GuidesGet cited by AI

How to run an AI visibility audit

An AI visibility audit answers one question: when a buyer asks an AI tool about your category, does it name you, say true things, and link to you? Write 20 to 40 buyer prompts, run them across ChatGPT, Gemini and Perplexity, score every answer for mention, accuracy and source, then fix the pages and gaps the results point to.

Updated November 16, 20265 steps5 min read

Check my AI visibility →

$1 today. Cancel anytime before day 7.

The audit, start to finish
  1. 1Write the prompts20 to 40 questions real buyers ask, by funnel stage.
  2. 2Run them everywhereChatGPT, Gemini, and Perplexity, with web search on.
  3. 3Score each answerMention, position, accuracy, and the sources cited.
  4. 4Find the gapsGroup misses by cause: no page, weak page, or no proof.
  5. 5Fix and rerunChange the pages, then repeat the same prompts later.
On this page
  1. Build the prompt set first
  2. Run and record every answer
  3. Turn the results into causes
  4. A simple scoring sheet
  5. Audit the competitors too
  6. Fix, then measure again
  7. Questions

Build the prompt set first

The audit is only as good as the questions. Write them the way buyers type: category searches (“best payroll software for a 10-person agency”), comparisons (“X vs Y”), problem questions (“why is my invoice not syncing”), and direct brand questions (“is Acme legit”). Aim for 20 to 40, spread across the funnel.

Take the wording from real sources: sales calls, support tickets, Search Console queries, and People Also Ask boxes. Keep the set fixed once you start, because you will rerun it. The AI search prompts guide goes deeper on choosing and tracking them.

Do not skip the brand prompts. What the engines say when asked directly about you shows how you are described to anyone who already knows your name, including prospects who researched you before a demo. Errors here are the cheapest to fix and the most damaging to leave.

Run and record every answer

Run each prompt in ChatGPT, Gemini, and Perplexity with web search on, and in a logged-out or clean session so personal history does not skew results. For each answer record four things: whether you are mentioned, where in the list, whether the facts are right, and which sources were cited.

Answers differ between runs, so run each prompt two or three times before you trust a result. A spreadsheet works: prompt, engine, date, mentioned yes or no, accuracy notes, sources. Expect to spend a few hours by hand, which is why most teams automate it after the first pass.

Be careful with personalized results. Some tools remember your history and preferences, and logged-in sessions can tilt an answer toward brands you have visited. Use a clean browser profile or a logged-out session, and note the location you ran from, since local intent can change what appears.

Does ChatGPT mention you yet?

Checks whether ChatGPT, Gemini and Perplexity mention your brand. Tracks it every week.

Turn the results into causes

Sort the misses into three buckets. No page: you have no content that answers the question. Weak page: you have one, but it is thin, buried, or contradicts another page. No proof: the page is fine but nobody else mentions you, so the model has nothing to cross-check.

Also read the sources that were cited for competitors. If the same review site, directory, or Reddit thread appears across several prompts, that is a place to earn a mention. Wrong facts, such as old pricing or a closed location, trace back to a page or listing that still says it.

Then check the basics a retrieval tool needs: indexing, crawl access, and a clear page for each product or service. The SEO site audit covers the technical half.

A simple scoring sheet

Give every answer a score across four columns. Mention: 0 if absent, 1 if named. Position: note whether you are first, in the top three, or lower. Accuracy: 0 if a fact is wrong, 1 if all is correct. Source: whether a page you own was cited.

Add up each column by engine and by prompt group. A pattern such as strong accuracy but low mention points to a visibility problem, while high mention with low accuracy points to stale or conflicting information on the web.

Keep a notes column for oddities: a competitor you did not know about, a feature you are credited with that you do not have, a source you did not expect. Those notes often yield the best actions in the audit.

Be ready for uncomfortable results. Many brands discover they are absent from prompts they assumed they owned, or described with a feature set from three years ago. That is the point of the audit, and the earlier you see it, the cheaper it is to fix.

Audit the competitors too

Run the same prompts and note which competitors the engines name, in what order, and with which descriptions. The competitors that matter in AI answers are not always the ones in your ad auctions, and the mismatch is informative.

Look at how the engines describe them. If a rival is consistently called the easiest to use or the best for small teams, you have learned the label buyers are being handed. Compare it with the story on your own pages and find where they differ.

Then read the sources behind those descriptions. They are your content brief: the review site that compares tools, the community thread that recommends one, the guide that names the leaders.

Fix, then measure again

Work the list in order of buyer intent. Write or rewrite the pages for the highest-intent misses, put the direct answer first, correct stale facts everywhere they live, and pursue the two or three outside sources that keep recurring. Get cited by ChatGPT lists the page shapes that get quoted.

Rerun the same prompts in four to six weeks. Changes to a model’s answers lag behind changes to the web, and not every prompt will move. The part you can hand off: Sprites can run your buyer prompts through ChatGPT, Gemini and Perplexity each week, show which brands they name, and log the change from the baseline. The free AI visibility checker gives a quick first read.

Questions

What is an AI visibility audit?

It is a structured check of whether AI answer engines mention, describe, and link to your brand for the questions your buyers ask. It ends with a list of fixes ranked by buyer intent.

How many prompts should I test?

Twenty to forty is enough for a first audit if they are spread across category, comparison, problem, and brand questions. Keep the set fixed so you can compare runs.

How often should I repeat it?

Monthly is a sensible default, weekly if you are actively changing pages. Answers vary run to run, so compare patterns rather than single results.

Is an AI visibility audit the same as an SEO audit?

No. An SEO audit checks how your pages are crawled and ranked. An AI visibility audit checks what answer engines say about you. The two overlap on indexing and content quality.

Keep going

See if ChatGPT recommends you

  • Checks whether ChatGPT, Gemini and Perplexity mention your brand
  • Shows which competitors they name instead
  • Tracks it every week