How to measure your brand’s visibility in AI search

Someone will ask how visible your brand is in AI search, and no tool can give you one true number. Google and Bing show part of it, third-party tools estimate the rest, and some rank trackers have been fed bad Google results since September.

The short answer: treat AI visibility as an estimate you build from three things. Use the reports Google and Bing give you, run a fixed set of prompts many times and report how often your brand appears, and check any tracker’s trend against Search Console before you believe it.

Can you measure brand visibility in AI search at all?

Yes, as a rate over many answers, not as a position in one answer. In late 2025, SparkToro and Gumshoe had 600 volunteers run 12 prompts through ChatGPT, Claude and Google’s AI results, 2,961 times in total. There was less than a 1 in 100 chance that ChatGPT or Google’s AI would give the same list of brands twice, and less than 1 in 1,000 of the same list in the same order.

Some brands still showed up again and again. Rand Fishkin concluded that a ranking position in AI is meaningless, while the percentage of answers that include you, across many prompts run many times, is a reasonable metric. The study isn’t peer-reviewed, and Gumshoe sells AI tracking, which Fishkin disclosed. I don’t report positions.

What do Google and Bing actually tell you?

They tell you when your links were shown or cited, and not much else. Start here anyway, because this data comes from the engines themselves.

• Search Console’s Generative AI performance report. Google says it reached all websites worldwide as of August 31, 2026. It covers AI Overviews and AI Mode, and it shows impressions by page, country, date and device. There are no clicks, no queries and no split between the two features. Sites with too few AI impressions may not see the report at all.
• The standard Performance report. Google’s help page says a click on a link in an AI Overview or AI Mode counts as a click. Those clicks are mixed in with the rest of your search data, so you can’t pull them out.
• Bing Webmaster Tools’ AI Performance report. It launched in public preview on February 10, 2026, and covers Copilot, AI summaries in Bing and unnamed partners. It shows citations, cited pages and grounding queries, the phrases the AI used to find your content. Bing says the grounding data is a sample. There are no clicks here either.
• ChatGPT. I couldn’t find any citation report from OpenAI for site owners. What you can measure is the visits, which I cover in how to track AI assistant traffic in GA4.

Google’s impressions count links to your site shown in an AI feature. An answer that names your brand without linking to you doesn’t appear, so the report measures citation, not mention. Gary Illyes said recently that the team wants to give more data but has nothing to announce.

How do I build a prompt set that gives a usable number?

Write the questions your buyers ask, freeze the wording, and run them often enough to see how much the result moves by chance. A tool or a script can do the running.

1. Write a few dozen questions, grouped by topic: category questions (“best X for Y”), problem questions and comparisons. Keep questions containing your brand name in a separate group, because they will inflate your rate. You’re ready when someone in sales recognises the questions.
2. Save the wording and settings in one sheet: engine, country and date. Don’t edit a prompt later; add new ones and report them separately, so months still compare.
3. Run each prompt many times per engine. Fishkin’s volunteers ran each one 60 to 100 times. If you use a vendor, ask how many runs sit behind its figure. One run per prompt isn’t worth reporting.
4. Record three things for each answer: whether your brand is named, whether your site is linked, and which competitors are named.
5. Calculate your mention rate (answers naming you, divided by all answers) and your citation rate. For share of voice, divide your mentions by the mentions of every brand you track.
6. Run the whole set twice in the same week, with nothing changed on your site. The gap between those two runs is your noise. Treat a later change as real only if it is bigger than that gap.

Report each engine separately, because a blended number hides the differences between them.

Why should I cross-check rank trackers with Search Console?

Because some of the rank data arriving in those tools since September was bad. Nozzle reported that Google has been serving rank trackers normal-looking pages with swapped results, and counted 8% of its pages as affected since mid-September, against 0.17% before. Ahrefs said on October 22 that it saw 5% to 20% depending on the data provider, and SE Ranking said on October 16 that about 2% of tracked results were still affected. The vendors measure this differently, and Google hasn’t commented.

AI visibility tools that scrape Google may share this exposure, but I haven’t seen anyone show it happening, so ask your vendor. Here is the check I’d run:

1. In Search Console, open the Performance report for Search results, click Add filter, choose Query and enter one important keyword. Set the country and device your tracker uses.
2. Set the date range to start before September 2026 and turn on Average position. Put the tracker’s chart for the same keyword next to it.
3. Look for falls in the tracker that don’t appear in Search Console. If Search Console is flat while the tracker drops and recovers, suspect the data before your site. Repeat for your most important keywords.
4. Ask your vendor how it detects bad pages and whether it corrected past dates.
5. Add a note to your report for October 31 to about November 3. Some sites saw an odd click spike in Search Console with flat impressions, and some reported it back to normal on November 4. Google hadn’t explained it when I wrote this, so don’t read it as a win.

Average position uses the topmost position of your link, so it won’t match a tracker’s reading exactly. You’re comparing direction, not digits.

Did the AI Overview move on your brand searches?

It may have, and that could change your reports. Search Engine Roundtable reported on November 2 that AI Overviews for many brand keywords moved to the right-hand column, and PPC Land said the same of desktop results on November 3. The examples are from the US, and Google hasn’t announced it or said whether it is permanent. Ahrefs had found AI Overviews on 82.91% of US desktop results for branded keywords on September 29, so this touches a lot of brand searches.

Google’s help page says an AI Overview takes a single position, and that positions are typically counted down the main side of the page, then the secondary side. By that logic, a right-column overview could carry a higher position number while sitting in plain sight. Google hasn’t said how it counts this placement, so this is my reading, not a fact. I’d do two things:

1. Search your brand name on desktop, in a private window, and save a screenshot with the date. Repeat on mobile.
2. Ask your tool whether it records where the overview appeared or only that one appeared.

Search Console’s branded queries filter helps too. It splits Performance data into branded and non-branded queries using Google’s own classification, on top-level properties with enough impressions. I haven’t seen how it treats AI features, so use it to watch brand demand, not AI visibility.

What should go in a monthly AI visibility report?

One short page, with the source next to each line, so nobody mistakes an estimate for a measurement:

1. Add AI impressions from Search Console, by top pages, with the trend. Label it “links shown”.
2. Add citations and the top grounding queries from Bing. Label the grounding queries as a sample.
3. Add mention rate and citation rate per engine from your prompt set, with the noise gap beside them.
4. Add AI assistant sessions and key events from GA4, because that is the line that connects to the business.
5. Add branded impressions and clicks from Search Console, if you have the filter.
6. Finish with one line on what none of this sees, such as unlinked mentions in Google’s answers. Note ads too, since Google is adding more of them to AI answers.

If the number looks wrong, check the basics first: what AI crawlers can actually see on your pages often explains more than any dashboard.

Start small this week: write the questions, run the set twice so you know your noise, and compare your tracker with Search Console once. An estimate with a stated margin is more useful than a precise figure you can’t defend.