How do you measure AI search visibility?

There is no rank in AI search, so measure presence instead

For twenty years the search scoreboard was a position. AI search has no equivalent. Run the same question through ChatGPT a hundred times and you get a hundred different answers, with businesses appearing, disappearing and changing order. A vendor once emailed one of our clients offering to rank him number one on ChatGPT. That is impossible. ChatGPT has no rankings.

What you can measure is presence rate, also called visibility percentage or share of voice: of all the times a buyer question is asked, how often does your business show up in the answer? A single answer is a slot machine payout. A pattern across fifty answers is data.

Build the prompt set from the questions buyers actually ask, not your old keyword list. The average ChatGPT prompt runs around 23 words versus three or four for a Google search, and when SparkToro asked 142 people to write a prompt with the same shopping intent, barely any two looked alike. So write ten to twenty questions in several phrasings, run each several times, and log how often you appear. Anyone selling more precision than that is selling the AI-era version of snake oil. How to earn those mentions is covered in What is generative engine optimization?

What Search Console shows for AI Overviews, and what it hides

Open Google Search Console for almost any business in 2026 and you see the same chart: impressions climbing, clicks flat or falling. That is often the signature of being cited inside AI Overviews, which counts as an impression without a click. Google folded AI Mode data into the main performance report with no separate filter, so the AI effect is baked into your headline numbers.

In June 2026 Google added a dedicated Generative AI performance report inside Search Console. It shows how many times links to your site appeared inside AI Overviews and AI Mode, by page, country and device. It answers the question I get asked first: are we showing up in the AI answers at all? When we shipped answer-first model research pages for a Honda dealer we work with in late July 2026, this report showed it first: weekly impressions inside Google's AI features went from roughly 350 in late June to about 1,400 by mid-August.

Know its limits. As of October 2026 it reports impressions only, with no query data and no separate clicks, and it is still rolling out gradually. Impressions tell you that you appeared, not what was said, which is why the manual check below still matters. And clicks from AI Overviews are counted as ordinary organic clicks, so this report shows Google's AI footprint, not traffic.

The GA4 direct-traffic problem and how to find AI referrals

Through 2025 and 2026, businesses everywhere watched organic traffic fall in Google Analytics 4 while direct traffic rose. Direct is supposed to mean someone typed your URL. Did twice as many people suddenly start typing business URLs? Of course not.

We tested it at SmartSites across nearly 1,500 client GA4 properties, year over year. Direct sessions jumped about 66 percent, and the jump was almost identical whether or not we manage the client's SEO. When something hits everyone equally regardless of what anyone did, it is a measurement artifact. When someone clicks a citation in an AI Overview or follows a link out of ChatGPT, the referrer is often stripped. GA4 sees a visitor with no origin and files it under Direct.

What to do in GA4:

• Use the built-in AI Assistant channel. Added in May 2026, it labels visitors from ChatGPT, Gemini, Copilot, DeepSeek and Grok. It only counts forward from launch, misses Perplexity, and catches a fraction of a percent of what sits in Direct.
• Build a custom channel group for the referrers it misses, including Perplexity.
• Watch deep-page direct landings. Real type-in traffic goes to your homepage. New visitors arriving “direct” on a product or research page are almost always re-routed search or AI clicks.
• Remember what is not there. Clicks from AI Overviews and AI Mode never appear in the AI channel; Google counts them as organic search.

It is not all mislabeling: real search clicks declined too, because the AI often answers completely. So stop treating GA4 organic as the SEO scoreboard. Use Search Console as ground truth, with leads and sales as the final word. And since only about one in ten people click an AI citation, 2,000 monthly AI referrals implies roughly 20,000 AI recommendations.

Bing Webmaster Tools: the free report most people have never opened

In February 2026 Microsoft launched an AI Performance report inside Bing Webmaster Tools, the first first-party view of where your content is cited in AI answers. It covers Copilot, Bing's AI summaries and Microsoft's partner AI experiences: which pages get cited, which questions triggered the citation, and the trend. By June 2026 Microsoft had added a citation share metric, your slice of citations for a question versus everyone else's. It is free.

Copilot and Bing are a smaller slice of AI search than Google, so treat it as directional. But much of ChatGPT search is currently powered by Bing, so it carries further than Bing's market share suggests. Verifying your site takes ten minutes.

The monthly AI spot-check and how often to re-measure

Dashboards tell you that you appeared. Only a human reading the answers tells you what was said. So every business should run a monthly spot-check that costs nothing and takes half an hour.

1. Ask ChatGPT, Gemini and Google AI Mode the questions your buyers actually ask: “best [category] near [city],” “should I buy from [your business]?”
2. Ask each several times, in different phrasings. Single answers are random; patterns are real.
3. Log how often you appear, what the AI says about you, and which competitors keep showing up.
4. Pull the Search Console and Bing AI reports the same day and note the trends.

Monthly is the right cadence because AI systems re-crawl continuously but not instantly. Fix a stale listing or publish a new page and the effect shows up in weeks, so weekly checks produce noise and quarterly checks hide the trend. Paid AI-visibility tools can automate the runs, but for a single-location business the manual routine is where to start. If a wrong claim keeps appearing, follow the citations: nine times out of ten it lives on a real page you forgot existed.

An AI search readiness checklist

Measurement only matters if it leads somewhere. Here is the checklist I ask marketing teams and agencies to score honestly, generalized from the dealer version in the book.

1. AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) are allowed in robots.txt.
2. Key page content and product data are readable without JavaScript, verified.
3. Structured data is in place: LocalBusiness or Organization, product and FAQ schema.
4. Every service, department and product line has its own detailed page.
5. Key pages answer the core question in the first sentence, with specifics and tables.
6. You produce the formats that get cited (numbered buying guides, comparison pages, pages built on first-party data) and have stopped funding generic how-to content.
7. Your business facts (name, address, phone, hours) are identical everywhere on the web.
8. You have an active plan for earning mentions in “best of” lists, local press and industry sites.
9. Review generation runs continuously, producing specific, quotable reviews.
10. Your Merchant Center feed is clean, complete and updating daily.
11. GA4 tracks AI referrals (AI Assistant channel plus a custom channel group), and you judge SEO by Search Console plus leads.
12. Someone runs the monthly spot-check and logs visibility over time.

Score twelve out of twelve and you are ahead of the overwhelming majority of businesses, including plenty twice your size.

Five questions to ask any vendor promising AI rankings

Being data-driven is your best defense against wasted spend. These five questions separate professionals from snake oil in ten minutes.

1. “How do you measure AI visibility?” Right answer: presence rate across many prompts and runs. Wrong answer: any promise about ranking number one in ChatGPT.
2. “What exactly would you change on our website, and why?” Right answer: crawler access, rendering, schema, dedicated pages, answer-first content. Wrong answer: proprietary magic they cannot explain.
3. “How does this build on our existing SEO?” GEO is a layer on the same foundation. Anyone selling GEO instead of SEO does not understand either.
4. “Show me a before-and-after for a client.” Real work has receipts; in 2026 that includes a Search Console Generative AI chart.
5. “What happens when the platforms change next quarter?” Right answer: humility plus fundamentals. Anyone claiming certainty is selling it.

Two more yellow flags: a pitch that leads with llms.txt (Google said in May 2026 it is not needed for its AI search features), and any tool reporting a “position” inside AI answers. The full chapter is in the book under Measuring AI Search; to have this explained live for your team, see speaking.

Questions people ask