When AI features take 21% of a site's Google clicks

As of November 2026, the best available estimate of how many Google search clicks arrive through AI Overviews and AI Mode comes from one site that switched the features off and back on. About 21% of that site's search clicks came from AI links, and a small lift in its regular listings took the AI-related share to 23%. If you are weighing a best seo singapore programme against AI search, that number is the starting point, not a benchmark you can copy.

The estimate belongs to Zyppy, an SEO company, and its founder Cyrus Shepard. Search Engine Journal reported it on 9 November 2026. Shepard was explicit about the limits: one experiment, one site, a directional yardstick rather than a rule.

Key takeaways:

AI impressions were 10% of the site's search impressions, but produced about 21% of its clicks, because AI links drew a click-through rate near 0.9% against the site's overall 0.4%.
Adding the halo effect on regular listings, the AI-related share reached 23%.
Google's Search generative AI control is the only lever publishers have, and Search Console reports AI impressions rather than AI clicks.
Google Search Console still does not split AI clicks from web clicks, so any figure you hear is an inference.

What the opt-out experiment actually measured

On September 1, 2026, Zyppy excluded its site from Google's AI features using the Search generative AI control in Search Console. Within an hour, the site stopped appearing in AI search. Its AI impressions, which had been 500-1,000 a day, fell by more than 99%. The site opted back in four weeks later.

Shepard then compared the 28 days opted out with the next 28 days opted in. Total clicks, impressions, average position, queries and landing pages were matched so that ranking and query shifts would not account for the gap. Anything left over was attributed to the AI placements.

That method is an interrupted time series with a single interruption. It cannot separate AI clicks perfectly from seasonal drift or a slow change in the query mix, which is why Shepard published a range instead of a point figure.

Measure (Zyppy, 2026)	Value	What it means
AI share of impressions	10%	How much of the site's visibility sat in AI features
Estimated AI click-through rate	0.9% (range 0.7%-1.1%)	Clicks per AI impression
Site-wide click-through rate	0.4%	The baseline AI clicks are compared with
Ratio of the two	2x	AI placements drew clicks at about double the average rate
AI share of clicks	21%	Share before the halo effect
With halo effect	23%	Share after adding the lift on regular listings

The arithmetic is simple once you accept the inputs: a placement with 10% of impressions but more than double the click rate produces a click share well above its impression share. On a site whose AI visibility is thinner, the same formula returns a far smaller number; Shepard says other sites could come in higher or lower.

An analyst compares two four-week search reports at a desk, notebook open beside the laptop.

The halo effect on regular listings

When the site appeared in AI Overviews, its ordinary organic results also earned more clicks. Zyppy puts that lift at around 2.3%, strongest on the URLs with the most AI visibility. The mechanism is borrowed from advertising: a site seen several times on one results page gets clicked at least once more often, and nobody knows precisely why. Brand reinforcement is the likely explanation.

The practical consequence is uncomfortable for anyone planning to opt out. Leaving AI features does not only remove AI clicks; in this dataset it also dented regular listing clicks, the traffic publishers already count on. A best seo singapore plan that only counts organic clicks from the web results will therefore understate what an opt-out costs.

How to read the numbers without overfitting

Four questions decide whether a site's own share looks like Zyppy's.

Does the site earn AI citations at all? Zyppy had done substantial work to win them, which Shepard offers as a reason its share was high. A site with weak source material may see almost nothing in AI placements.
What is the denominator? AI share of clicks depends on whether you count all search clicks or only organic ones, and on whether brand queries are included.
What is the position profile? AI features push regular results down the page, so sites that rank in the top few positions lose less and gain less than sites sitting lower.
Are your definitions checked by someone who will use them? A best seo singapore programme and an in-house analyst should agree on the denominator before either reports a share to management.

Gap filled: g1, how much of a site's Google search clicks come from AI Overviews and AI Mode

The gap the search results leave open is measurement. Google confirmed the exact scope in its own help page on the Search generative AI control: the Generative AI performance report exists so you can see how your content performs in these features and how a control change may affect traffic. It reports AI Overviews and AI Mode impressions.

Clicks on an AI Overview or AI Mode link are counted by Google as standard clicks in Search Console's Performance report. No help page lists a way to separate them from web clicks. So an AI click share is an estimate built from a before-and-after comparison, or from your own analytics when AI referrers are visible, and it should be labelled that way in any report you send upward.

That is the honest reading: the Zyppy figure is a plausible size for one site, not a Search Console metric, and not a target. A seo consultant who quotes a share without saying which comparison produced it is describing a hypothesis.

A publisher studies a printed traffic report through a magnifying glass, one section left blank.

What Google's control does and does not change

Google's help page sets out the scope clearly. The control covers AI Overviews, AI Mode and generative AI features in Discover. It was rolled out to all websites worldwide as of August 31, 2026. The default is to include a site.

Excluding a site removes its links and content from those features. Other sites' content stays available, and it may look similar to yours. The control is not used as a ranking or inclusion signal for other parts of Search. It does not affect AI training; Google-Extended handles that. To leave Google Search entirely you need noindex.

Timing matters for anyone planning an experiment of their own. Google says content is excluded within 1-2 days after the control goes live, with some pages taking longer because of caching and propagation. Zyppy saw the change within an hour, which is the fast end of that window and a reminder that a single site is not a sample.

Parent and child properties inherit the setting. A property follows its closest parent that has stopped inheriting, and owners of a child property can override it. On a Singapore site with subdomains or URL-prefix properties, that inheritance is worth checking before you conclude the control applies where you think it does.

Where this leaves Topkee's reading of AI clicks

Topkee treats an AI click share as an outcome to observe, not as an acceptance metric, and that distinction shapes how its reporting is built. Topkee SEO targets natural top-10 rankings for commercial keywords and reports AI answers and other special placements separately from organic web results, so a client can see which surface produced a ranking. Its effectiveness reporting sits in four layers: service delivery, then search and AI visibility with per-keyword rank and brand mention rate, then visits and engagement where identifiable AI referral visits are counted apart from ordinary organic visits, and finally business results such as qualified enquiries. An AI click share belongs in that third layer and has to be evidenced by site analytics or logs, not inferred from visibility metrics alone. The same caution runs through Topkee's AI browsing analysis, which reads retrieval sources and cited pages and is not equal to actual visits. Topkee's Singapore company was established in October of 2024, and its SEO work is scoped to a site's baseline and its competitive gap and the workload involved, with no unconditional ranking guarantee.

What to do with the control this week

Start by reading your own numbers rather than copying Zyppy's.

Open the Generative AI performance report in Search Console, if your account has it, and note AI impressions as a share of total search impressions.
Compare that share with your overall click-through rate. A high impression share paired with a low site-wide click rate is the combination that makes AI clicks matter.
Pull landing pages from the Performance report for the last 28 days and match them against the pages you know AI answers cite.
If you run analytics, filter referral traffic for the AI features and record what is identifiable. Treat anything invisible to your tooling as unknown.
Decide whether an experiment is worth the risk of losing the halo effect, and set a fixed window before you start.

A consultant earns their fee here by fixing the definitions before the comparison. For a team that wants that work scoped properly, an independent seo consultant review of the same 28-day window is the cheaper first step than a live opt-out experiment you cannot reverse cleanly. Where the site competes on SEO in Singapore, a best seo singapore programme should say at the outset which of its reports cover AI placements.

Two colleagues agree on a measurement plan at a table with printed charts and a marked calendar.

Check AI referrals in analytics first

Before editorial work on new pages, make sure your analytics separates AI referral visits from organic ones. The guide to first-party data in growth planning is a useful starting point for setting the buckets, and the analysis of AI answer ad counts in apparel and electronics shows how differently commercial placements behave across categories.

What the UK rules may change

The UK's Competition and Markets Authority gives Google until Dec. 3 to provide publishers with impressions, clicks and click-through rate for AI search features, aggregated across features such as AI Overviews and AI Mode. Google has three more months after Dec. 3 to roll out page-level controls.

If those reporting duties arrive, the estimate stops being necessary in the UK. It also sets an expectation for what publishers elsewhere will ask for: a click figure from Google rather than a before-and-after calculation. Until then, a seo consultant working with a Singapore client has to build the estimate from site data and say plainly that it is one.

FAQ

Is the 21% figure something Search Console reports?

No. Search Console reports AI impressions in the Generative AI performance report, and it counts AI link clicks as ordinary clicks. The 21% came from comparing total clicks in an opt-out period with an opt-in period.

Does the Search generative AI control affect my Google rankings?

Google says the control is not used as a ranking or inclusion signal for other parts of Search. On Zyppy's site, the page stayed at number 1 in regular results while it was excluded from AI features.

How fast does the control take effect?

Google says content is excluded within 1-2 days after the control goes live, though caching can delay some pages. Zyppy's site dropped out of AI search within an hour.

Should a Singapore SME opt out to test?

Only with a fixed window and a clear measurement plan. The Zyppy result shows the halo effect can reduce regular listing clicks, so an opt-out test carries a real cost. Most teams get more from checking their AI impression share and referral traffic first.

Why does a low AI click-through rate still produce a large share of clicks?

Because the click rate is compared with the site's own baseline. At 0.9% against an overall 0.4%, AI links drew more than double, so a 10% impression share turns into roughly a fifth of all clicks.

Can I compare my site to Zyppy's 23%?

Not directly. AI visibility varies widely between publishers, and Shepard describes his figures as a directional yardstick. Use the method on your own site and keep your denominators consistent.