Google AI features sent one site an estimated 21% of its clicks. That is the headline finding of an opt-out experiment by Cyrus Shepard, founder of Zyppy SEO, who removed his site from Google’s AI features for 28 days and measured what happened to its traffic. The number is striking — and it comes with a warning that matters as much as the number itself: this was a single experiment on a single site, and Shepard says to treat the findings as a directional yardstick, not as Google gospel. Its value is not that 21% applies to your site — it is that, for the first time, someone measured what AI search visibility is worth in clicks.
What did the Zyppy experiment actually measure?
The experiment measured how much of one website’s Google search traffic depended on Google AI features — specifically AI Overviews and AI Mode. Shepard used a Search Console control to exclude Zyppy’s site from those features, then opted back in and compared the two periods to estimate AI features’ share of clicks. Source: Search Engine Journal, “Google’s AI Features Sent One Site An Estimated 21% Of Its Clicks” (Nov 9, 2026), by Matt G. Southern, reporting on the primary source, Cyrus Shepard, “We Opted Out of Google AI Search – Here’s What Happened”.
Most reporting counts impressions — how often a site appears in an AI Overview or an AI Mode answer. Impressions show you were visible; they do not show whether anyone clicked. Shepard’s experiment answered the click question, because clicks turn visibility into visits, leads and revenue.
How did the Zyppy opt-out experiment work?
The design was simple. Shepard excluded Zyppy’s site from Google’s AI features using Search Console’s Search generative AI control on September 1, 2026, and opted back in on September 29, 2026 — 28 days. The control had rolled out worldwide on August 31, 2026, the day before.
- Measure the baseline. Record how often the site appeared in AI Overviews and AI Mode, and how its overall search traffic looked.
- Opt out. Exclude the site from Google’s AI features in Search Console.
- Observe for 28 days. Track impressions and clicks while excluded.
- Opt back in. Restore eligibility on September 29, 2026, and watch whether the earlier pattern returns.
- Compare and estimate. Combine the impression data with an estimated AI click-through rate to estimate the share of total clicks AI features had produced.
The speed of the change was itself a finding. Within an hour of opting out, AI-feature impressions went to zero; when Shepard opted back in, they returned almost as fast. The site’s AI visibility was real and ongoing — not a one-off spike.
What were the experiment’s numbers?
| Measure | What Shepard found |
|---|---|
| AI-feature impressions before opt-out | 500–1,000 per day, about 10% of all Google search impressions |
| AI-feature impressions during opt-out | Zero, within an hour of opting out |
| AI-feature impressions after opting back in | Returned almost as fast as they had disappeared |
| Estimated AI search CTR | About 0.9%, with a reasonable range of 0.7–1.1% |
| Site-wide organic CTR | About 0.4% |
| Estimated share of total search clicks from AI features | About 21% |
| Estimated total share including the halo effect | About 23% |
| Opt-out period | September 1 to September 29, 2026 (28 days) |
The most surprising line is the click-through rate. Shepard estimated an AI search click-through rate of about 0.9%, with a reasonable range of 0.7% to 1.1% — about double the site’s overall organic click-through rate of about 0.4%. That runs against a common assumption: that an AI answer satisfies the searcher and removes the reason to click. One reason may be selection: pages in AI features are ones Google already considers relevant and trustworthy. Another is that a citation inside an AI answer acts as a recommendation, and a recommended link is a link people follow.
How did Shepard get from 10% of impressions to 21% of clicks?
AI features accounted for about 10% of the site’s Google search impressions. If AI-feature clicks arrived at the same rate as ordinary clicks, they would account for about 10% of clicks too. But Shepard estimated AI-feature clicks arrived at about double the site’s overall rate — about 0.9% against about 0.4% — so a 10% share of impressions became about 21% of total search clicks.
That estimate depends on the click-through rate, which is the one number Shepard could not read directly from Google’s reporting. Google counts a click on an AI Overview or AI Mode link as a standard click in Search Console’s Performance report, so AI clicks are mixed in with ordinary clicks, and the generative AI report shows impressions only. Shepard estimated it from the change in total clicks during the opt-out period, which is why he gives a range of 0.7% to 1.1%. The 21% figure is the best reading of the evidence from one site over 28 days — not a figure Google published, and not one any other site should expect to reproduce exactly.
What is the halo effect, and why does it add another 2%?
Shepard found a second, subtler effect on top of the direct clicks. During the opt-out period, the site also lost some clicks on its regular listings. His explanation is a halo effect: appearing in AI features seems to lift clicks on a site’s ordinary results as well, perhaps because searchers who see a brand cited are more likely to click its listing below. The effect was strongest on the URLs with the most AI visibility. Adding that halo effect, estimated at about 2%, puts the total share of clicks tied to AI features at about 23%.
The halo effect is an inference from the data, not a number Google reports, and it is small next to the direct effect — but being present in AI features may make your ordinary listings perform better too.

Why can’t you measure this in Search Console?
Google’s reporting leaves publishers unable to answer the most basic question about AI search: how many clicks did it send me?
- Clicks are counted, but not separated. A click on a link inside an AI Overview or an AI Mode answer counts as a standard click in Search Console’s Performance report — you cannot tell which clicks came from AI features.
- The AI report shows impressions only. Search Console’s generative AI report shows how often your pages appeared in AI features — no clicks, no queries, no click-through rate. You cannot turn those impressions into a traffic estimate without guessing the rate.
- The opt-out control is new. The Search generative AI control rolled out worldwide on August 31, 2026 — before that, switching AI features off to measure the difference was not possible.
- Regulators are pushing for more. The UK’s Competition and Markets Authority gives Google until December 3, 2026 to provide publishers with impressions, clicks and click-through rate for AI search features.
Until then, track what you can: AI-feature impressions, total clicks, and clicks on the pages that appear most in AI features. Look for Shepard’s pattern: when AI visibility rises or falls, do clicks on those same pages move with it? Our guide to how to measure AI visibility covers the reporting that does exist.
What are Google AI features and how do they show links?
Google AI features, in the sense Shepard tested, means the two places where Google’s AI answers appear in search: AI Overviews, the AI-generated summaries at the top of many search results, and AI Mode, the conversational experience where users ask follow-up questions and receive AI-composed answers. Both cite the web pages they draw on, with links a searcher can click.
Those citations are the mechanism behind the experiment’s result. An AI Overview does not hide its sources; it links to them. A searcher who wants more detail or the original data follows the link — a citation is not only a mention, it is a doorway. A site cited often, on queries where searchers still want to read further, can receive real traffic from AI search, as Zyppy did. Our comparison of AI Mode versus AI Overviews explains how the two formats differ — but for a publisher, the goal is the same on both: be the page the AI chooses to cite.
Should you opt out of Google AI features?
For most sites, the Zyppy experiment points to no. If AI features were producing about a fifth of one site’s search clicks, opting out means giving up roughly a fifth of your search traffic, plus the halo effect on your ordinary listings.
Opting out is worth considering in narrow cases. If your content is summarised so completely that an AI answer replaces the visit, your click-through rate may be far lower than Zyppy’s. If you object in principle to your content appearing in AI answers, the control exists and works quickly — Shepard’s impressions went to zero within an hour.
Three facts should shape the decision. First, exclusion is not a ranking signal for regular search: opting out does not change how your pages rank in ordinary results. Second, it does not opt your site out of AI model training — that needs Google-Extended in robots.txt. Third, opting out is reversible: Shepard opted back in after 28 days and his AI impressions returned almost immediately. If you test it, test it the way he did — with a clear baseline, a defined period, and a decision made on your own numbers.
What does this mean for link building?
The Zyppy experiment matters to link builders because it changes what a good link is worth. A link has always had two values: the authority it passes, and the traffic it sends directly. AI search adds a third — a link from a page that AI features cite can put your brand inside the AI answer itself, and that citation can send clicks at a rate that beats ordinary listings.
- Citations are now a traffic channel, not just a trust signal. A placement on a site that appears in AI Overviews puts your brand in front of searchers as they ask the question. Judge placements partly by whether the host site’s pages show up in AI features.
- The linked page must deserve the click. A citation earns a click when the searcher wants more than the AI summary gave them. A page that answers the question directly, then goes deeper than the summary, is worth clicking.
- Relevance matters more than ever. AI features cite pages that match the query closely. A link from a niche-relevant site, on a page genuinely about your topic, is far more likely to feed an AI answer than a link from an unrelated page.
- Measure AI visibility alongside rankings. If a placement’s host page starts appearing in AI features, report that to the client alongside the link itself.
None of this replaces the foundations. Links still matter for AI search: our article on whether backlinks matter for AI search covers the evidence that the pages AI systems cite tend to be pages other reputable sites link to. The Zyppy experiment adds the other half: once you are cited, the citation itself can send traffic — and that traffic may be a larger share of your search visits than anyone expected.

When should you hire a service like VonixCo?
Hire help when the work above is more than your team can do consistently: earning placements on niche-relevant sites whose pages AI features cite, writing guest posts direct enough to be quoted and deep enough to be worth clicking, and tracking whether those placements show up in AI answers as well as in rankings. One guest post is easy; a steady programme on the right sites takes a team and a process.
VonixCo is a niche-relevant guest posting and SaaS link building service based in Lahore, Pakistan. Its services are SaaS guest posts, niche guest posts, link building and guest post content writing, with plans at $500, $950 and $1,800 per month, content included and a live link report with every placement. Contact: info@vonixco.com and WhatsApp +92 340 0096621.
If your needs are simple — a few placements, a few pages rewritten so AI systems can cite them — do that work yourself this week. Bring in a service when volume, consistency or outreach grows beyond what your team can sustain. And keep the experiment’s caveat in mind: a single site’s 21% is a yardstick, not a promise. No honest provider can guarantee how much AI traffic your site will receive.
Frequently asked questions
What is the Zyppy experiment?
A 28-day test by Cyrus Shepard, founder of Zyppy SEO. He excluded his site from Google’s AI features using a Search Console control from September 1 to September 29, 2026, then opted back in. Comparing the two periods, he estimated AI features produced about 21% of the site’s total search clicks.
How much of a site’s clicks can Google AI features send?
For Zyppy’s site, about 21% of total search clicks, or about 23% including a halo effect on ordinary listings. That figure applies to one site only — AI visibility varies widely across publishers.
What click-through rate do AI Overviews and AI Mode get?
About 0.9%, with a reasonable range of 0.7% to 1.1% — about double the site’s overall organic rate of about 0.4%. The rate is an estimate, because Google’s reporting does not separate AI-feature clicks from ordinary clicks.
Can I see AI search clicks in Google Search Console?
Not directly. Google counts clicks on AI Overview and AI Mode links as standard clicks in the Performance report, so they are mixed in with ordinary clicks. The generative AI report shows impressions only.
Should I opt out of Google AI features?
For most sites, no. The experiment suggests opting out can cost a large share of search clicks, and exclusion does not change regular search rankings. Consider it only if your own data shows AI answers replacing visits — and test it with a clear baseline and a defined period.
What does the Zyppy experiment mean for link building?
A citation inside an AI answer can be a real source of clicks, not just a mention. Placements on niche-relevant sites whose pages AI features cite can put your brand in front of searchers and send visitors to your linked pages. Links still matter as trust signals, so relevance and quality remain the deciding factors.