An AI visibility audit is a structured check of your visibility in AI search: whether AI tools can reach your site, cite your pages, describe your company correctly and name you when buyers ask about your category.
Run it in that order, because each layer depends on the one before. Free tools cover every step, and the output is a short list of fixes, each with an owner, plus a baseline that every later round is compared with.
Set the scope before you open a tool
Write down three lists first, because every later step uses them: the engines your buyers use, the rivals you meet in deals and lose to, and a fixed set of buyer questions, asked the way buyers see each tool.
Then pick the pages that matter commercially: the home page, each service or product page, pricing if you publish it, and the articles that bring in the most qualified visits. The audit checks those closely and samples the rest.
Access: can the engines reach you?
Start here, because nothing later counts if a crawler is turned away. Google says AI Overviews and AI Mode have no requirements beyond ordinary Search: a page must be indexed and eligible to show with a snippet.
So the Google check is plain SEO. Confirm the key pages are indexed in Search Console and that no nosnippet or max-snippet tag limits them. Google also asks that crawling be allowed by any CDN or hosting setup, which robots.txt alone will not show.
For ChatGPT, the crawler to check is OAI-SearchBot. OpenAI says sites that opt it out “will not be shown in ChatGPT search answers”. Blocking GPTBot, its training crawler, is a separate choice that does not affect search.
Then look past robots.txt: at a firewall or bot rule that turns away crawlers robots.txt allows, and at the Search Console setting that can exclude a whole property from AI features. A month of server logs settles what reaches you.
Fail if a key page is not indexed, carries a snippet limit nobody intended, or turns away a search crawler.
Accuracy: what do AI tools say about you?
Ask each engine the plain questions a buyer asks before a first call: what your company does, who it is for, what it costs, where it works, and how it compares with a named rival. Read the answers against your own facts, line by line.
Log every error with its likely source. An old price usually traces back to an archived page or a directory listing, a retired service to an old article, a wrong location to an outdated profile. Google tells site owners to keep their Business Profile current, and the same goes for every listing an answer cites.
Fix the source, then tell the engine. Google’s AI Overviews carry a thumbs down button with a Report a problem option for inaccurate answers. Feedback does not change the page an answer drew on, so the source comes first.
Fail on any wrong price, service, location or category, and on any answer that confuses you with another company.
Presence: how often are you named?
Now run the category questions, the ones that do not name you, several times on each engine, and count how often you are named against the rivals on your list. That count is your AI share of voice, the baseline every later month is compared with.
Note which rival is named most, and how the engine describes it. Those words show the position you are competing with, and often the page or listing they came from.
There is no pass or fail here, only a number and its trend. A share near zero on questions you should win is the finding.
Citations: which pages do engines use?
Bing Webmaster Tools’ AI Performance report shows how often your pages are cited in Copilot and Bing’s AI summaries, which URLs, and the grounding queries behind them. Search Console adds how often your URLs appeared in AI Overviews and AI Mode.
Read both against your commercial pages. A service page that is indexed but never cited needs work on its content. A post cited often for a question close to your offer is a page to strengthen and link from.
Then list the outside sites the answers cite for your category: review platforms, directories, trade publications and forums. They are where AI answers look for the names they recommend, so whether you appear there, and correctly, belongs in the audit.
Fail if a key page is never cited while a weaker one is, or if you are missing or wrong on a site that keeps being cited.
Pages: does each one answer its question?
For each commercial page, write the buyer question it should answer. Then check whether its first two sentences answer it in plain words, with the facts an engine needs: who it is for, what it includes, and what it costs or how pricing works.
Check that the answer is in the HTML the server sends, not added later by scripts, and that structured data matches the visible text. Google says its AI features need no new files and no special markup.
A page that ranks can still be passed over, because ranking and being quoted are separate jobs, and an opening that answers nothing is usually the cheapest fix on the list.
Fail any commercial page whose opening does not answer its question.
Traffic and agents: what reaches the site
Check that analytics can see AI visits at all. Google Analytics files some assistants under its AI Assistant channel, but others still land in Referral, so set up the channel before you report a number.
Last, try your main form the way an agent would. Can a browser agent find it, fill it in and submit it without guessing? If not, a standard that lets a site hand agents its forms as tools is one for later, not a first fix.
Fail if AI visits cannot be told apart from the rest of your traffic.
The findings sheet: an example
The audit ends as one sheet: each finding, the layer it sits in, who owns the fix, and how urgent it is. The rows below are made up, to show the shape.
| Finding | Layer | Fix, and who owns it | Priority |
|---|---|---|---|
| A bot rule on the CDN blocks OAI-SearchBot | Access | IT: allow the crawler and OpenAI’s published IP ranges | First |
| Two engines quote last year’s price | Accuracy | Marketing: correct the pricing page and the listing they cite | First |
| Main service page indexed, never cited | Citations | Content: rewrite the opening to answer the buyer question | Next |
| Missing from the review site most answers cite | Citations | Marketing: complete the listing and ask customers for reviews | Next |
| Named in 2 of 60 category answers | Presence | Marketing: track monthly as the fixes above land | Track |
| AI visits counted as Referral | Traffic | Analytics: add the AI channel | Next |
Order the fixes by layer, not by size. An access fix can change everything after it, and content work is wasted on a page a crawler cannot reach.
When to run it again
Run the full audit once a quarter, and after any migration, redesign or change to CDN or firewall rules, since those are the changes that quietly undo access. In between, track share of voice and the two citation reports monthly.
Keep the questions, engines and method the same each round, and note what changed on the site since the last one. An audit compared with nothing is only a snapshot.
● About the author
Hamza Hashim is the marketing strategist at Katama. He owns the channel mix and the order it gets built in: what ships first, what waits for evidence, and what gets stopped when the evidence says stop. Meet the team.