What Is in an AI Visibility Audit, and How to Run Your Own

An AI visibility audit checks whether AI tools such as ChatGPT, Perplexity, Copilot, Gemini and Google AI Overviews can reach your website, whether they mention your organisation, whether what they say is accurate, and which sources they rely on instead of you. It ends with a prioritised list of fixes. A small organisation can run a useful first audit in about half a day with free tools and a spreadsheet.

An audit is different from monthly tracking. Tracking tells you whether the numbers move; we explain how in how to measure AI search visibility. The audit comes first and answers a harder question: why you are, or are not, in the answers.

What an AI visibility audit covers

A complete audit has six parts. Run them in this order, because the early ones can explain the later results.

Part	Question it answers	Main tools
1. Access	Can AI crawlers reach and read the site?	robots.txt, CDN settings, a browser with JavaScript off
2. First-party data	What do Google and Bing already report?	Search Console, Bing Webmaster Tools, analytics
3. Prompt tests	Are you named when buyers ask?	The assistants themselves, a spreadsheet
4. Accuracy	Is what they say about you right?	Your own fact sheet
5. Sources	Who do they cite, and why them?	The citations in each answer
6. Page readiness	Are your key pages worth quoting?	Your top service pages

How to run your own audit

Step 1: Check access

Start with the plumbing, because a blocked crawler makes everything else pointless.

• Open yourdomain.com/robots.txt and look for rules naming OAI-SearchBot, PerplexityBot, Claude-SearchBot, Bingbot or a blanket Disallow: /.
• Check your CDN, firewall or security plugin for “block AI bots” settings. These are sometimes switched on by default or by a previous supplier.
• Load your three most important pages with JavaScript disabled. If the main text disappears, some AI crawlers may not see it either. We explain why in what is LLM SEO.
• Confirm the pages are indexed in both Google and Bing. Copilot draws on Bing’s index, so a site missing from Bing is missing from Copilot.

Step 2: Pull the first-party data

Two search engines now report AI visibility directly, and most organisations have never opened either report.

• Google Search Console. The Generative AI performance report shows how many times links to your site were shown in AI Overviews and AI Mode, broken down by page, country, device and date. Google says it had rolled out to all websites by 31 August 2026, though sites with very few AI impressions may not see it. Note which pages earn AI impressions and which important pages earn none.
• Bing Webmaster Tools. The AI Performance report shows how often your pages are cited in Copilot and Bing’s AI summaries, and the “grounding queries” the AI used when it retrieved them. Those queries are a rare view of how an assistant phrases its own searches.
• Analytics. Filter referral traffic for chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com and claude.ai over the last six months. Treat it as a minimum, because some AI visits arrive without a referrer.

Write down three numbers: AI impressions in Google, citations in Bing, AI referral visits. That is your starting point.

Step 3: Run the prompt tests

Write 15 to 25 prompts across three types: category (“best [service] for [audience] in [place]”), problem (“how do I [problem you solve]?”) and brand (“what does [your name] do?”, “is [your name] any good?”). Run each one in ChatGPT, Perplexity, Copilot and Gemini with web search on, in a logged-out or clean session, and check Google for an AI Overview. Run the category prompts twice, because answers vary.

For each answer, record whether you appear, where, and who else is named.

Step 4: Check accuracy

Before you start, write a one-page fact sheet: legal name, trading name, location, services, sectors, prices or price ranges, founding year, key people, accreditations. Then compare every brand answer against it.

The errors we see most often are an old address, a service you stopped offering, a founder who left, prices from an old page, and confusion with a similarly named business. Each error usually traces back to a specific source, which is why the next step matters.

Step 5: Read the sources

List every domain the assistants cited across all your prompts and count how often each appears. You will usually find three groups:

1. Your own pages. Good. Note which ones.
2. Third-party pages that mention you. Directories, reviews, client sites, press. Check each for accuracy.
3. Third-party pages that mention competitors but not you. Comparison articles, “best of” lists, sector directories, forum threads. This group is your outreach list.

If an answer contains a wrong fact about you, look for the cited page that contains it. Correcting that source is often the fastest fix in the whole audit.

Step 6: Check page readiness

Take the five pages that should win your most valuable prompts, usually your main service pages. For each, check:

• Does the first paragraph say what you do, for whom, where and roughly at what price?
• Do the headings match the questions buyers ask?
• Would each paragraph make sense if quoted on its own?
• Is there evidence: named clients, figures, dates, credentials?
• Does it answer the obvious follow-ups, such as cost, timescale and how it differs from the alternative?

Pages that fail these checks are the ones to rewrite first. How to appear in Google AI Overviews covers the rewriting.

Turning findings into a plan

Put every finding in one table with five columns: the finding, the evidence, the fix, who owns it, and priority. Rank by a simple rule:

1. Access problems first. Blocked crawlers or missing indexing cancel everything else.
2. Wrong facts next. Inaccurate answers can cost you enquiries today.
3. Page rewrites third. These move retrieval-based answers within weeks.
4. Third-party mentions last. They matter most over time but take longest to earn.

Re-run the same prompts in a month to see what changed.

What an agency audit adds

You can do everything above yourself. An agency audit is worth paying for when you need scale or a second opinion. In practice it adds a larger prompt set run several times per engine, a direct comparison with named competitors, a review of your entity details across every platform, and someone who has seen the same patterns across other clients and can tell a quick fix from a long project. It should also hand you the raw answers, not only a score.

The GEO service we run starts with exactly this audit, then fixes the gaps on and off your site, with the page rewriting covered by AEO. If you run the audit yourself first, send us the spreadsheet. It makes the conversation much shorter.