Google's AI links drew 21% of one site's clicks from 10% of its impressions. How to estimate your own share

Every brand team that has looked at Google Search Console this year has asked the same question and found no column that answers it: how many of our clicks came from the AI answer? Google counts a click on a link inside an AI Overview or AI Mode answer as an ordinary click. Its own documentation says traffic from those features is "included in the overall search traffic in Search Console," reported in the Performance report under the Web search type. You can see that you appeared. You cannot see what it was worth.

Cyrus Shepard, who runs the SEO consultancy Zyppy, found a way to estimate it on his own site. He pressed the button most publishers have been told not to press. His write-up in the Zyppy Signal newsletter on November 5, 2026 is the first public before-and-after test we have seen of what Google's AI citations are worth in clicks. It is one small site and he says so plainly. It is still the best number available, and the method is one a CPG brand can borrow without switching anything off.

What Zyppy did

In August 2026 Google made a control available worldwide that lets a site exclude its content from AI features in Search. Using it removes the site from AI Overviews and AI Mode, including the links, while leaving its regular listings in place. It does not stop Google from using the site for AI training; that is a separate control.

Shepard opted Zyppy out on September 1 and back in on September 29, then compared the 28 days with the control on against the 28 days after. According to his write-up, the site disappeared from Google's AI features within an hour. Before the switch it had been getting 500 to 1,000 AI-feature impressions a day, about 10% of its total search impressions. During the opt-out those impressions fell by more than 99%. When he opted back in they returned almost as fast.

Because Search Console does not split AI clicks out, he inferred them: total clicks in the AI-off period against total clicks in the AI-on period, checked against matched queries and positions so that ranking changes and a shifting query mix would not drive the result.

The numbers he reported

Show the data

Result type	Click-through rate
AI Overview and AI Mode links (estimate)	0.9%
Site average, all Google Search	0.4%

Source: Zyppy Signal, November 5, 2026. Single site. The AI figure is an estimate with a stated range of 0.7% to 1.1%.

The estimated click rate on AI links was about 0.9%, with what Shepard calls a reasonable range of 0.7% to 1.1%. The site's average across all of Google Search is around 0.4%. So, on this site, a link in the AI answer drew clicks at a bit more than twice the rate of an average result.

That is where the headline comes from. AI-feature impressions were about 10% of the site's total impressions, and because they clicked at roughly double the usual rate, Shepard estimates they produced about 21% of the site's search clicks. He also measured a small lift, around 2.3%, in click rate on Zyppy's regular listings when the site was cited in the AI answer as well, which he calls a halo effect. Add that in and he attributes about 23% of search clicks to AI features.

Show the data

Measure	Share of site total
Share of search impressions	10%
Share of search clicks	21%
Share of clicks incl. halo	23%

Source: Zyppy Signal, November 5, 2026. All three figures are Shepard's estimates for one site.

An earlier version of these findings, shared on LinkedIn at the end of October and covered by Search Engine Land, drew criticism for leaving out the dates, the method and the size of the drop. The November write-up fills in most of that.

How much weight the 21% can carry

Not much on its own, and the reasons are worth understanding before anyone pastes it into a budget deck.

The base is small. AEO Sherpa's Alex Brufsky did the arithmetic on Shepard's published inputs: 500 to 1,000 AI impressions a day at a 0.9% click rate is roughly 4.5 to 9 AI clicks a day. The share is real, but it sits on a handful of clicks.

The share moves a lot inside the stated range. If AI impressions are 10% of the total and the site average is 0.4%, the AI share of clicks is simply 10% times the AI click rate divided by 0.4%. We ran that across Shepard's range.

AI click rate	AI clicks per 1,000 total impressions	All clicks per 1,000 total impressions	AI share of clicks
0.7% (low end)	0.70	4.0	17.5%
0.9% (central)	0.90	4.0	22.5%
1.1% (high end)	1.10	4.0	27.5%

The central case gives 22.5%, a little above the 21% Shepard reports; he does not show the exact sum behind his figure, which is a point AEO Sherpa makes too. Either way the honest range is something like a sixth to more than a quarter of clicks.

The test removed the citation, not the AI Overview. AI Overviews kept appearing on the results page throughout; Zyppy was just missing from them. So the test measures what losing your place in the answer costs, which is the question a brand actually faces, not what Google's AI features do to clicks in general.

The site has worked to be cited. Shepard says Zyppy has done a fair amount of work to win AI citations, which may be why so many of its clicks came from them. A site that has done none of that work would likely see a smaller share.

None of these points undo the finding. They change what it is: a ratio from one site, not a forecast for yours. The ratio, AI links clicking at about twice the average rate, is the part we would carry forward.

What this means for a CPG brand site

Most packaged goods brands do not depend on organic search clicks the way a publisher does. The purchase usually happens at a retailer, and the brand site is a reference shelf. That makes it tempting to shrug at this test. We think three parts of it apply directly.

First, if your site is cited in Google's AI answers, those citations are probably your best-converting search placements, not your worst. The answers sit at the top of the page, and the people who click through from them have already read a summary and still want more. Google's own documentation says it has seen that clicks from results pages with AI Overviews are "higher quality," meaning people spend more time on the site.

Second, opting out is now easy, and that makes it dangerous. A nervous legal or brand review can remove a site from AI features in minutes. Zyppy's test is the clearest evidence that the cost of doing so is more than the clicks you can see. We covered the wider case in our think piece on opting out and in tip 34.

Third, the clicks you can measure on your own site are a fraction of the influence. When AI answers recommend a product, the next step is often a retailer, as WISLR's study of ChatGPT's Shop at buttons showed. Your AI click share is a floor, not the whole effect.

How to estimate your own AI click share without opting out

You do not need to switch anything off to get a rough number. You need three figures and an honest range.

1. Get your AI-feature impressions. In Search Console, use the report Zyppy used to see AI-specific impressions, for the last 28 days. Note the share of your total Web impressions they represent.
2. Get your overall click rate. From the Performance report, Web search type, same 28 days.
3. Apply a range, not a point. Zyppy's 0.7% to 1.1% is one site's estimate. Use it as a starting range and widen it if your average click rate is very different from 0.4%.
4. Run the sum. AI clicks are AI impressions times the click rate. Divide by your total clicks.
5. Write the result down as a range with the date. Then repeat it monthly. The trend tells you more than any single month.

Input	Value
Total Web impressions, 28 days	300,000
AI-feature impressions	36,000 (12%)
Total clicks, 28 days	1,800 (0.6% average click rate)
AI clicks at 0.7%	252
AI clicks at 1.1%	396
Estimated AI share of clicks	14% to 22%

This is a sketch, not a measurement. It assumes your AI links click at a rate similar to Zyppy's, which nobody can confirm for your site. What it gives you is a defensible order of magnitude and a reason to watch the AI impressions line as closely as the rank line.

If you want a firmer answer, the only clean method we know is Zyppy's: a deliberate switch with dates on both sides. If you try it, do it on a section where losing a month of AI visibility will not hurt, such as an old blog archive, not product pages or the brand's about page. Annotate the dates in analytics, and do not run it across a holiday or a site migration.

Better data may arrive, but not for US brands first

The UK's Competition and Markets Authority has given Google until December 3 to provide publishers with impressions, clicks and click-through rate for AI search features, according to Search Engine Journal's summary of the rules. Those rules cover AI features shown to people in the UK. Nothing equivalent has been announced for the US. Until it is, estimates like this one are what a US brand has to work with.

What we would do this week

• Pull your AI-feature impressions and run the five-step estimate above. Record it beside the date.
• Check which of your pages earn those impressions. If they are FAQ, how-to or ingredient pages, keep them current; they are doing more work than their traffic suggests.
• Make sure nobody on the team plans to use the opt-out without a written reason and a test plan.
• Keep a short list of shopper questions and check whether Google's AI answer cites your site for them, as in tip 36. Seeing the answer itself matters as much as the click; teams that do this across assistants sometimes use tools such as SearchPanel from XENA Intelligence, though a spreadsheet works for a first pass.
• Put the AI share on the same page as the caveats from our holiday measurement piece, so it is read as a range.

The test does not prove AI citations will send your brand site a fifth of its clicks. It does show, on the only site anyone has been willing to switch off, that a link in the answer is worth more than an average result. That is enough to protect the citations you have while the better data arrives. Our AI shelf stats tracker now carries the figure with its caveats.