Google Just Published Real Numbers on Its AI Search Tools.

Google spent two years telling everyone that AI in Search was going fine. In July 2026 it finally put a few real numbers behind that claim, and the interesting part is which numbers it chose.

What Google published, and when

There were really two disclosures, five days apart, and they were aimed at different audiences. On 17 July 2026, Nick Fox, the senior vice president who runs Knowledge and Information at Google, posted publicly that Google's AI features in Search send billions of clicks to websites every week, and that Search overall still sends billions of clicks daily. Liz Reid reshared it. That was for publishers and the search industry, and it was the first time Google had ever attached a volume figure specifically to its AI surfaces rather than to Search as a whole.

Then on 22 July, Sundar Pichai read the quarterly script to investors. That one is the richer document, because earnings calls come with legal exposure and the numbers in them tend to be the ones Google is most willing to defend. Alphabet's own transcript confirms revenue up 24 percent year over year, Search and Other up 17 percent, Cloud up 82 percent, and, buried in the Search paragraph, the milestone the search industry had been waiting on since AI Mode went global.

Worth separating the two, because they get quoted interchangeably and they are not the same class of statement. The earnings figures are audited. The click figure is a marketing assertion with no methodology attached. Both are true statements from Google. Only one of them can be checked.

The five numbers that matter

Strip out the cloud and hardware material and the Search story compresses into a handful of figures. Here they are with the source and the date, because half the confusion in this debate comes from people quoting a May statistic in July as though nothing moved.

The Gemini app number is the sleeper here. At 950 million monthly users it is within striking distance of AI Mode, and Pichai noted daily active users tripled over the year, which is a much better signal of habit than the monthly count. A lot of brand visibility work still treats the Gemini app as a side channel. On these numbers it is not a side channel.

AI Mode at one billion users

AI Mode went global last October. Nine months later it has more monthly users than most countries have people. Pichai's framing was that AI features are driving an expansionary moment, that AI Mode is producing an incremental increase in total Search queries rather than cannibalising them, and that Search usage hit an all time high during the World Cup. Whether or not the causal story holds, the scale is not in dispute.

The structural change announced alongside it matters more for anyone doing visibility work. Google has merged AI Overviews and AI Mode into what it calls one seamless Search experience. For two years those were treated as separate surfaces with separate optimisation logic, and there was decent evidence for treating them that way: Ahrefs found across 540,000 query pairs that AI Mode and AI Overviews cite the same URLs only 13.7 percent of the time despite reaching the same conclusions 86 percent of the time. Same conclusions, almost entirely different citations.

AI Mode also works differently at the retrieval layer. It uses query fan out, meaning a single question gets decomposed into multiple related searches across subtopics before anything is synthesised. That is why content built to rank for one head term often fails to show up in AI Mode while thinner pages that happen to answer a specific sub question get pulled in. The unit of competition stopped being the keyword some time ago and became the sub question.

The billions of clicks claim

Now the part that generated the arguing. Fox said AI features in Search send billions of clicks weekly. He also repeated the older line that Search overall sends billions of clicks daily. Both are billions. Both lack an exact figure, a baseline, a denominator, or a year over year comparison.

Search Engine Journal made the sharpest version of this point when it noted that Google gave precise revenue figures and click claims nobody can check in the same week. Revenue to the decimal. Clicks to the nearest order of magnitude. A company that can report 63.27 billion dollars can report a click count.

The other thing to notice is what Google is and is not asserting. Fox claimed the absolute volume is large. He did not claim it went up. That distinction gets lost every time the quote is recycled. Billions of clicks a week across the entire web is compatible with a handful of very large publishers absorbing most of them while everyone else loses ground, and nothing in the statement rules that scenario out.

Google's quality argument has been consistent since August 2025, when Reid wrote that total organic click volume was relatively stable year over year and that average click quality had slightly increased, with quality defined as visits where users do not immediately click back. The implied model is that AI answers strip out bounce traffic and leave the visits that were going to convert anyway. It is a reasonable hypothesis. It has never been accompanied by data.

What outside measurement found

Independent work landed on a different answer, and it got more rigorous over time rather than less. The most important study of the past year came from Saharsh Agarwal at the Indian School of Business and Ananya Sen at Carnegie Mellon. They built a Chrome extension that hid AI Overviews from a randomly assigned group of 1,065 desktop users and showed them normally to everyone else, then collected data from 7 January to 10 February 2026.

Randomisation is the whole point. Every earlier study compared queries that happened to trigger an AI Overview against queries that did not, which leaves an obvious objection: those query sets are different in kind. This design removed that objection. Outbound organic clicks fell 39.8 percent when an AI Overview was shown, zero click searches rose 34.5 percent, and clicks on sponsored results stayed flat. The damage sits entirely in organic. Ads were untouched.

The finding that should end the quality argument is the one nobody quotes. Hiding AI Overviews did not reduce user satisfaction, did not lower perceived information quality, and did not change downstream engagement on the sites people landed on. If AI answers were filtering out low intent visits and passing through better ones, downstream engagement should have moved. It did not.

Ahrefs re run is the one that shows the trend hardening. Its April 2025 pass over 300,000 keywords put the hit to top ranking pages at 34.5 percent. Its February 2026 update, using December 2025 data, measured position one click through at 0.016 with an AI Overview present against 0.039 without, a 58 percent reduction. The cost of an AI Overview roughly doubled as the feature matured.

Seer Interactive ran the longest series, tracking 2.43 billion impressions across 53 brands and 5.47 million queries. Organic click through on AI Overview queries fell from 1.76 percent in June 2024 to 0.61 percent by September 2025, then recovered to 2.4 percent by February 2026. That rebound is real and it deserves attention, since it suggests the floor was a rollout artefact rather than a permanent state. The gap against non AI Overview queries stayed open.

Pew Research reached the same neighbourhood by a different route in July 2025: users clicked a traditional result 8 percent of the time when an AI summary was present against 15 percent when it was not, and clicked a link inside the summary itself only 1 percent of the time. That last figure gets overlooked. Being cited inside an AI answer is worth a great deal for brand recall and almost nothing for direct referral traffic.

Search Console starts reporting AI visibility

The most useful thing Google shipped this year was not a stat, it was a report. On 3 June 2026 the company launched Search Generative AI performance reports in Search Console, giving site owners a dedicated view of impressions inside AI Overviews, AI Mode, and generative features in Discover. Data starts 18 May 2026 with no retroactive fill. The initial rollout went to a subset of UK properties, with access widening from 23 June.

The shape of that omission is telling. Impressions are the metric that makes AI features look generous. Clicks are the metric under dispute. Releasing one without the other lets Google answer the transparency complaint without settling the traffic argument, and site owners are left with a visibility number they cannot convert into a business case.

A control shipped alongside it: a toggle that removes a site's content from AI Overviews, AI Mode, and generative Discover without harming normal organic ranking. That existed because the UK Competition and Markets Authority forced it in June 2026, the first regulator anywhere to require a separate opt out. Most sites should leave it alone. It exists for publishers who have concluded the exchange no longer works for them.

Discovery starts in classic Search and finishes inside the conversation.

Reading both datasets at once

The temptation is to pick a side. Google is lying, or the studies are flawed. Neither is necessary, because both sets of numbers can be true simultaneously and mostly are.

More queries, fewer clicks per query, and a redistribution of whatever clicks remain. A rising denominator with a falling rate produces roughly flat aggregate volume, which is exactly what Reid claims, while individual sites lose ground, which is exactly what their analytics show. The publisher complaints and Google's stability claim are not contradictions. They are the same phenomenon measured at different altitudes.

Ad revenue is the third leg. Search and Other grew 17 percent to 63.3 billion while organic click through on AI Overview queries collapsed. Google can grow ad revenue on fewer, more expensive clicks for a long stretch before any strain appears in the reported numbers, and the randomized experiment noted that sponsored clicks were unaffected by AI Overviews. Financial health and ecosystem health are separate questions.

What changes for anyone who wants to be cited

The practical consequence of a merged Search experience with 13.7 percent citation overlap is that visibility has to be tracked per surface. A brand can hold position three on classic organic, appear in most AI Overviews for its category, and be entirely absent from AI Mode, and no single tool in a standard marketing stack will reveal that. This is the specific gap InsightWonder was built to close: prompt level tracking of whether a brand actually surfaces across AI Overviews, AI Mode, Gemini, and the other answer engines, rather than inferring it from rank.

Some documented signals worth acting on. Google extended Preferred Sources into AI Overviews and AI Mode on 27 May 2026 and reported that click through for labeled links roughly doubled, with more than 345,000 unique sources selected by users. Muck Rack's May 2026 analysis found 84 percent of AI citations come from earned editorial coverage in third party publications rather than brand owned pages or paid placement. Read together, those two point the same direction: getting written about elsewhere now outperforms writing about yourself.

Query fan out changes what a page needs to look like. Since AI Mode decomposes a question into sub questions before retrieval, a page organised around one broad topic competes badly against a page that answers a specific sub question cleanly and early. Direct answer first, then the elaboration. That is not a stylistic preference, it is a retrieval mechanic, and it is why the answer block sits at the top of this article rather than the bottom.

Also worth adjusting: what gets counted as success. If the report Google shipped only exposes impressions, then impressions inside AI answers are becoming the honest leading indicator, with brand search volume and direct traffic as the lagging ones. Anyone still judging content purely on organic sessions is measuring a channel that is structurally shrinking while ignoring the one that replaced it.

Questions people keep asking

Sources

Figures reflect what had been published as of 30 July 2026. Google has changed AI Overview triggering behaviour repeatedly, so coverage percentages in particular move quickly.