How to measure AI visibility: choosing questions, how many runs, and reading the numbers
You measure AI visibility by choosing the questions your customers would actually ask, putting each to an AI several times, and counting the share of answers that name your business or cite your site as a source. Three things matter most: keep a fixed set of questions that do not contain your name, ask them repeatedly under the same conditions, and read the trend across runs rather than any single answer.
Written October 31, 2026. What OpenAI, Perplexity and Google say was checked in their own documentation that day. It can change, so check the current pages too.
AI answers change between days and between runs of the same question. The numbers this method produces are a sample, not a view of every answer. And nothing you do can guarantee that an AI will mention you or rank you anywhere.
What "AI visibility" means: four numbers to track
Different tools use "AI visibility" for slightly different numbers. If you measure it yourself, keeping these four apart makes your results much easier to compare later.
| Number | How to count it | What it tells you |
|---|---|---|
| Mention rate | Answers whose text names your business (including common alternative names) ÷ answers received | Whether the AI suggests you as an option |
| Citation rate | Answers whose sources include your domain ÷ answers received | Whether the AI uses your site as a source |
| Appearance rate | Answers with either a mention or a citation ÷ answers received | Whether you show up in the answer at all |
| Citation share against competitors | Citations of your domain ÷ (citations of your domain + citations of the competitor domains you chose) | How often you are chosen as a source compared with named competitors |
An AI may name you while citing a review site or a map service instead of your own site, and the reverse happens too. Record the two separately.
How to measure it, in six steps
- Pick the AI platforms: start with the one or two your customers are most likely to use. Record each platform separately and never pool them. Each one uses its own search system and its own model, so a result on one does not carry over to another.
- Write the questions: 5 to 20 "discovery" questions that do not contain your name, such as "Can you recommend a dentist in Umeda?" Questions that name you are kept out of the rate and used to check whether the AI states your facts correctly (next section).
- Decide how many runs: answers change each time, so ask each question two or three times. The fewer runs, the more a lucky or unlucky answer moves the number.
- Keep the conditions the same: the same wording, the same platform and mode, signed out, every time. Consumer apps can change their answers with sign-in, location and chat history.
- Record each answer: date and time, platform, question, whether you were named, whether your domain was among the sources, who appeared instead, and an excerpt (see the sheet below).
- Repeat monthly with the same set: compare trends only between runs with the same questions, the same platform and the same conditions. If you change the questions, start a new series.
Choosing questions: keep discovery and named questions apart
| Type | Example (a dental clinic in Umeda, Osaka) | Use |
|---|---|---|
| Discovery | Can you recommend a dentist in Umeda? | The core of the rate: someone who does not know your name asks an AI |
| Discovery (specific need) | I'm looking for a dentist in Umeda who is careful about pain. Where should I go? | Shows whether you appear for the strengths you actually have |
| Named | What are the address and opening hours of "XYZ Dental"? | Not part of the rate. Checks whether facts such as address and hours are correct and whether your site is cited |
Named questions return your name almost every time. Mixing them into the rate only makes it look higher, so keep them separate. Write discovery questions in the words your customers use, and leave out questions about services you do not offer.
An example recording sheet
One row per answer in a spreadsheet makes the counting easy later.
| Column | Example |
|---|---|
| Date, platform, mode | 2026-10-31 10:00 / Perplexity / standard search |
| Question ID, wording, run number | d1 / Can you recommend a dentist in Umeda? / run 2 |
| Named | Yes / no |
| Own site cited | Yes / no (was your domain among the sources?) |
| Who appeared instead | Competitor A, review site B, etc. |
| Excerpt and check notes | The part that mentions you. If a fact is wrong, note the correct fact and where it is published |
Do not mark an answer as correct automatically: a person should check it against your own facts. If you find an error, check that the fact in question (hours, address, services) is stated clearly on your site. See also how businesses appear in ChatGPT.
Five things to keep in mind when reading the numbers
- Do not act on a single run. Look at the trend under the same conditions. A rate based on two or three answers means very little.
- The order of sources is not a ranking. A source list comes back in the order the platform returns it; it is not a search ranking or a "best first" order.
- The numbers cannot tell you why you are missing. All they show is that the answers you observed did not cite you.
- Site improvements and appearance do not always move together quickly. When Nurevo studied 62 local businesses, its on-site AEO score showed no relationship with how often Perplexity named or cited them (ρ = 0.03). A study that size can only detect fairly strong relationships, so this does not prove there is none, but do not assume that fixing your site alone will raise the rate. Read the study
- Read each platform on its own. Appearing on Perplexity does not mean you appear on ChatGPT or in Google's AI features.
Other ways to measure, and what each can and cannot show
| Method | What it shows | What it cannot show |
|---|---|---|
| Web analytics (referrers) | Visits that came from AI answers. OpenAI says ChatGPT automatically adds utm_source=chatgpt.com to links from ChatGPT search results (OpenAI: publishers and developers FAQ) | Appearances nobody clicked, such as a mention of your name in the answer |
| Google Search Console | Sites shown in Google's AI features (AI Overviews, AI Mode) are counted in the Performance report under the "Web" search type (Google: AI features and your website, version updated December 31, 2025). Google also announced generative AI performance reports in Search Console in June 2026 (Search Central blog) | How you appear on AI platforms other than Google's, such as ChatGPT or Perplexity |
| Server access logs | Whether AI search crawlers read your pages. OpenAI documents OAI-SearchBot (for search) and ChatGPT-User (visits a page when a user asks) (OpenAI: overview of crawlers); Perplexity documents PerplexityBot and Perplexity-User (Perplexity: crawlers) | Being read is not the same as appearing in an answer. It gives no appearance rate |
| Monitoring tools | Send a fixed set of questions to AI platforms on a schedule and record mentions, citations, competitors and trends automatically | Platforms and question counts differ by tool and plan. AEO tools compared |
For reference: Nurevo AEO has a free AI check for connected Free and Standard sites. Once a month it asks Perplexity three questions built from your business type and area (never your name) and shows whether you were named, whether your site was cited and who appeared instead. Ongoing measurement of a fixed question set under the same conditions, with competitor comparison and history, is Pro (from $99/month, beta) and currently runs on Perplexity.
FAQ
How many questions and runs are enough?
There is no single right number, but 5 to 20 discovery questions asked two or three times each is a practical start. The fewer questions and runs, the more chance moves the result. What matters most is measuring the same questions under the same conditions every month.
Can't I just ask ChatGPT myself?
It is a useful first look. But answers can change with sign-in, location and chat history, and one answer carries a lot of chance. Record what you see in a sheet and repeat it under the same conditions, and you get numbers you can compare.
If the AI names me when asked about me by name, am I fine?
Named questions return your name almost every time, so they say little about visibility. Use them to check that facts such as your address and hours are right and that your site is cited, and measure visibility with questions that do not contain your name.
My rate is low. What should I do first?
First check that AI can read your site: structured data, llms.txt, and that robots.txt is not blocking AI search crawlers. Then look at what the businesses and sites that appear instead say about themselves; it often shows which facts you are missing. Fixing your site does not guarantee the rate will rise quickly.
Will measuring make AI show my business?
No. Measuring tells you where you stand; it does not change the answers. AI answers change daily, and no method can guarantee a mention or a position.