The 23% AI Traffic Stat Rests on 5 to 9 Clicks a Day

You may be about to put a number in a budget deck: Google's AI features drove 23% of one site's search clicks. Before you do, look at how small that site is.

Cyrus Shepard, who runs the Zyppy Signal newsletter, published the full results of his opt-out experiment on November 5, 2026. He says AI search accounts for about 23% of Zyppy's search clicks. He reports 500 to 1,000 AI impressions a day and estimates a 0.9% click rate on them. Taking those two numbers at face value gives roughly 5 to 9 AI clicks a day. That is my arithmetic, since the post gives no daily count.

The study is useful. The 23% is a poor number to carry into a planning meeting.

Shepard laid out the experiment in a Zyppy Signal post on November 5, a week after a first version of the numbers drew coverage from Search Engine Roundtable.

Google rolled out its Search generative AI control worldwide on August 31, 2026. It lets a site exclude itself from AI Overviews, AI Mode and generative features in Discover. Google warns that excluding a site means no traffic or impressions from those features. Shepard wanted to know how much traffic that would cost.

So he opted Zyppy out on September 1. Within an hour, he says, the site stopped appearing in Google's AI features. Google's documentation says content is excluded within 1 to 2 days of the control going live, and his hour beat that. AI-feature impressions fell by more than 99%. On September 29 he opted back in, and impressions returned almost as fast.

How the clicks were counted

Shepard says Google's Search Console does not report clicks from AI features separately. He inferred them by comparing total clicks in the 28 days with the control on against the 28 days after he switched it off. He says he checked the result against matched-query and position-level data to limit the effect of shifts in rankings and query mix.

The estimate is an AI click rate of about 0.9%, with a range of 0.7% to 1.1%. Zyppy's overall click rate in Google is around 0.4%. He also estimates that regular organic clicks got a lift of around 2.3% when the site was cited in AI answers, and he calls it a halo effect. That estimate comes from the same before and after comparison. He puts AI impressions at about 10% of the total, which he turns into about 21% of clicks, and about 23% with the halo. The inputs as published give 22.5%, and he does not show how he got to 21%.

He is open about the limits. His write-up says the data comes from "a single experiment on a single site" and should be read as "a directional yardstick." In the same post he says Zyppy has done a lot of work to win AI citations, which may be why so many clicks came from them. His conclusion is blunt. Cyrus Shepard writes in the Zyppy Signal post: "But now it not only seems like the smart move to include your site in AI search and optimize for AI recommendations/citations, but the only reasonable action to take."

The size of the base

Take his published estimates as given. If AI impressions are 500 to 1,000 a day and make up about 10% of all impressions, total impressions are 5,000 to 10,000 a day. At the 0.4% site click rate, that is 20 to 40 search clicks a day. At 0.9%, the AI impressions produced 4.5 to 9 of them.

The share is also sensitive to the range he gives. Run the same inputs at 0.7% and at 1.1% and the AI share, before any halo, lands between 17.5% and 27.5% of clicks. In October he put AI impressions at about 8% of the total, and at 0.9% that gives about 18%. None of this breaks his finding. It shows that the 23% sits on a handful of clicks a day, and that a small change in an input moves it by several points.

What the test does and does not isolate

Barry Schwartz of Search Engine Roundtable pointed out a limit when the first numbers came out. In his words: "The issue is, again, AI Overviews still showed up on the search results page, so it is not comparing clicks on web results with and without AI Overviews on the search result page." That was Schwartz at Search Engine Roundtable on October 30, 2026.

Shepard confirmed to Schwartz that AI Overviews were on the page before, during and after the experiment. His site was missing from them for 28 days. The test measures what losing the citation cost one site, not what losing AI Overviews would do to the page.

The opt-out period came first and the opt-in period second, and anything that changed in Google between early and late September sits inside the comparison. Shepard says his matched-query and position checks reduce that risk. They do not remove it.

On the halo, Greg Sterling of Near Media commented on the Schwartz article that his research points the same way. Sterling said: "We've seen in our research at Near Media that AI overviews create a kind of brand lift or awareness that translates into subsequent searches and clicks, not consistently but many times."

Sterling says it does not happen every time. Schwartz himself wrote that he has not seen anyone else produce data like this, probably because few sites will risk opting out.

My read

I would not call this a benchmark. It is one small site that has worked on AI citations, measured over two 28-day windows, with the clicks inferred instead of reported. The share would look different on a site with ten times the traffic and no citation work.

The part that holds up better is the ratio. AI links got more than double the click rate of the site as a whole. That fits my view that a citation is a lead indicator you can use, not an outcome you bank. The 0.4% baseline also leaves an open question for me: how much of the AI click rate is just the AI answer sitting above everything else on the page? Nobody has run this on a site large enough to answer it.