Skip to main content

AI Search

How to Measure AI Visibility: A Practical 2026 Method

A practical method for measuring AI visibility across ChatGPT, Google AI Overviews, Perplexity and Gemini: fixed prompt sets, Search Console's generative AI report and GA4's AI Assistant channel, with a real dental clinic example.

  • 10 November 2026
  • 13 min read
  • By Kaan TURK
Kaan TURK, author
Key Takeaways
  • AI visibility is whether AI tools name, describe, recommend and link to your business when people ask the questions you should be the answer to
  • Measure it in three layers: answers (a fixed prompt set per platform), impressions (Search Console's generative AI performance report) and visits (GA4's AI Assistant channel and referral sources)
  • Record the sources every AI answer cites, because a missing third-party source can be the fastest fix available
  • Clicks from Google AI Overviews and AI Mode are counted in Search Console's Web totals and in GA4's Organic Search channel, not in the AI Assistant channel
  • In my dental clinic case study, ChatGPT drove 89.66% of AI referral sessions, and those visitors engaged at 81%

AI visibility is how often, and how accurately, AI tools such as ChatGPT, Google AI Overviews, Perplexity, Gemini and Microsoft Copilot name, describe and link to your business when people ask questions you should be the answer to. It is harder to measure than a Google ranking, because there is no single position to check. After 15 years of Search Engine Optimisation and 250+ projects, I measure it the way I measure anything noisy: a fixed sample, the platform data that does exist, and an honest view of what each number can and cannot tell you. This guide sets out that method, step by step.

What is AI visibility?

AI visibility is the degree to which AI tools name, describe, recommend and link to your business in their answers. It covers ChatGPT, Google AI Overviews and AI Mode, Perplexity, Gemini, Microsoft Copilot and Claude. Unlike a Google ranking, it is a pattern across many prompts and platforms, so you measure it with repeatable samples rather than single positions.

I find it useful to split visibility into four levels, because each one is worth more than the last:

  1. Named. Your business appears somewhere in the answer, often in a list.
  2. Described accurately. The answer gets your services, location and specialty right.
  3. Recommended. The answer suggests you for a specific need, such as "a dentist in Blacktown who sees nervous patients".
  4. Cited. The answer links to your website, so a reader can click through.

The work of improving these levels has several names. Generative engine optimisation (GEO) and answer engine optimisation (AEO) both describe it, and in practice it overlaps heavily with good SEO. The measurement, though, is its own discipline, and most businesses skip it. They either guess from a few searches on their own phone or buy a dashboard score they cannot interpret.

Why can't you measure AI visibility like Google rankings?

AI answers change between sessions, users, locations and model versions, and most platforms share no query data with site owners. There is no equivalent of a ranking position. So measurement combines a repeatable sample of prompts with the platform data you do get: Search Console impressions for Google's AI features and referral visits in GA4.

Four things make AI answers unstable:

  • Wording. "Best dentist Blacktown" and "I'm nervous about the dentist, who's good near Blacktown?" can produce different lists.
  • Location and personalisation. Results for a user in Parramatta can differ from those for a user in Melbourne, and chat memory or past conversations can colour the answer.
  • Retrieval. When a tool searches the web before answering, the sources it pulls can change from one run to the next.
  • Model updates. A new model version can reshuffle which sources a platform trusts.

None of that means measurement is pointless. It means a single screenshot proves nothing, and a number from any tool is an estimate built on a sample. Across the Australian businesses I work with, the useful question is never "are we visible?" but "for which questions, on which platforms, and is that improving?". Three layers answer it: answers from a fixed prompt set, impressions in Search Console, and visits in GA4.

How do you build a prompt set for each AI platform?

Write 20 to 40 prompts that real customers would ask, grouped by stage: discovery, comparison, evaluation and brand. Use their wording, include a location where it matters, and keep the set fixed so results stay comparable. Run the same set on each platform your customers use, starting with ChatGPT, Google AI Overviews and Perplexity.

A prompt set is your sample. If it changes every time, you are measuring your own creativity, not your visibility. Here is how I build one.

Start with real language. The best sources are your own: questions from sales calls and enquiry forms, long question-style queries in Google Search Console, People Also Ask boxes for your main keywords, and discussions in forums or Reddit threads in your industry.

Group prompts by stage. For a dental clinic in Western Sydney, a set might look like this:

Stage Example prompt
Discovery "Who is a good dentist in Blacktown?"
Need-specific "Which Blacktown dentist is good with anxious patients?"
Comparison "Is [clinic] or [competitor] better for implants?"
Evaluation "What do reviews say about [clinic] in Blacktown?"
Brand facts "What are [clinic]'s opening hours and services?"

For a B2B service, replace the location with the buyer's situation: "Which SEO consultant in Melbourne understands eCommerce migrations?" works the same way.

Keep the set between 20 and 40 prompts. With fewer than 20, one odd answer swings the whole result. With many more than 40, manual tracking stops being done consistently, and an inconsistent sample is worse than a small one.

Choose platforms from evidence. Start with ChatGPT, Google AI Overviews and Perplexity, then add Gemini, Copilot or Claude if your GA4 data shows visits from them. There is little value in tracking a platform your customers do not use.

Control the conditions. Use a clean session with chat memory off where the platform allows it, note the date and the location you ran from, and run each prompt the same way every time. I also run each discovery prompt twice in separate sessions. If the two answers disagree, that disagreement is itself a finding: your position is unstable.

What should you record for every AI answer?

For each prompt and platform, record the date, whether your business is named, its position in the answer, whether the description is accurate, whether your site is linked, which competitors appear and which sources are cited. The source column matters most, because it shows where the AI tool is getting its information about your market.

A simple spreadsheet does the job. These are the columns I use:

Column What to record
Date, platform, prompt The fixed identifiers, so results can be compared over time
Named Yes or no
Position First, in the list, or last
Accuracy Correct, partly wrong, or wrong, with a note on what was wrong
Linked Whether the answer links to your site, and to which page
Competitors Every other business named
Sources cited Every source the answer shows or links to, including directories and review sites

From that sheet, four numbers are enough for a small business:

  • Mention rate: prompts where you are named, divided by prompts run.
  • Recommendation rate: prompts where you are suggested for the specific need.
  • Citation rate: answers that link to your site.
  • Accuracy issues: a count of wrong facts, because a confident wrong answer about your hours or services costs you customers.

What counts as progress? A rising mention rate on the same prompts across at least two consecutive runs, fewer accuracy issues, and a shorter source gap list. A single run that improves is noise until the next run confirms it. A drop is the same: check whether the answers changed for everyone, including competitors, before you assume something broke on your side.

Then build a source gap list: every source that appears in answers where competitors are named and you are not. This list is the most practical output of the whole exercise. It tells you which review platforms, directories, publications or comparison pages the AI tools lean on, and where you are missing. A text file on your own server will not close that gap, which is why I rank something like llms.txt well below this work.

How does Search Console report AI Overviews and AI Mode?

Google counts AI Overviews and AI Mode inside the Web search type of the Search Console Performance report. Search Console also has a generative AI performance report, rolled out to all websites by August 2026, showing impressions in AI features by page, country, device and date. Pair it with clicks from the main report.

Google's page on AI features and your website explains that traffic from AI Overviews and AI Mode is reported in the Performance report within the Web search type, and counted in the overall Search Console data. On its own, that means AI Overview clicks are blended into your normal totals.

The newer generative AI performance report adds a dedicated view. Google's help page describes impressions for AI Overviews and AI Mode, broken down by pages, countries, dates and devices, drawn from the same Web search data. The report also covers generative AI features in Discover, which matters if you publish articles or news. Google's own guide to optimising for generative AI features points site owners to this report for measuring performance in those features.

Here is how I use it:

  1. Filter to Australia if your customers are local. Global impressions can hide a weak domestic picture.
  2. List the pages that earn AI impressions. They tell you which content Google already treats as answer material.
  3. Compare against the main Performance report. A page with rising AI impressions but falling clicks may be answering the question so completely that fewer people need to visit. That is not automatically bad, but it changes what the page should do for you.
  4. Look for gaps. Important service pages with no AI impressions are candidates for clearer answers, better structured data and stronger internal links.

If you are new to the main report, my Google Search Console guide covers the basics. One limit to keep in mind: none of this covers ChatGPT, Perplexity, Gemini or Copilot. For those, your evidence is the prompt set and GA4.

How do you track AI referral traffic in GA4?

GA4's default channel group now includes an AI Assistant channel for visits referred by tools such as ChatGPT, Gemini, Copilot, DeepSeek and Grok. Clicks from Google AI Overviews and AI Mode are excluded and stay in Organic Search. Check the session source too, because some assistants, such as Perplexity or Claude, may still appear as ordinary referrals.

According to Google's default channel group definitions, GA4 sets the medium to "ai-assistant" when the referrer matches its list of AI assistants, and groups those sessions in the AI Assistant channel. The same page states that Google's AI Overviews and AI Mode are excluded and are classified as Organic Search. That single rule explains a lot of confusion: AI traffic from Google's own features never shows up in the AI Assistant row.

As a Google Analytics 4 certificate holder, this is the setup I check in every property I audit:

  1. Open Reports, then Acquisition, then Traffic acquisition, and look at the AI Assistant row in Session default channel group.
  2. Add Session source as a secondary dimension. This shows which assistants actually send visits, and whether any sit in Referral instead.
  3. Build an exploration with a source filter for chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. This catches sessions from periods before the channel existed in your property, and any assistant that is not on Google's list.
  4. Look at landing pages and key events, not just sessions. AI visitors often arrive on a specific page that answered their question, and the real test is whether they enquire, book or buy.

Two limits matter. Some AI apps strip the referrer, so a visit arrives as Direct. And many people read an AI answer, then search your brand name on Google, so the visit lands in Organic Search or Direct even though AI did the persuading. In my experience GA4 undercounts AI influence rather than overcounting it. My GA4 for SEO guide covers the wider reporting setup.

Do you need a paid AI visibility tool?

Not to start. A spreadsheet, a fixed prompt set, Search Console and GA4 give a small business a reliable baseline. Paid tools such as Semrush's AI Visibility Toolkit and Ahrefs Brand Radar help when you track hundreds of prompts, several markets or many competitors, but their scores are estimates from sampled prompts.

I use tools when the volume justifies them. Tracking 30 prompts across three platforms by hand takes a couple of hours. Tracking 500 prompts across five platforms in three countries does not work by hand. Before you pay for one, ask the vendor five questions:

  1. Which platforms does it query, and does it query them from Australia?
  2. Where do the prompts come from, and can you add your own?
  3. How often are prompts re-run, and is each one run more than once?
  4. Can you export the raw answers and the cited sources, not just a score?
  5. How does it calculate its headline visibility number?

Expect different tools to report different numbers for the same brand, because they sample different prompts. Compare trends inside one tool over time. Do not compare one tool's score with another's.

What did measuring AI visibility show for a dental clinic?

A Blacktown dental clinic with more than 500 five-star Google reviews was missing from ChatGPT's answers for local dentist prompts. Tracing the sources showed ChatGPT drew dental recommendations from BirdEye, a review platform where the clinic had no profile. Two weeks after the profile was claimed and completed, ChatGPT listed the clinic among its top recommendations.

This is the clearest example I have of why the source column matters. On Google, the clinic was strong on reputation: hundreds of five-star reviews. In ChatGPT, it did not exist. A ranking report would never have revealed the problem, and neither would a dashboard score on its own. What revealed it was asking the same local prompts, reading the answers, and tracing the sources behind them.

The fix was not content and not code. I created and claimed the clinic's BirdEye profile and completed every field with consistent name, address and phone details. Within two weeks, the clinic began appearing in ChatGPT's Blacktown dental recommendations, and later in Perplexity and Google's AI results.

GA4 then showed the result in visits. Of 58 sessions from AI platforms, 52 came from ChatGPT, which is 89.66%. Perplexity accounted for 6.90% and Google's AI products for 3.45%. AI referral visitors engaged at 81%, higher than any other traffic source. Those are small absolute numbers, and I would not dress them up as anything else. But they came from a channel that did not exist for the clinic twelve months earlier. The full numbers, including the Google rankings work that ran alongside, are in the dental clinic AI search case study.

Three lessons from that project shape how I measure every business now:

  1. Strong Google signals do not transfer automatically. Reviews on one platform were invisible to an assistant that trusted another.
  2. Sources beat scores. The useful finding was a named missing source, not a percentage.
  3. Measure before and after each fix. Re-running the same prompts two weeks after the profile went live is what turned a theory into evidence.

If you want this baseline built for your business, my AI visibility audit starts with the same three layers: a fixed prompt set, Search Console's AI data and GA4 referrals, followed by a source gap list you can act on.

Frequently Asked Questions

How long does it take to improve AI visibility?
It depends on the fix. Filling a missing source, as in the dental case, showed up in ChatGPT within two weeks. Content and authority work moves at the pace of SEO, where results usually lag the work by about three months. Nobody can promise a specific AI mention.

Is AI visibility the same as Google visibility?
No, but they overlap. Google's AI Overviews and AI Mode draw on its normal Search index and ranking systems, so strong SEO helps there. ChatGPT, Perplexity and other assistants run their own crawlers and draw on their own sources, which is why a business can rank well on Google and still be missing from ChatGPT.

Can I see the prompts people use to find my business in ChatGPT?
Not from the assistants I track. GA4 shows the visits and landing pages, and Search Console's generative AI report shows which pages earn impressions in Google's AI features, but neither shows the prompt behind a visit. A fixed prompt set is the closest substitute.

Why does GA4 show less AI traffic than I expected?
Three reasons are common. Clicks from Google's AI Overviews and AI Mode are classified as Organic Search, some AI apps strip the referrer so visits arrive as Direct, and many people read an answer and then search your brand instead of clicking. Treat GA4's AI numbers as a floor, not the full picture.

How often should I re-run my prompt set?
Often enough to see change without chasing noise. Monthly suits most small businesses, plus an extra run two to four weeks after any significant fix, such as a new review profile or a rewritten service page. Keep the prompts, platforms and conditions identical each time.

AI VisibilityAI Search OptimisationGenerative Engine OptimisationGoogle Search ConsoleGA4ChatGPTGoogle AI OverviewsPerplexity
Share:
Kaan TURK
Kaan TURKAbout
SEO & AI Search Consultant

15 years of SEO expertise. Former SEO Lead for Louis Vuitton, LC Waikiki and Vakko. Mathematics graduate bringing data-driven precision to search engine optimisation.

Ready to Implement These Strategies?

Get expert help implementing these SEO strategies for your business. With 15 years of experience and work with brands like Louis Vuitton, I can create a customized plan tailored to your goals.

If you want to know what is holding your site back, the first step is a discovery call. Before the call I look at your site and your competitors, and on the call I show you what I would fix first.

Book a discovery call