Article · AI search · Google · AI Overviews · Search Console · measurement

In one site's test, Google's AI answers brought an estimated fifth of its search clicks

Zyppy, an SEO research site, used Google's opt-out for AI search features to leave AI Overviews and AI Mode for 28 days in September 2026, then opted back in. Its AI impressions dropped to zero within an hour. By comparing the two periods, it estimated that AI answers brought about 21% of its Google search clicks, or 23% counting a small lift to its regular results. It is one test on one site. What it found, what it can't tell you, and how to read your own numbers.

The question
What happened when one site opted out of Google's AI answers for 28 days?
The short answer
Zyppy, an SEO research site, opted out of Google's AI search features on 1 September 2026 and opted back in on 29 September. Within an hour of opting out, its AI-feature impressions went to zero; they had been 500 to 1,000 a day, about 10% of its search impressions. Comparing clicks across the two 28-day periods, Zyppy estimated that links in AI answers were clicked about 0.9% of the time, against about 0.4% for the site overall, and that AI answers brought about 21% of its search clicks, or 23% counting a small lift to its regular results. Zyppy calls this a single experiment on a single site and a directional yardstick. Our read: it is a reason to measure before you opt out, not a number to assume for your own site.

TL;DR: Zyppy, an SEO research site, switched off its own appearance in Google's AI Overviews and AI Mode for 28 days in September 2026, then switched it back on. Its AI-feature impressions went to zero within an hour. By comparing the two periods, Zyppy estimates that AI answers brought about 21% of its Google search clicks, or 23% counting a small lift to its regular results. Zyppy is clear that this is one experiment on one site, and that the site was already set up to be cited by AI. Our read: treat it as a reason to check your own numbers before you opt out, not as a figure for your site.

The likely default: assume AI answers take clicks and give few back, so leaving them costs little.

On this one site, that was wrong. Whether it is wrong for yours is something only your own data can show.

What changed

The test

Zyppy writes that in August 2026 Google shipped a feature worldwide that lets publishers opt out of having their content appear in its AI search features. As we explained in Can AI search skip your website?, that control is a Search Console setting, under Settings > Search generative AI. Zyppy, which publishes SEO research, used it on its own site.

From Zyppy's write-up, published on 5 November 2026:

  • 1 September 2026: Zyppy opted out. Its AI-feature impressions fell by more than 99%. That started a 28-day period with AI answers off.
  • 29 September 2026: it opted back in, followed by a 28-day period with them on. Impressions "returned almost as fast."

It compared the two periods using Search Console's web performance data (clicks, impressions, click rate, average position, queries and landing pages) and Google's Generative AI features report for AI impressions. Zyppy calls it a "28-day interrupted time-series design." In plain words, it compared the same site before and after one deliberate switch, and checked the result against matched queries and positions to reduce the effect of other ranking changes.

Zyppy notes that the opt-out removed its links from AI answers while it still ranked first in regular results for the example search it showed. It also notes that the setting does not stop Google using a site for AI training. Training has separate controls, including the shared robots.txt rule for opting out of AI training, which we covered earlier.

What it found

  • AI visibility went to zero within an hour. Before the switch, the site saw 500 to 1,000 AI-feature impressions a day, about 10% of its total search impressions.
  • By Zyppy's estimate, AI links were clicked at more than twice the site's overall rate. Search Console doesn't split AI clicks from regular clicks, so Zyppy estimated them from the change in total clicks between the two periods. The estimate: about 0.9% of AI impressions led to a click, with a likely range of 0.7% to 1.1%. The site's overall click rate in Google Search is around 0.4%.
  • Its regular results did a little better when it was also in the AI answer. Zyppy measured a small lift of around 2.3% in click rate on regular results, strongest on the pages with the most AI visibility. It calls this a "halo effect" and says nobody knows for sure why it happens.
  • Together, about a fifth of its search clicks. AI impressions were about 10% of the site's impressions but, by this estimate, about 21% of its clicks. Adding the halo, Zyppy puts AI answers at about 23% of its search clicks.

Zyppy has opted back in and says it will stay in for now.

The caveats Zyppy gives

Zyppy says this "is data from a single experiment on a single site," that "AI visibility varies wildly across publishers," and that readers should treat the findings "more as a directional yardstick than Google gospel." It also says it has done "a fair bit of optimizing" to win AI citations, which "may be why so many of our clicks came from AI search."

What it means for your business

This section is our read, not Zyppy's.

The number is not the lesson. Zyppy publishes SEO research, and it says it has worked to be cited in AI answers. Your site could see a much larger or much smaller share, depending on what people search for and how often you appear in AI answers.

The lesson is about method. Google doesn't show you how many clicks come from AI answers. Zyppy found a way to estimate it: change one thing, on a known date, and compare equal periods either side. You don't have to opt out to use the same idea. A dated change you didn't choose, such as an AI feature launching in a country you sell to, gives you a similar before and after, though with more noise.

The result also cuts against an easy assumption: that AI answers only take clicks away. On this one site, Zyppy estimates links in AI answers were clicked at more than twice the site's overall click rate, and that appearing in them slightly lifted clicks on its regular results too. That doesn't prove the same for anyone else. It does mean "AI answers send us nothing" is a claim to check, not assume.

Our default stays the one in our opt-out explainer: for a business that sells something, stay in unless your own data says otherwise. If you want to test it, plan the test the way Zyppy did, with a fixed start, a fixed end and equal periods.

One timing note: Google's November core update started on 10 November and may take up to two weeks. A test that overlaps it will mix two changes, so wait until it is marked complete.

What to do this week

  1. Look at your Generative AI performance report. In Search Console (our explainer shows where), find how many impressions your site gets from AI features, and what share of your total search impressions that is. If it's tiny, an opt-out test won't tell you much.
  2. Don't opt out to save traffic without a test. On the one site in this test, AI answers were estimated at about a fifth of its search clicks. Find out what they're worth to you before switching them off.
  3. If you test, copy the design. Pick a start date, keep the setting off for a fixed period such as 28 days, then turn it back on for the same length. Avoid periods with a known ranking update or a seasonal peak. Write down the dates before you start.
  4. Compare clicks, not just impressions. Losing AI impressions is expected. The question is what happens to total clicks, and to clicks on your regular results, across equal periods.
  5. Keep the training question separate. If your real concern is AI training, the Search Console setting is the wrong control. Read what the robots.txt training opt-out does first.

Sources

Want to know what AI says about your business?

We check it for you.