ExplainerAI visibility
How to Run an AI Visibility Audit (and What You Will Find)
by the AEO GEO Labs team3 min read
An AI visibility audit answers four questions: for the things your buyers ask, which engines name you, which cite you, who gets the spot when you are absent, and what the engines seem to be reading. It is the starting point for every piece of AI visibility work, because without it you are guessing which of a dozen possible problems you have.
Here is the method, scaled down so a single person can run it in a day.
Step 1: Build the question set
Write 30 to 50 questions real buyers ask, across the journey:
- Category questions. "What is [category] software?" "How does [category] work?"
- Selection questions. "Best [category] for [buyer type]." "[Category] tools under $X." "[Category] with [integration]."
- Comparison questions. "[You] vs [competitor]." "[Competitor] alternatives."
- Fact questions. "How much does [you] cost?" "Does [you] integrate with [tool]?"
Pull them from sales calls, support tickets, and the questions engines themselves suggest. Fix the set; you will reuse it.
Step 2: Run every question through every engine
ChatGPT (with search), Perplexity, Gemini, Google AI Overviews, Claude, and Copilot. Record for each answer: whether you are named, whether a page of yours is cited, which other brands are named, every source cited, and any factual error about you.
A spreadsheet works. One row per question per engine.
Step 3: Compute share of answer
Per engine, the percentage of questions where you are named or cited, with named and cited tracked separately. Do the same for your top three competitors. This is your baseline.
Step 4: Build the source log
Count every cited domain across all answers. Sort by frequency. This list is the most valuable output of the audit. It tells you which review sites, communities, publications, and vendor pages engines actually read for your category. In most categories, five to eight domains account for most citations.
Step 5: Check the plumbing
For your own site: robots.txt and CDN rules for each AI crawler, whether key pages render without JavaScript, whether schema is present, valid, and matches the page, and whether an llms.txt exists. A block or a rendering gap can explain a zero share on its own.
Step 6: Diagnose
Pattern-match your results:
| Finding | Likely cause | First fix |
|---|---|---|
| Zero share everywhere | Crawl block or rendering | Fix robots.txt, CDN, SSR |
| Named but not cited | Weak or missing pages | Write answer-ready and comparison pages |
| Cited but not named | Reputation gap | Get into cited sources |
| Named with wrong facts | Stale third-party source | Find and correct the source |
| Strong on one engine, absent on another | Engine-specific source preference | Check that engine's source log |
What audits usually find
At AEO GEO Labs the first-month findings are strikingly consistent: missing from most ChatGPT answers, a competitor cited several times more often on Perplexity, Gemini quoting last year's pricing from a directory, no structured data on key pages, and absence from the handful of sites the engines cite most. The order of fixes follows from that: unblock and render, write the pages, fix the facts, get into the sources, then measure weekly.
Step 7: Repeat
Run the same question set monthly at minimum, weekly if you are actively working on it. Share of answer only means something as a trend tied to what you shipped.
Shortcut
The one-day version above is worth doing. The full version runs a few hundred questions through every engine, logs sources over 30 days, and tracks weekly. AEO GEO Labs runs that version live on a 20-minute call and shows you, engine by engine, who the AI recommends instead of you.