How to measure AI search visibility
How to measure your brand's visibility in ChatGPT, Gemini, Perplexity and Google AI Overviews: the six numbers to track and how to collect them.

Why normal SEO reports miss it
Rank trackers tell you where a page sits in Google's list of links. Search Console tells you about clicks and impressions from Google Search. Neither tells you whether ChatGPT recommended you this morning, or what Perplexity said about you.
AI answers are written fresh for each question, and they differ by engine. So AI visibility needs its own measurement: you ask the questions your buyers ask and look at the answers.
The six numbers to track
| Metric | The question it answers | How to record it |
|---|---|---|
| Mention rate | How often are we named? | Answers that name you ÷ answers collected, per engine |
| Position | When named, are we first, in the list, or a footnote? | First, top three, later, or not named |
| Share of voice | How do we compare with rivals? | Your mentions ÷ all brand mentions in the same answers |
| Citation rate | Do engines link to our site? | Answers that cite a page on your domain ÷ answers collected |
| Accuracy | Is what they say about us right? | Correct, partly wrong or wrong, with the wrong claim noted |
| AI referral traffic | Do AI answers send us visitors? | Visits from chatgpt.com, perplexity.ai, gemini.google.com and similar in your analytics |
Together they show whether you are visible, how strongly, against whom, on what evidence, how accurately, and whether it brings visitors.
How to collect the data
1. Fix your question set
Write 10 to 30 questions your buyers really ask. Mix category questions ("best [service] for [type of company]"), problem questions ("how do I fix [problem]") and brand questions ("what is [your brand]?"). Keep the list fixed so every run is comparable. Our AI visibility audit guide explains how to choose them.
2. Run them on each engine
At minimum: ChatGPT, Google AI Overviews, Google AI Mode, Gemini and Perplexity. Add Claude and Microsoft Copilot if your buyers use them. Use fresh sessions so past chats do not colour the answers, and save the full answer and its sources each time.
3. Score every answer
Record the six metrics above for each answer. A spreadsheet works for a small set; a tracking tool helps once you have dozens of questions across six engines.
4. Add the traffic view
In your analytics, filter referral traffic for AI assistants. This is the only number on the list that shows a visit, not just a mention, and it is often small at first. Watch which pages those visitors land on.
5. Repeat on a schedule
Run the same set monthly, or weekly if you are changing a lot. Answers vary from one run to the next, so judge trends over several runs rather than one result.
Reading the results
- Named rarely, rivals named often: look at the sources engines cite for those questions. Those are the sites you need to be on.
- Named, but described wrongly: find where the wrong fact comes from and correct it at the source, then on your own site.
- Named, never cited: your site may be hard for engines to read or quote. Check crawler access and add plain, answer-first pages.
- Good on one engine, absent on another: each engine leans on different sources. Treat them as separate channels.
Doing it with less effort
You can do all of this by hand. It just grows fast: 20 questions on 6 engines is 120 answers a run. Our AI Visibility Score runs your buyer questions on each engine and gives you this baseline, and every LeadAIO engagement tracks it per engine and reviews it with you every quarter.