compare the substance, not just a count in a spreadsheet.
Fix what would help a person, too
Check that the relevant pages can be crawled and that they answer the question directly. Don’t make the reader navigate three pages to learn whether you offer something. Give the answer early, explain the tradeoffs, and back up claims with sources or examples that you can stand behind.
Prioritize the gaps that matter to a buyer and can be verified after a change. An outdated specification or unclear service page is a better first fix than rewriting the whole site because one assistant cited a competitor.
Keep visibility separate from business impact
Use repeat mentions and citations as directional evidence. Report referral visits or conversions separately where you can measure them. Some answers will be read without a click, and referral data won’t capture every visit. The point is to learn where discovery happens and which improvements are worth testing, not to turn a screenshot into an ROI claim.
Further reading
Dillon Jones
Digital marketing strategist and founder of FoundHive. More about me.