Search rankings are public: anyone can look up where a page sits. AI answers are not. Each person sees their own answer, and you will never see the ones that recommend someone else unless you go and ask. So the first step in AI visibility is simply to find out what buyers are being told.
Why one answer is not enough
Ask an AI engine the same question twice and it can name different businesses. Wording, location, the account you are signed in to and the time of day can all change the result. If you ask once and see your name, you may feel safe. If you ask once and do not, you may panic. Neither tells you much.
What does tell you something is a pattern: the same questions, asked the same way, several times, with the results written down and dated. That is all this method does.
Six steps
- Write your buyer questions. Eight to ten is enough to start. Use the sentences customers use on the phone and on WhatsApp, with the place and the need in them: “best dental clinic near Kondapur” or “custom birthday cake in Gachibowli today”, not “dentist” or “cake shop”.
- Pick the engines and a fixed way of asking. Choose the AI products your buyers are likely to use, for example ChatGPT, Perplexity, Gemini or Google's AI Overviews. Decide how you will ask: a fresh session each time is closer to what a new buyer sees than your own long-running account.
- Ask each question more than once. Three runs per question per engine is a sensible minimum. More is better, since answers vary. Ten questions, three runs and four engines is 120 answers to record, so begin with fewer questions or engines and grow.
- Record what came back. Use the table below. It takes a few minutes per question.
- Work out your share and look for gaps. See the arithmetic in the next section.
- Fix a page, then ask again. Change the page that should answer a question you are missing, then repeat the same questions later and compare with your earlier rows.
What to write down
One row per answer. These columns are enough.
| Column | Why it is there |
|---|---|
| Date and time | Answers change, so every row needs a date. |
| Engine and how you asked | For example, a fresh session with no account. Keep it the same each run. |
| Question | Word for word, so the next run is comparable. |
| Named? | Yes or no. Count a mention only if the answer clearly refers to your business. |
| Who else was named | Shows who is winning the answers you are missing. |
| Source shown | If the engine links a page, note which one. It tells you what the answer was drawn from. |
| Facts right? | Name, location, hours, services. A wrong fact is a finding even when you are named. |
Working out your share
Count the answers you collected, then count how many of them name your business. Divide the second number by the first. Omnishown calls this Answer Share. As an illustration only: if you collected 20 answers and 8 named you, your share for that set of questions is 40%.
Treat it as a trend to follow, not a verdict. The useful comparison is your own number last month against your own number this month, using the same questions, the same way of asking, the same engines and the same number of runs per question. If those change, compare each question and engine separately, because more runs on an engine that names you often can lift the total even when nothing has improved. Comparing with a competitor's number, or with a figure from a different set of questions, tells you little.
When you are not named
Look at the questions where you are never named. For each one, ask four things:
- Does one page on your site answer this question in plain words near the top, with your location and what you offer?
- Do your name, address, phone number and hours match everywhere they appear: your site, Google Business Profile and directories?
- Can crawlers read the page? Check what your robots.txt allows.
- Is the key information real text rather than an image, and is it described with structured data such as FAQ markup? The FAQ schema generator can help.
Fixing these makes it easier for an engine to find and quote you. It does not guarantee that it will. The AI visibility guide has a longer checklist.
When doing it by hand stops working
The method above works well for a first look. It gets tiring when you track many questions across several engines, repeat each one several times, and do it every week so you can see the trend. People skip runs, change the wording without noticing, and forget to date the results.
That repetitive part is what Omnishown Watch does. It asks your chosen buyer questions on a schedule, samples each one more than once, marks each answer named or not named, dates it, and rolls the results into an Answer Share. The list of engines and the cadence are being set with the first pilot clients, so ask before you assume a particular engine is covered.
Knowing where you are not named is only half the job. Omnishown Check reviews each push to your repository for SEO, AEO and GEO issues, and Omnishown Fix sends proposed changes as pull requests that a person approves. Omnishown Diff shows what changed after each release. If you would like to try it on your own site, you can join the pilot.