A site reported in Search Engine Journal opted out of Google's AI features - AI Overviews and AI Mode - specifically to estimate how much of its traffic those features were generating. The result: an estimated 21% of its total clicks were attributable to Google's AI surfaces. That is not a rounding error. That is a meaningful slice of organic performance that most measurement setups are currently failing to capture properly.
The method here is blunt but instructive. By removing the site from AI features and observing the traffic change, the site's team could back-calculate what AI was contributing. It is essentially a controlled subtraction. The fact that this kind of manual experiment is what practitioners are resorting to says a great deal about where measurement tooling currently sits.
Why This Number Matters More Than It Appears
Twenty-one percent is not a marginal effect. If you think about a site receiving, say, 50,000 clicks a month from Google Search, that figure implies roughly 10,500 of those clicks were coming via AI features rather than conventional blue-link results. Strip out AI Overviews and AI Mode and that traffic disappears - not gradually, but immediately upon opt-out.
What the experiment also hints at, though the site's methodology limits how precisely we can say this, is that AI features may provide a halo effect on conventional listings. When a site appears prominently in an AI Overview or AI Mode response, users who then scroll down and click an organic listing may already be primed towards that result. Separating genuine AI-referred clicks from that lift is genuinely difficult, and this experiment surfaces that complexity without fully resolving it.
The practical implication is straightforward: if you have not attempted to quantify what AI features are contributing to your organic performance, you are making strategy decisions with incomplete data. That includes decisions about content investment, AEO programme prioritisation, and how to frame the value of organic search to stakeholders.
The Opt-Out Method: Useful, But Not a Sustainable Measurement Strategy
Opting out of AI Overviews is possible via robots directives and the noai or nosippet meta tags, depending on what you are trying to block. The site in question appears to have used this approach as a deliberate diagnostic tool - take yourself out of AI features, measure the traffic delta, then draw conclusions about AI's contribution. It is creative and it generates a real signal.
But this is not something most businesses can do continuously or at scale. Opting out means forgoing real traffic during the measurement window. For a site where AI features are driving 21% of clicks, that is a meaningful revenue sacrifice. You are essentially running a live experiment with actual customers to get a data point your analytics platform should, in theory, already be giving you.
The more sustainable path is building proper AI referral tracking into your measurement infrastructure. Google Search Console now surfaces some AI Overview impression and click data. Beyond that, monitoring referral strings and UTM patterns associated with AI-referred sessions in GA4, and cross-referencing with CRM-attributed leads and pipeline, gives a picture that does not require you to turn off a channel to understand it.
What Sites With Strong AI Visibility Are Doing Differently
The site in this case was evidently receiving a substantial proportion of its clicks via AI features. That does not happen by accident. Sites that earn consistent citation and click volume in AI Overviews and AI Mode tend to share a set of structural characteristics: content that answers specific questions directly and concisely, strong entity associations that help AI systems understand what the site is authoritative on, and a technical setup that does not block AI crawlers.
Schema markup plays a role, particularly for structured content types - FAQs, how-to guides, product specifications, review aggregations. But schema alone does not explain a 21% AI click share. The more likely explanation is that the site had built topical depth and credibility in its niche such that AI systems were routinely drawing on its content when generating responses to relevant queries.
Digital PR and third-party citations feed into this too. When authoritative external sources reference a site's content or data, AI systems pick that up as a trust signal. It reinforces the probability that the site will be cited in AI responses. That is a different kind of link-building rationale from traditional SEO, and it requires a different production mindset - one focused on creating content that other credible sources want to reference.
The Attribution Problem This Experiment Exposes
The fact that a site had to opt out of AI features to estimate their contribution highlights a measurement gap that the industry has not fully closed. Standard GA4 setups do not neatly separate AI-referred traffic from organic search traffic. Google Search Console's AI Overview data is improving but still incomplete. And for businesses where the goal is revenue attribution rather than click counting, connecting an AI Overview citation to a CRM opportunity is still largely a manual or semi-manual process.
This matters because if you cannot attribute AI-referred leads to revenue, you cannot make a rational case for investing in AEO programme work. The 21% click share finding is striking, but clicks are not revenue. The question every commercial team should be asking is: what proportion of our pipeline-stage leads first arrived via an AI search surface? That requires UTM discipline, session source tagging, and CRM mapping that most organisations have not yet set up.
Start with what is measurable now. Tag AI-referred sessions where referral strings are identifiable. Use Google Search Console's AI Overview performance data as a baseline. Build a simple revenue bridge between those sessions and CRM-stage contacts. It will not be perfect immediately, but it will be more defensible than estimating by temporarily removing yourself from a channel.
The Strategic Read: AI Features Are Not Supplementary
The temptation when you see a figure like 21% is to treat it as interesting but secondary - a nice bonus on top of conventional organic. That framing is wrong. For the site in question, AI features were delivering more than a fifth of its total click volume. Remove that and you have a meaningful hole in traffic and, almost certainly, in pipeline.
If your AEO programme is still positioned internally as experimental or exploratory, this kind of evidence should change that conversation. AI search visibility is not an add-on to organic strategy. For many sites, it is already a primary traffic driver - they just do not have the measurement in place to prove it yet. That is the gap worth closing first.