An AI visibility audit answers one question: when a buyer asks an AI assistant who to hire for what you do, is your company in the answer, and is it described correctly? Then it answers the question that matters more: if not, why not?
"Run an audit" can mean anything from a screenshot of one ChatGPT answer to a scored, repeatable measurement. Here is what a useful one checks, in the order we run it on our own site and for clients.
1. Can the engines read your site at all?
Before content, plumbing. An engine can only cite a page it can fetch and understand.
- Crawl access. Your robots.txt and firewall let the AI crawlers in, and pages return their real content to them, not a bot wall or an empty shell that only fills in with JavaScript.
- Indexing. This is the check most audits skip. Each answer engine retrieves from a search index, often someone else's. Copilot draws on Bing. Gemini and Google's AI answers draw on Google. Claude's web search draws on Brave. If your site is missing from the index an engine searches, no content you publish can get you cited by it. We found our own site was absent from Brave's index entirely, which explained why Claude named us in none of 24 answers; we wrote it up in The AI that reads our site most has never heard of us.
- Basics that still count. Real "page not found" responses, a sitemap with true dates, canonical URLs, a permanent redirect between your www and bare domains.
2. Do you answer the questions buyers ask?
AI answers lift short, self-contained answers to specific questions, and they prefer pages that show they are current. The audit checks whether your key pages answer the questions a buyer would type, in plain sentences an engine can quote, with a visible updated date. A service page that explains the offer in a few vague lines gives an engine nothing to work with.
3. Do you say the same thing about yourself everywhere?
Engines assemble a description of your company from many places: your title tags and meta descriptions, your about page, structured data, directory profiles, review sites, and machine-readable files. When those disagree, the engine picks, and you do not get a vote. The audit lists every place your company is described and flags each contradiction: an old product name, a stale price, a different one-line description.
4. Do outside sources mention you?
AI tools lean on third-party sources to decide who to name: review sites, directories, forums, news. A company that only describes itself on its own site is easy to leave out. The audit checks which of those sources mention you, and whether they mention you consistently.
5. Are you actually named? Measure it properly.
This is the part most tools get wrong. AI answers vary from one run to the next, so one answer is a sample, not a result. A useful measurement:
- asks a fixed set of real buyer questions, written for your business,
- runs each question on each engine more than once and saves the full answer,
- classifies every answer as named and correct, named but wrong, or absent,
- sets aside runs where the engine never searched the web, instead of counting them as a zero,
- repeats on a schedule, so you see a trend instead of a snapshot.
What you should get at the end
A score you can track, the reason behind it, and a short, ranked list of fixes, each with what it changes and how you will know it worked. If an audit hands you a single confident number with no way to check the answers behind it, ask for the answers.
We run this method on our own site first and offer a free first read for yours on our AI visibility page. For the broader playbook, see How to improve your AI visibility.