How to measure AI visibility: share of answers, citations and AI traffic

AI visibility is how often AI tools such as ChatGPT, Gemini and Google AI Overviews name your brand, what they say about it and which sources they cite. You measure it with a fixed set of real customer questions, run repeatedly, plus the traffic and branded search that follow.

By Andrew Holland, Director of SEO, and the JBH team · Updated · How we wrote this

AI visibility in one paragraph

AI visibility is the share of AI answers in your category that name your brand, describe it accurately and cite your pages or coverage about you. It covers ChatGPT, Gemini, Perplexity, Copilot, Claude, Google AI Overviews and Google AI Mode. Because these tools give a different answer almost every time, there is no position one to track. You measure it by asking a fixed set of real customer questions many times, counting how often you appear against competitors (your share of answers), and checking the downstream signals: AI referral traffic in Google Analytics 4, generative AI impressions in Search Console, and branded search. It is the scoreboard for GEO (generative engine optimization, or optimisation in UK spelling) and for digital PR aimed at AI search.

AI visibility in numbers

  • There is a less than 1 in 100 chance that ChatGPT or Google’s AI gives the same list of brands twice when asked the same question 100 times. (SparkToro, January 2026)
  • 142 people asked to write a prompt with the same intent produced prompts with an average semantic similarity of just 0.081, yet the AI answers still converged on a similar set of brands. (SparkToro, January 2026)
  • People who are recommended a brand by AI are 2.5 times more likely to visit its site in the next seven days, and over half of those visits arrive through branded search. (Similarweb, June 2026)
  • ChatGPT carries 92.4% of trackable referral traffic from standalone AI assistants, while Google’s AI Overviews and AI Mode drive more AI-influenced traffic than every assistant combined. (Previsible, July 2026)
  • ChatGPT includes citations in 96% of responses, Gemini in 82% and Claude in 55%. (Muck Rack, May 2026)
  • Brands that own a ChatGPT category had higher branded search volume than the runner-up in 55.7% of pairs, the only one of three domain measures Semrush tested that made a statistically significant difference. (Semrush, July 2026)

From JBH’s work: we report share of search beside AI visibility. Spacegoods climbed to 30% share, up from 15%, according to MyTelescope search share data. (Spacegoods case study)

What to measure: mentions, share of answers, citations and sentiment

“Are we showing up in AI?” is really four questions: were we named, how often against competitors, were we cited, and what did the answer say? A brand can do well on one and badly on another, so each needs its own measure.

MeasureWhat it countsWhat it tells you
MentionAn answer names your brand in its textWhether you are in the consideration set at all
Share of answersAnswers that name you, divided by all answers for your prompt setYour visibility rate, the headline number to trend
AI share of voiceYour mentions, divided by mentions of every tracked brandYour position against named competitors
CitationAn answer links to a page as a source: your site, or coverage and reviews that mention youWhether the engines use your pages and your coverage as evidence
Sentiment and descriptionThe words used about you, and whether they are positive, neutral, negative or wrongWhat a shopper actually reads about you

Share of answers and AI share of voice are often used interchangeably, but they move differently. If AI tools start naming more brands per answer, your share of voice can fall while your share of answers holds steady. Report both. SEO vs GEO vs AEO sets these AI measures beside what success looks like in classic SEO.

Split citations in two. A citation of your own site shows your pages are being read. A citation of an article that names you shows your coverage is doing the work, and that is where digital PR shows up in AI answers. Engines differ a lot here: Muck Rack found ChatGPT cites sources in 96% of responses, averaging about five, while Claude cites in only 55% but averages 13 when it does. (Muck Rack)

Note prominence too. Is the brand recommended as the answer, listed among several, or mentioned in passing? That matters more than where in the list it appears.

Skip “AI rankings”. SparkToro found AI tools almost never return the same brands in the same order: less than 1 in 1,000 runs. Its conclusion was that visibility percentage across many prompts is a reasonable metric, and “any tool that gives a ‘ranking position in AI’ is full of baloney.” (SparkToro)

Build a prompt set from real customer questions

Your measurement is only as good as the questions you ask. A prompt set invented by a marketing team tends to be flattering: it mirrors the brand’s own language and the products it wants to sell. Start from what customers actually ask.

Where to find the questions

  • Search Console: filter the Performance report for queries that start with how, what, which, best or vs.
  • Customer service and sales: the questions people ask before they buy, in their own words.
  • Reviews and Reddit: the comparisons, worries and “is it worth it” questions in your category.
  • Google’s People Also Ask boxes for your main category terms.

How to shape the set

  • Mostly unbranded. Questions such as “best dog food subscription UK” show whether you are recommended to people who do not yet know you. Keep a smaller group of branded questions (“is [brand] any good”) to check how you are described.
  • Cover each stage. Category (“best…”), problem (“how do I stop…”), comparison (“[brand] vs [brand]”), and price or trust (“is it worth it”).
  • Write them the way people type into AI. Longer and more specific than a search query, with context such as budget or situation.
  • Fix the set. Once agreed, change it only quarterly, and log every change. Otherwise changes to the prompts will skew the trend.

Twenty to fifty prompts is a sensible start for one category. You do not need to guess every possible wording. When SparkToro asked 142 people to write their own prompt for the same need, the prompts had almost nothing in common, yet the AI answers still drew on a similar set of brands. (SparkToro) Covering each intent matters more than matching exact phrasing.

For competitors, use the three or four brands you lose sales to, then add any brand the AI tools keep naming that you had not considered.

Why answers change from run to run, and how to sample

Ask ChatGPT the same question twice and you will often get two different shortlists. The tools generate a new answer each time and search the web live for many questions. SparkToro and Gumshoe had 600 volunteers run twelve prompts through ChatGPT, Claude and Google’s AI 2,961 times. The chance of the same list of brands appearing twice in 100 runs was below 1 in 100. (SparkToro)

The same study found a pattern under the noise. Across many runs, the leading brands for a question appeared in a consistent share of answers: one agency was named in 85 of 95 Google AI responses to the same prompt. (SparkToro) One screenshot of ChatGPT proves very little; you need a hundred answers to see a pattern.

A practical sampling plan

  • Run every prompt several times in each engine each month, not once.
  • Keep the conditions the same: UK location, the same model where you can choose, and a clean session with no history.
  • Record the date, engine and model with every answer, so a model update does not look like a campaign result.
  • Report each engine separately as well as combined. ChatGPT and Google’s AI surfaces build answers differently.
  • Judge trends over three months, not month on month.

Sample size sets how much a change can mean. As a rough guide, if 100 answers name you 40 times, your true rate could sit anywhere from about 30% to 50%. With 400 answers, the range narrows to about 35% to 45%. A move of a few points on a small sample is noise, so say so in the report.

Treat tool vendors with the same care. Google’s own guidance warns: “No third-party tool has access to our internal ranking or AI systems.” (Google Search Central) A good tool shows you the prompts, the raw answers and how often each prompt was run. If it only gives you a score, ask how the score is built. Different prompt sets give different numbers for the same brand, so pick one method and keep it.

Track AI referral traffic in GA4

Since 13 May 2026, Google Analytics 4 has had an AI Assistant channel in its default channel group. (Google Analytics, What’s new) When a visit’s referrer matches Google’s list of AI assistants, GA4 sets the medium to “ai-assistant” and groups the session under that channel. Google’s examples include ChatGPT, Gemini, Claude and Copilot. (Google Analytics Help; What’s new)

What it does not include

  • Google AI Overviews and AI Mode. Google states the AI Assistant channel excludes them. Clicks from those features are counted in Organic Search, and GA4 does not report them separately. (Google Analytics Help)
  • Visits without a referrer. A click that arrives with no referrer information, which can happen from some apps, is counted as Direct.
  • Influence without a click. Much of the effect of an AI recommendation never shows as a referral. Similarweb found people recommended a brand by AI were 2.5 times more likely to visit within seven days, and over half arrived through branded search. (Similarweb)

How to set it up

  • Open the Traffic acquisition report and look at the AI Assistant row under session default channel group. Add session source as a secondary dimension to see each assistant.
  • Build an exploration with AI Assistant sessions by landing page. This shows which pages the assistants send people to, which is often a guide or comparison page rather than the homepage.
  • Compare engagement rate, key events and revenue for AI Assistant against Organic Search. The volume may be small; the value per visit is the point.
  • Check Referral for any AI source GA4 has not yet picked up, and add it to a custom channel group if you need it.

Treat AI referral traffic as a minimum. It only shows the AI influence that left a trace.

What Search Console shows for AI Overviews and AI Mode

Google launched a Generative AI performance report in Search Console on 3 June 2026 and says it was rolled out to all websites worldwide by 31 August 2026. (Google Search Central) It covers AI Overviews and AI Mode on Google Search, with a separate report for Discover.

What it shows: impressions, meaning how many times links to your site appeared in those AI features, broken down by page, country, device and date. You can filter by text or multimodal search, such as Google Lens. (Search Console Help)

What it does not show, at the time of writing:

  • Clicks or click-through rate for AI features. Google’s help page describes impressions only.
  • The queries that triggered the AI answer.
  • Mentions of your brand without a link to your site, or coverage about you on other sites.
  • Any other AI tool. ChatGPT, Gemini and Perplexity are outside Search Console entirely.

The same AI feature data is also included in the standard Performance report, under the Web search type. (Google Search Central) Use the new report to see which pages Google’s AI features link to, and whether that number is growing. Use your prompt tracking for everything it cannot see.

If AI tools are recommending you, more people should search for you by name. Branded search is the most reliable signal that AI visibility is turning into demand, because it captures the people who read an answer and came looking later.

Several studies point the same way. Ahrefs found branded search volume correlates with AI Overview brand visibility at 0.392, behind web mentions at 0.664. (Ahrefs) Semrush found branded search volume was the only statistically significant difference between brands that own a ChatGPT category and their runners-up. (Semrush)

Two ways to measure it

  • Branded queries in Search Console. The branded queries filter in the Performance report splits clicks and impressions into branded and non-branded queries. Google made it available to all eligible sites on 11 March 2026. (Google Search Central)
  • Share of search. Searches for your brand divided by searches for all brands in your category. Les Binet’s work for the IPA found it correlates with market share and tends to lead it, by up to a year for cars. It can be pulled free from Google Trends. (IPA, 2020)

Spacegoods shows what this looks like. After 92 placements, its share of search rose from 15% to 30% between March and August 2025, according to MyTelescope data. (Case study)

Branded search moves for many reasons, including TV, social and seasonality. Report it alongside share of answers, and note anything else that ran that month, rather than crediting AI alone.

A simple monthly AI visibility report

One page, the same every month. Show each measure for this month, last month and the three-month average, with a line of commentary on what changed and why.

MeasureHow to calculate itWhere the data comes from
Share of answersAnswers naming you ÷ all answers, by engine and combinedPrompt tracking
AI share of voiceYour mentions ÷ mentions of all tracked brandsPrompt tracking
Citations of your siteAnswers citing a page on your domain, and which pagesPrompt tracking
Earned citationsAnswers citing coverage or reviews that name youPrompt tracking, matched to your coverage log
Description and sentimentPositive, neutral and negative mentions; errors and out-of-date claims listedA person reading the answers
Source gapsThe most-cited sites in your category where you are absentPrompt tracking
Google AI impressionsImpressions and top pages in AI Overviews and AI ModeSearch Console, Generative AI performance report
AI Assistant trafficSessions, engagement rate, key events and revenueGA4, AI Assistant channel
Branded searchBranded clicks and impressionsSearch Console, branded queries filter
Share of searchYour brand searches ÷ all brand searches in the categoryGoogle Trends or a keyword tool

The source gaps row is the one that turns a report into a plan. It lists the publications, review sites and threads the engines cite when they answer your customers’ questions. For Clifton Private Finance, JBH mapped 21 sites that AI cites for bridging finance, and five key queries now surface Clifton. (Case study) Closing those gaps is the job of digital PR.

Report the results of the work as well. Nature’s Best’s AI citations rose from 784 to 1,307 in six months alongside 83 placements, 56 of them linking to category pages. (Case study)

Keep the old measures. Report AI visibility next to links, rankings, organic traffic and revenue. Google says optimising for its generative AI features is “still SEO”. (Google Search Central)

How JBH’s free AI visibility check works

Our free AI visibility check is a baseline built on the method above. You tell us your website and what you sell.

  • 10 real customer questions, from “which is best” to “is it worth it”, each asked in ChatGPT.
  • An AI visibility score out of 100, built from how often ChatGPT names you, so you can track it month by month.
  • How you are described, word for word, and the sources behind the answers, including anything out of date.
  • Three moves we would make first, written by a strategist.

A strategist reads every check before it is sent, within two working days, and you can book a 15-minute walkthrough if you want one. We only follow up if you ask. A single check is a snapshot. Re-run the same questions monthly and the trend is what tells you whether your coverage, content and SEO are working.

The AI visibility audit goes further: 100 customer questions in ChatGPT, Gemini, Perplexity and Google AI Overviews, each asked on three different days.

What building the tool taught us is in What I learned building an AI visibility checker. If you want the tracking run for you, see our GEO service.

Sources:

Questions people ask about measuring AI visibility

What is AI visibility?
AI visibility is how often AI tools such as ChatGPT, Gemini, Perplexity and Google AI Overviews name your brand when people ask questions in your category, how they describe you and which sources they cite. It is measured as a share of answers across a fixed set of real customer questions, run repeatedly, because the answers change every time.
What is AI share of voice?
AI share of voice is your brand’s mentions in AI answers divided by the mentions of every brand you track in your category. It shows your position against competitors. Report it alongside share of answers, the percentage of answers that name you, because the two can move in different directions.
Can GA4 track traffic from ChatGPT?
Yes. Since May 2026, GA4 has an AI Assistant channel that groups visits from assistants such as ChatGPT, Gemini, Claude and Copilot. It excludes Google AI Overviews and AI Mode, which count as Organic Search, and it misses visits that arrive without a referrer and people who search for you later.
Does Search Console show AI Overviews data?
Yes. The Generative AI performance report shows impressions for links to your site in AI Overviews and AI Mode, by page, country, device and date. Google’s help page describes impressions only, with no query data, and it does not cover ChatGPT, Gemini or Perplexity.
How can I check my AI visibility for free?
Ask ChatGPT, Gemini, Perplexity and Google the questions your customers ask, and note who is named and which sources are cited. JBH’s free AI visibility check does this for 10 real customer questions in ChatGPT, and a strategist reads every report.
How many prompts should I track?
Twenty to fifty real customer questions is a sensible start for one category, grouped by intent. Run each one several times in each engine every month, because a single answer is close to random. The number of answers you collect matters as much as the number of prompts.
Which AI search visibility tool should I use?
Pick one that shows you the prompts, the raw answers and how many times each prompt was run, and stick with it so your trend is comparable. Google warns that no third-party tool has access to its internal ranking or AI systems, so be wary of any tool that promises AI rankings.
How often should I measure AI visibility?
Monthly, with the same questions each time. Judge the trend over three months, because answers move from run to run and AI models are updated often. Review the prompt set quarterly.

Let’s make your brand the one AI recommends

Tell us where you are now and where you want to be. We’ll reply within one working day.