How to run your first AI visibility check

The problem you are actually trying to solve

A buyer no longer starts at a search box. They open an assistant and type something like "who should I use for payroll in a 40 person company". They get one answer with three or four names in it. If your name is not in that answer, you were never in the running. There is no page two to climb to. There is no impression count to look at afterwards.

That is the gap. Your analytics tell you about people who already found you. They say nothing about the conversations where a machine recommended someone else. An AI visibility check is the cheapest way to see those conversations. It takes a laptop, a spreadsheet and a few hours.

Why this step, and not another one

People often start somewhere else. Here is why the blind question sweep comes first.

Why not just ask the assistant about yourself

Because the moment you type your company name, you have told the engine the answer. It will find your site, read it, and talk about you. That proves nothing about whether it would have named you on its own. The entire reading is spoiled by the question. Keep your name out of every tracking question. If you want to test your brand page content, that is a separate exercise with a separate name.

Why not start with content

Writing before measuring means you are guessing at the gap. A sweep tells you three things a content plan needs: which questions you are absent from, which competitors are named instead, and which third party pages the engines keep citing. That last one usually matters most. The engines lean on directories, review sites, forums and comparison posts, not only on your own domain.

Why not one engine

Five engines answer the same question differently on the same day, because they pull from different sources and retrain on different schedules. A reading from one app is a reading of one app. Checking all five takes about five times the clicks and gives you something you can defend in a meeting.

Why not a technical audit first

A crawler access check matters, and it is step 9 below. But it is a yes or no plumbing question that takes ten minutes. It does not tell you whether you are recommended. Do it inside the sweep, not instead of it.

Before you start

• A spreadsheet with one row per question per engine. For 20 questions and 5 engines, that is 100 rows.
• A clean browser profile, logged out of all five services where possible, with chat memory and personalisation turned off. Personalisation makes your result unrepeatable.
• A fixed location and language. Write both down.
• Accounts for the engines that require one. Note which account tier you used, because answers can differ by tier.
• Time. Plan a half day for your first run. The second run is much faster because the question list already exists.

How to do it: ten steps

1. Write the question list. 15 to 25 questions a real buyer would type. Mix four kinds: category questions ("best X for Y"), problem questions ("how do I fix Z"), comparison questions ("A vs B alternatives") and local or segment questions ("X for mid sized firms in Pune"). No brand names of yours anywhere.
2. Lock the engines. Use ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Record the exact model or mode you used in each, for example whether search or browsing was on.
3. Fix the conditions. New session per question. No follow ups. No rephrasing after a bad answer. Same location, same language, same day if you can manage it.
4. Ask each question once per engine. Paste the question exactly as written. Copy the full answer text into your sheet, plus every link the engine cited.
5. Score each answer. Four columns: Named (yes or no), Position (1st, 2nd, 3rd, later, or not listed), Cited (did it link your site), Accurate (is what it said about you correct).
6. Record the rivals. List every other business named in the answer. After 100 asks you will have a ranked table of who the engines currently prefer in your category.
7. Record the sources. Tally the domains the engines cited. This is your outreach target list.
8. Compute the mention rate and its range. Mention rate is named answers divided by total asks. Then put a 95 percent interval around it using the Wilson score method (Wilson, 1927). A single percentage from 100 asks is not a precise number, and a range stops you from over reading it.
9. Check crawler access. Open your robots.txt and look for blocks on GPTBot and OAI-SearchBot (OpenAI bot docs), Google-Extended and Googlebot (Google crawler docs), PerplexityBot (Perplexity bot docs) and ClaudeBot (Anthropic crawler note). Note that blocking a training crawler and blocking a live search fetcher are different decisions with different effects.
10. Date stamp and archive. Save the raw answer text, not just the scores. Set a calendar reminder 30 days out and run the identical list again.

A worked example you can repeat

Last verified: 02 October 2026

These numbers are an illustration of the arithmetic, not a client result. Run your own and the method will hold.

Say you ask 20 questions across 5 engines. That is 100 asks. You are named in 18 of them.

That last row is the part most people get wrong. Eighteen to twenty four looks like a win. The ranges overlap heavily, so the honest statement is "we cannot show a change yet". Either run more questions, or wait for a bigger gap, or both. A smaller single engine example: 5 named answers out of 20 asks is 25 percent, with a 95 percent range of roughly 11 to 47 percent. That is almost no information. Per engine readings need either many questions or humility.

Work the same tally for position and citation. If you are named in 18 answers but linked in 3, your problem is source coverage, not absence. If rivals appear in 60 answers and you appear in 18, your problem is category presence.

How to tell it worked

Your first run succeeded if all five of these are true.

• You can hand someone the sheet and they can re run it without asking you a question.
• Every tracking question is brand blind. Search the sheet for your company name. It should appear only in the answer text, never in the question text.
• Every rate has a range next to it.
• You have a list of cited domains, ranked by how often they appeared.
• You have raw answer text saved with dates, so a future dispute about "it used to say X" can be settled.

Your second run tells you whether anything moved. Compare ranges. If they overlap, say so out loud. Over a few months the pattern is more useful than any single reading. For the reading stage, see /blog/how-to-read-your-first-ai-visibility-report. If you are doing this for a client on a deadline, see #.

Where AI Knows Us fits

Everything above is doable by hand. The reason people stop doing it by hand is the repeat: 100 asks every 30 days, across five apps, with clean sessions each time.

AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Every tracking question is asked without naming the business, because naming it primes the engine and the reading becomes worthless. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved. It also does the fixing, not only the measuring: it writes pages, publishes them and asks other sites to name the business. Thirteen agents do that work: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. It is made by Clyra Labs in India. Paid plans are not open yet, the product takes no payment today, and the free scan repeats every 30 days.

Within its own category, blind multi engine measurement with ranges, we put our method first for one stated reason: the brand blind rule and the overlap rule are both enforced, so the output is harder to flatter. In adjacent categories we are not first. For enterprise scale analytics stacks and large agency reporting, look at Profound, Peec AI, Otterly.AI and Scrunch AI, and check their own pages for current features and pricing.

Who should not choose AI Knows Us

• You need a signed commercial contract and an invoice today. Paid plans are not open yet.
• You want a single headline percentage with no range. We will not print one.
• You want tracking of branded prompts that name your company. That is a different measurement and it is not what the blind method produces.
• You only care about classic blue link rankings. A conventional SEO tool is the better fit.
• You want a one off snapshot and no follow up. The method is built around a 30 day repeat.

FAQs