How to Audit Your AI Visibility: A Step-by-Step Guide

An AI-visibility audit answers one question: when buyers ask AI what to buy in your category, does it name you, a competitor, or no one? You can run a useful version yourself with a spreadsheet, a fixed set of buyer questions, and the free tiers of each engine. This guide walks through the five steps, from building the question set to turning the findings into a plan.

It is worth the hour because the research is now happening where you cannot see it. In G2's 2026 survey, 51% of B2B software buyers said they begin product research in an AI chatbot more often than in Google. That activity rarely shows up in your analytics, so an audit is often the only way to find out whether AI is quietly handing your buyers to a competitor.

What an audit covers

A complete audit looks at three things, not just the first. Many teams check only whether AI mentions them and stop, which tells you the symptom but not the cause.

• Answer-level visibility. Across your priority questions and engines, are you cited, and how do you compare to competitors?
• On-site extractability. Do your pages state clear answers an engine can lift, or is the answer buried?
• Off-site authority. Are you present, and accurate, in the third-party sources AI draws on for your category?

Step 1: Build your question set

Write down ten to twenty real buying-intent questions, phrased the way a buyer would ask, not the way your marketing describes you. The most important rule: never name your own brand in a question. The test is whether AI recommends you unprompted, so a question that already contains your name is a gimme that proves nothing.

Use category, use-case, and comparison framings. For a software product, that looks like best [category] software for [use case], top alternatives to [a competitor], or which [category] tool do teams actually recommend. Keep this set fixed, because consistency from month to month is what lets you see trends.

Step 2: Run them across the engines

Buyers do not all use the same assistant, so check more than one. Run each question across ChatGPT, Perplexity, Google AI Overviews, Gemini, and Claude. Two practical notes make the results trustworthy.

• Use a logged-out or fresh session. Personalization and chat history skew answers toward what you have searched before. A clean session approximates what a new buyer sees.
• Run each question more than once. Generative answers vary by session, so a single result can mislead. Sampling each question a few times gives you a fairer picture.

Step 3: Log citations and gaps

Record the results in a simple grid so patterns become visible. One row per question, one column per engine, plus columns for who got named and which sources the engine cited.

Two patterns matter most: the questions where you never appear, and the sources the engines keep citing. The first is your gap list. The second tells you where your off-site authority needs to be.

Step 4: Check on-site extractability

For each priority question, look at your own page that should answer it. Ask whether the answer is stated plainly in the first few sentences, supported by specific facts, and structured with clear headings. Engines favor content they can lift with confidence, so a buried or hedged answer is easy to skip over.

Check the basics too: is the page actually indexable, and does it carry schema markup for your organization, products, and FAQs? If an engine cannot retrieve or parse the page, it cannot cite it, no matter how good the content is.

Step 5: Check off-site authority

Now look outside your own site, because off-site sources often decide AI answers more than your pages do. Using the sources you logged in step three, check where you stand.

• Review platforms. Are you present with real, recent reviews on the sites your category uses, for software typically G2 and Capterra?
• Communities. Do you appear, fairly and authentically, in the Reddit and forum threads where your buyers compare options?
• Reference and roundups. Is your reference presence accurate, and do you appear in the best-of lists and comparison articles the engines keep citing?

Turning the audit into a plan

The output of an audit is not a score. It is a prioritized list: the questions where you are absent, the on-site pages that need a clearer answer, and the off-site sources where you are missing or inaccurate. Rank them by how close the question is to a buying decision, and start there.

Then make it repeatable. Re-run the same question set monthly so you can see citation share move as you work. A baseline you never revisit is trivia; a baseline you track is a program. If you would rather not run it by hand, a free scan can give you a fast first read of where you stand before you commit the time.