By Mehdi HassanHow do you measure AI search visibility?
You measure AI search visibility by asking AI tools the same questions your buyers ask, on a fixed schedule, and recording whether and how your brand shows up. It's closer to rank tracking than to analytics: you sample the answers, because the engines don't report impressions to you.
The reason this needs its own method is simple. GA4 only sees visits. An AI answer that recommends you, names a competitor instead, or describes your pricing wrongly leaves no trace in your analytics unless someone clicks. Even then, many app visits arrive as Direct, as our guide to tracking AI traffic in GA4 explains.
This article is part of our hub on marketing website analytics. It defines a small set of metrics, gives a manual tracking sheet you can run in a couple of hours a month, and explains when a paid tool starts to earn its cost.
Which AI visibility metrics actually matter?
Four AI visibility metrics matter for most B2B companies: mention rate, citation rate, share of voice and accuracy. Each answers a different question, and together they tell you whether AI tools know you, trust your pages, prefer you over competitors and describe you correctly.
Metric | Definition | Question it answers |
|---|---|---|
Mention rate | Share of prompts where your brand is named in the answer | Do AI tools know we exist for this topic? |
Citation rate | Share of prompts where one of your URLs is linked as a source | Do they trust our pages as evidence? |
Share of voice | Your mentions divided by all brand mentions across tracked competitors | Are we winning against the alternatives? |
Position | Whether you're named first, in the middle or last in a list | How prominent are we when we appear? |
Accuracy | Share of mentions where pricing, features and category are correct | Is what they say about us true? |
Position is optional, and useful mainly for "best tools" style prompts where order matters. Accuracy is the metric teams skip and later regret. An AI answer that says your product lacks SSO, or quotes a plan you retired, can cost deals quietly. For a fuller definition of the competitive metric, see what AI share of voice means.
How do you build a prompt set?
Build a prompt set from the questions real buyers ask at each stage, not from your keyword list. Twenty to thirty prompts is enough to see a pattern without making the monthly run a chore. Write them the way a buyer would type them into a chatbot: longer and more specific than a Google search.
- Category prompts: "What are the best tools for [job] for a 50-person B2B company?"
- Problem prompts: "How do I fix [problem] without hiring [role]?"
- Comparison prompts: "[Your brand] vs [competitor], which is better for [use case]?"
- Brand prompts: "What does [your brand] cost?" and "Is [your brand] secure?"
- Use-case prompts: "Which [category] tools work for [industry or team]?"
Pull wording from sales call notes, support tickets and the answers to your "How did you hear about us?" field. Avoid prompts that name your brand except in the brand and comparison groups, or you'll measure recall of a name you supplied. Keep the set fixed for at least two quarters; changing prompts every month destroys the trend.
What does a manual prompt-tracking sheet look like?
A manual prompt-tracking sheet has one row per prompt per engine per month, and a handful of columns you can fill in by reading the answer. It's slow the first time and fast after that, and it forces you to read what AI tools actually say about your category.
Column | What to record | Example entry |
|---|---|---|
Date and engine | When you ran it and in which tool | 2026-10-01, ChatGPT |
Prompt | The exact wording, copied from your fixed set | Best CRM for a 20-person agency |
Mentioned? | Yes or no, and position if it's a list | Yes, 3rd of 6 |
Cited URL | Which of your pages was linked, if any | /pricing |
Competitors named | Every other brand in the answer | Brand A, Brand B, Brand C |
Accurate? | Yes, no, or partly, with a note | Partly: old pricing tier |
Sources cited | The third-party domains the answer drew on | Review site, Reddit thread |
Use clean sessions
Run prompts logged out or in a fresh account with memory and personalization off, so earlier chats don't shape the answer.
Run each prompt once per engine
Cover ChatGPT, Perplexity, Gemini and a Google search that shows an AI Overview. Add Claude or Copilot if your buyers use them.
Record, don't interpret
Fill in the columns as you read. Save a screenshot or the shared link for anything surprising.
Calculate the rates
Mention rate is prompts with a mention divided by total prompts, per engine. Do the same for citations. Share of voice is your mentions over all brand mentions.
Note the sources
List which third-party sites keep appearing as sources. Those are the places where a mention or a correction would move the numbers.
Hypothetical example: reading one month of results
For example, suppose a startup runs 25 prompts in four engines, so 100 answers. It's mentioned in 30 of them, a 30% mention rate, but cited in only 6. Three competitors share the remaining mentions, giving it roughly a quarter of the share of voice.
The pattern says AI tools know the brand from elsewhere but don't use its own pages as evidence. The sources column shows the same two review sites and one comparison article appearing again and again. The useful actions are clear: add direct, quotable answers to the pricing and comparison pages, and get the brand described accurately on those recurring third-party sources. These numbers are illustrative, not from a real company.
Why do results differ so much between engines?
Results differ between engines because each one retrieves and cites sources differently. The Semrush AI Visibility Index, built on 126 million prompts, found ChatGPT cites about 15 sources per answer against about 3 for Gemini. More citation slots give smaller brands more chances to appear, so a mention rate that looks strong in ChatGPT and weak in Gemini can both be accurate.
Sources also shift over time. Reddit's share of ChatGPT citations reportedly fell sharply in August 2026 while staying stronger in Google, which shows how quickly one engine's habits can change. That's why you report each engine separately and avoid blending them into a single score. For whether community threads are worth your effort, read our view on Reddit and AI search.
Answers also vary between runs of the same prompt. One sample per month is noisy for a single prompt but stable enough across 25 or more. If a specific prompt matters a lot, run it three times and record the majority result.
What moves AI visibility numbers?
AI visibility numbers move mainly with what other sites say about you and with how quotable your own pages are. Off-site mentions have the strongest evidence: in Ahrefs' study of 75,000 brands, web mentions correlated 0.664 with AI Overview visibility, against 0.218 for backlinks and 0.392 for branded search.
- Mentions on the sources AI tools already use. Your sheet's sources column tells you where to start.
- Clear answers on money pages. Pricing, comparison and security pages with direct, self-contained answers are easier to cite. Our guide to structuring content so AI engines quote it shows the pattern.
- Visible FAQs. Real buyer questions answered on the page, as covered in FAQ sections that get cited.
- Crawlable, server-rendered pages. If AI crawlers can't read the content, nothing else matters. Run an AI visibility website audit to check.
When is a paid AI visibility tool worth it?
A paid AI visibility tool is worth it once your manual sheet has shown which prompts, engines and competitors matter and you need more frequent or broader tracking than a person can run. Tools sample more often, cover more prompts and markets, and keep history automatically.
Pros
- Daily or weekly sampling smooths out answer-to-answer noise
- Covers hundreds of prompts and several markets or languages
- Tracks competitors and source domains automatically
Cons
- Monthly cost that's hard to justify before you know what to track
- Each tool samples differently, so numbers don't match across vendors
- Dashboards can replace reading the answers, which is where the insight is
Our recommendation for most startups: run the manual sheet for two or three months first. You'll learn more from reading 100 answers yourself than from a score, and you'll buy the right tool when you do. Report the results monthly alongside your Search Console and pipeline numbers, as in our SEO reporting template.
