“Billions of AI Clicks”: Google’s Claim vs Your Missing Data

Google’s “billions of clicks” claim about AI search landed on Friday, July 17, 2026, when Nick Fox — Google’s head of search — wrote on LinkedIn that AI features in Search alone now send billions of clicks to websites every week. It is the first time Google has attached an absolute number to AI-feature-specific traffic, and it arrived with no baseline, no denominator, and no methodology.

What makes the claim remarkable is not its size — it is that no publisher on earth can check it. Google’s own Search Console report for generative AI features exposes impressions, pages, countries, devices, and dates. It does not expose clicks. The one metric Google chose to publicize is the one metric its own verification tool was built without.

This guide separates what Google actually said from what it can support, lays the competing click claims of this news cycle into a single verification grid, reconciles the seemingly contradictory numbers from independent research, and ends with the instrumentation plan we run for clients who need real answers while the official data stays dark.

The primary source is a LinkedIn post from Nick Fox, Google’s head of search, published Friday, July 17, 2026, and first reported by Search Engine Land. Fox wrote: “We’re now sending billions of clicks to websites every week through AI features in Search alone.” In the same post he framed the design intent directly: “We’ve designed our AI features in Search to connect people to websites.”

Crucially, Fox made two distinct claims that most coverage has blurred together — and per Search Engine Roundtable’s follow-up, they operate at different scopes and different cadences:

The rollout was coordinated. Liz Reid, VP of Google Search, reshared Fox’s post on LinkedIn, adding: “It’s awesome to see how our AI features are helping people explore the web and connect with the creators and publishers they love.” Fox followed up on X — as transcribed by Search Engine Roundtable, since X posts are not independently fetchable — saying: “We’ve been continuing to improve how we show links in our AI features in Search, to make it easy for people to click to websites.”

Note what neither executive offered: an exact figure, a comparison to any prior period, or a definition of what counts as an AI-feature click. “Billions per week” is compatible with AI traffic growing, shrinking, or collapsing — it depends entirely on the denominator Google declined to share. Any specific number you see attached to this claim elsewhere is invented; Google gave none.

Here is the structural problem. Google’s Search Console added a “generative AI features performance” report — the official surface for publishers to monitor visibility in AI Overviews and AI Mode. Per Search Engine Roundtable’s coverage of the rollout (June 23, updated July 9, 2026), that report exposes exactly five dimensions: impressions, pages, countries, devices, and dates. Clicks are not among them.

Search Engine Journal’s Matt G. Southern put the consequence plainly in his July 18 analysis: “The AI reports give you no click number to check [this] against.” SEJ’s summary of the wider criticism is just as direct: “Neither figure includes a baseline, denominator, or methodology, and Google had not published the data behind the weekly number at the time of publication.”

The report itself is still reaching sites. John Mueller of Google Search Relations, asked about the rollout on Bluesky, said: “Yeah, nothing super exciting. We’re just rolling these out incrementally to sites, and reviewing the feedback along the way.” Google has separately said it is continuing to work with website owners to understand which insights would be most helpful — phrasing that implies click data has not yet been judged necessary. If your property already has the report, our companion walkthrough covers how to actually track AI Mode traffic in Search Console with the fields that do exist.

To see how unusual this situation is, lay the competing click claims of this news cycle side by side and ask one question of each: can a site owner check it in Search Console today?

Four claims, four different sources, and not one of them is checkable in the tool Google positions as the publisher’s window into AI search. That asymmetry — aggregate assertions from the platform, structural silence at the site level — is the actual story of this news cycle, and it is the reason the rest of this post focuses on instrumentation rather than outrage.

Why publish an unverifiable number now? Barry Schwartz of Search Engine Roundtable offered one reading — and it is worth being precise that this is Schwartz’s own speculation, relayed via his X post, not a confirmed Google motive: he suggested Fox’s post was reactive, noting that a New York Times piece titled “Google Is Building an A.I. Fence Around the Internet It Once Championed” was reportedly in progress at the time.

What is documented is Google’s pattern on this topic. Search Engine Roundtable’s archive records Google denying AI-Overview-driven click declines multiple times: claiming clicks are stable with AI Overviews, calling third-party CTR studies flawed, and asserting that web publishing is not dead. The “billions of clicks” post is the affirmative version of the same defense — a positive assertion in place of a denial, still without published data.

Alongside the messaging, Google has shipped real product changes aimed at increasing click-throughs from AI features: preferred sources controls, Top Stories surfaced inside AI Overviews, recipe-specific link improvements in AI Mode, and general link-display improvements across both surfaces. We covered the most consequential of those — the news carousel — in our same-day analysis of Top Stories inside AI Overviews. The product work is genuine; it just does not substitute for verifiable click data.

The strongest independent evidence in this debate is a randomized field experiment by Saharsh Agarwal (Indian School of Business) and Ananya Sen (Carnegie Mellon University), published as an SSRN working paper. The design matters: 1,065 U.S. participants recruited via Prolific, with a Chrome extension randomly assigning each user to an AI-Overviews-on or AI-Overviews-off condition. That is a true randomized experiment — not an observational correlation that Google could dismiss as confounded.

The results: on queries where an AI Overview appears, organic clicks to external websites fell by roughly 38% (reported as 39.8% in some write-ups of the same study), and zero-click searches rose by about 34.5% — without materially changing sponsored clicks or how often people searched. In other words, the ad business and the search habit held steady while the open web absorbed the loss.

Survey-adjacent data points the same direction. Pew Research Center figures cited in Forbes’s coverage of the claim found users clicked through to traditional web results on roughly 8% of visits to results pages showing an AI summary, versus 15% without one — and ended their browsing session entirely on 26% of summary pages versus 16% of non-summary pages. We cite those numbers via Forbes’s framing rather than a primary Pew document, so treat them as directional corroboration rather than a precise benchmark.

None of these figures is Google-confirmed, and Google has disputed third-party CTR studies before. But that is precisely the point: when the platform publishes no data, third-party measurement — with its stated methods and limitations — is the only falsifiable evidence anyone has.

Most coverage treats “billions of clicks” (Google), “−38% on AIO queries” (the field experiment), and “68% zero-click” (SparkToro) as contradictory. Our read is that all three can be simultaneously true, because each operates at a different level of aggregation. Google’s claim is a whole-platform numerator with no denominator: across trillions of annual searches, even a sharply reduced click-through rate still yields billions of clicks in absolute terms. The experiment is a per-query effect: on the specific queries where an Overview appears, clicks drop by more than a third. SparkToro’s estimate is a share-of-searches metric: what fraction of all searches end with no click at all. A big absolute number, a negative per-query effect, and a rising zero-click share describe the same system from three angles.

That is also why Fox’s framing is rhetorically effective and empirically empty. A fixed, huge numerator says nothing about the trend line, and nothing about any individual site. Per SparkToro’s June 2026 zero-click study — built on Similarweb panel data for U.S. searches, January through April 2026, and a modeled estimate rather than Google-confirmed measurement — 68.01% of Google searches end without any click, up from 60.45% in 2024. That 7.56-point jump in two years is the fastest acceleration SparkToro has recorded since it began tracking the metric. And of the clicks that do happen, only about 27.6% land on the open web, versus roughly 66% two years prior.

SparkToro’s same methodology estimates that AI Overviews now appear on more than 20% of Google searches, correlating with roughly a 60% CTR drop on queries where one is present — again, SparkToro’s own modeled figures, not Google-confirmed, and worth labeling as such every time they are cited. If you want the adjacent question — which AI surfaces actually send measurable referral visits — we broke that down separately in our analysis of how AI referral share compares across Gemini and ChatGPT; this post stays on the verification question.

The publisher-side numbers are forecasts, and should be read as forecasts. Press Gazette projects that Google Search traffic to leading UK publishers is set to halve by Q3 2027 — a forward projection flagged the same day by Search Engine Roundtable, not a measured outcome, and one that could be wrong in either direction. But it lands in a news cycle where the platform’s answer to falling-traffic reports is an unverifiable aggregate, which is exactly why forecasts like it get traction.

Looking forward, we expect this asymmetry to widen before it narrows. Google is publicly “continuing to work with website owners to understand what insights will be most helpful” for the AI report — language that signals more dimensions may come, but commits to nothing on clicks. Our projection: aggregate reassurance will keep arriving faster than site-level verification, because the incentive runs one way — a clicks column in Search Console would let every publisher falsify the next “billions” post with their own export. Plan your measurement stack assuming the official click data does not come, and treat any future arrival as upside.

The productive response to an unverifiable claim is not to argue with it — it is to build measurement you own. Three levers are available to every site today, and none of them depends on Google adding a clicks column. Together they triangulate the question Fox’s post raises and Search Console cannot answer: is AI search visibility producing value for this site?

The first lever has a full setup guide in our companion playbook on GA4’s AI Assistant channel for measuring AI traffic properly — we will not restate it here. The point of running all three together is triangulation: landed traffic, latent demand, and upstream presence. Any one alone can mislead; the three in combination give you a site-level answer that does not depend on trusting or doubting Google’s aggregate. This is the instrumentation layer we build inside our analytics engagements, and the presence-tracking half is core to how our agentic SEO service treats AI surfaces as a distribution channel rather than a mystery.