How do I audit my own AI visibility without buying a tool?

Start with the questions your buyers ask, not your brand name

Begin by writing down the real questions a customer types before they buy. Not your company name, but the problem they have. Think "best accountant for freelancers" or "how do I reduce warehouse returns". Aim for fifteen to twenty questions that cover your services, your locations and your common objections. These prompts are the backbone of the whole audit, so spend proper time getting them close to how people actually speak.

Group the questions into three types. First, category questions where someone wants a recommendation. Second, comparison questions that pit two options against each other. Third, problem questions where someone describes a symptom and wants a fix. Each type triggers a different kind of AI answer, and you want to see how you show up across all three. Write each question in plain language, exactly as a person would.

Keep the list in a spreadsheet with a column for the question and a column for its type. You will reuse these same prompts every month, so a stable list matters more than a long one. Do not tune the wording to favour your business. The point is to see the honest answer a stranger gets, not the flattering one you could coax out with a leading prompt.

Run the same prompts across every major assistant

Ask each question in ChatGPT, Perplexity, Google AI Overviews and Microsoft Copilot, one at a time. Use a fresh session or a logged out window so past chats do not skew the answer. Perplexity and Copilot show their sources on screen, which makes them the easiest place to see which pages an assistant trusts. Google AI Overviews appears above normal search results for many questions and links out to the pages it summarised.

For each answer, record three things. Whether your business was named at all. Whether a competitor was named instead. And which web pages the assistant cited as its sources. Copy the cited links into your sheet. These citations are the real prize, because they tell you exactly which pages the model reads when it forms an opinion about your market. Named competitors show you who currently owns the answer.

• ChatGPT: ask with browsing on so it pulls live sources you can inspect.
• Perplexity: read the numbered citations under each claim.
• Google AI Overviews: note the links in the expandable source panel.
• Copilot: check the footnote links it attaches to each sentence.

Score what you find in a simple sheet

Turn your notes into a score you can track over time. For every question, mark whether you were named, whether you were cited with a link, and whether a rival appeared instead. A count of "named" versus "not named" across twenty questions gives you a baseline visibility rate in one glance. You do not need fancy maths. A plain tally tells you where you stand and where you are invisible.

Now sort your questions by outcome. The questions where rivals get named but you do not are your priority list. Look at the pages those rivals rank through. Often it is a clear, well structured page that answers the exact question in its opening lines. Read those pages and note what they do that yours does not. Frequently the answer is simply that they state a direct answer early and back it with specifics.

Check whether the sources are pages you control

Look hard at the citation links you collected. Sort them into pages you own, pages you could influence and pages you cannot touch. Your own website, your blog and your help centre are things you can change this week. Directory listings, review sites and industry roundups are things you can often get added to. Forums and news articles sit further out of reach. This split tells you how much of the problem is fixable directly.

When assistants cite third party pages instead of yours, it usually means your own pages do not answer the question cleanly. Models favour pages that give a direct answer, use clear headings shaped like questions, and include concrete detail such as who a service suits and how it works. If your site buries the answer under marketing language, the model skips it and quotes someone who wrote plainly instead.

Also confirm the assistants can actually read your site. Check that your key pages are not blocked to crawlers like GPTBot, PerplexityBot, Google-Extended and OAI-SearchBot in your robots file. A page that answers the question perfectly is useless if the crawler is told to stay out. Ask your web developer to confirm these are allowed, then re-run your audit a few weeks later to see the effect.

Repeat the audit on a schedule

Run this audit once a month, using the exact same question list every time. AI answers shift as models update and as new pages get published, so a single snapshot goes stale fast. A monthly rhythm shows you the trend, which matters far more than any one result. Keep every month in its own dated tab so you can see your named rate climb or slip over time.

Watch for two patterns. First, questions where you move from absent to named, which tells you a change worked. Second, questions where a new competitor suddenly appears, which tells you someone else is investing in the same answers. Both are early signals you would miss without a steady cadence. Treat the audit as a recurring habit, not a one off project you finish and forget.

Over a few months this manual process teaches you exactly how the assistants see your market. You learn which questions matter, which pages get quoted and what plain answering looks like in your field. That knowledge is the real output. A paid tool can speed up the counting later, but the judgement you build by reading the answers yourself is what actually moves your visibility.

Published 16 September 2026 by Carpe DM Strategies. We run this work on our own business every day, which is where the examples come from.