Short answer: in July 2026 Google published real numbers on its AI search tools: AI Mode has passed 1 billion monthly active users, the Gemini app has 950 million, and AI features in Search send “billions of clicks” to websites every week. The user and revenue figures are precise. The click figure has no baseline, no denominator and no comparison period, so it cannot tell a site owner whether its own traffic is rising or falling. The independent studies that do measure clicks per search, from Pew, Ahrefs, Seer and a randomized experiment, all find fewer organic clicks when an AI answer appears.

Both things can be true at once. Reading the two datasets together is more useful than picking a side, and it changes what you should measure.

What Google disclosed about its AI search tools

The richest source is Sundar Pichai’s Q2 2026 earnings remarks, published on 22 July 2026, with the revenue table in Alphabet’s quarterly filing. Strip out Cloud and hardware and the Search story comes down to five statements of very different kinds.

StatementFigureCan you check it?
AI Mode monthly active usersOver 1 billion, since AI Mode expanded globally last OctoberReported to investors, with no breakdown by country or query type
Gemini app monthly active users950 million, with daily users tripling in the last yearReported to investors, same limits
Google Search & other revenue$63,271 million in the quarter, against $54,190 million a year earlier, growth of 17%Yes: it is in Alphabet’s filing with the SEC
Clicks from AI features in Search“Billions” every weekNo: no exact figure, baseline or trend
Search queriesAI Mode is driving an “incremental increase” overall; usage hit an all-time high during the World CupNo growth rate given

Read down the right-hand column: the numbers that matter to investors are exact and audited, and the one that matters to publishers is the vaguest in the document. Google also said it has brought AI Overviews and AI Mode together into one seamless Search experience. For anyone working on visibility, that structural line matters more than the user counts, and we come back to it below.

Why “billions of clicks” cannot settle the argument

“Billions of clicks every week” is a numerator without a denominator. It does not say how many AI-feature impressions produced those clicks, how the number compares with a year ago, or how it is spread across sites. It is compatible with traffic rising for a few large publishers and falling for most others. And it says the volume is large, not that it went up.

Google’s earlier position, set out by Liz Reid in an August 2025 post, is that total organic click volume from Search to websites has been “relatively stable” year over year, and that Google is sending slightly more quality clicks, which it defines as clicks where users do not quickly click back. The post gives no figures for either claim. The implied argument is that AI answers remove low-value visits and keep the ones that matter: a reasonable hypothesis, not yet published with data.

Revenue is reported to the million dollars; clicks to the nearest order of magnitude. A site owner can plan with the first number and not with the second.

What independent AI search click studies measured

Studies that measure clicks per search point the same way, with different methods and different sizes. The table gives each one’s design, because the design decides how much weight a number can carry.

StudyDesignFinding
Pew Research Center, July 2025Browsing data from 900 U.S. adults in March 2025Users clicked a traditional result in 8% of visits when an AI summary appeared, against 15% without one, and clicked a link inside the summary in 1% of visits
Ahrefs, April 2025300,000 keywords, March 2024 against March 2025An AI Overview correlated with a 34.5% lower average clickthrough rate for the top-ranking page
Ahrefs update, February 2026Same design on December 2025 data58% lower; position-one CTR of 0.016 on AI Overview keywords, against 0.039 on informational keywords without one
Seer Interactive, April 202653 brands, 5.47M tracked queries, 2.43 billion organic impressionsOrganic CTR on AI Overview queries recovered from a floor of 1.3% in December 2025 to 2.4% in February 2026; queries without one reached 3.8%
Agarwal and Sen, 2026 working paperRandomized: a Chrome extension removed AI Overviews for some usersWhere an AI Overview appeared, outbound organic clicks fell 39.8% and zero-click searches rose 34.5%; sponsored clicks were unaffected

Pew’s result is the easiest to read, because all three of its click figures share one base: the share of visits that ended in a click. Put side by side, the drop is roughly half, and the clicks that AI summaries send through their own links are a small fraction of what they displace.

Google visits that ended in a click, March 2025 Share of visits; browsing data from 900 U.S. adults Clicked a traditional result, no AI summary on the page 15% Clicked a traditional result, AI summary on the page 8% Clicked a link inside the AI summary 1%
Figure 1. Among Google visits by Pew’s panel of 900 U.S. adults in March 2025, a traditional result was clicked in 15% of visits without an AI summary and 8% with one; a link inside the summary was clicked in 1%. All three bars use the same base, the share of visits. Source: Pew Research Center, July 2025.

The randomized study deserves the most weight on the causal question. Observational studies compare queries that trigger an AI Overview with queries that do not, and those differ in kind; randomly hiding the feature removes that objection. It is still a working paper by Saharsh Agarwal and Ananya Sen, already revised since first posted, so treat its exact figures as provisional.

Two of its findings speak directly to Google’s quality argument. Engagement on the sites people clicked through to was not different between the groups, and survey answers showed no measurable improvement in perceived search quality. If AI answers were filtering out poor visits, downstream engagement should have moved; here it did not. An exploratory AI Mode condition suggested even greater reductions in traffic to publishers, which is a direction rather than a size.

Ahrefs’ re-run shows how the gap widened over two years on one fixed measure, the clickthrough rate of the page ranked first. Treat it as a correlation, as Ahrefs does, but the direction is hard to miss.

Clickthrough rate of the page ranked first Ahrefs keyword studies; CTR as a fraction of impressions Keywords with an AI Overview, December 2023 0.073 Keywords with an AI Overview, December 2025 0.016 Informational keywords without an AI Overview 0.039
Figure 2. Position-one clickthrough rate for keywords that show an AI Overview fell from 0.073 in December 2023 to 0.016 in December 2025; informational keywords without an AI Overview stood at 0.039. Ahrefs reports these as correlations, not causal effects. Source: Ahrefs, February 2026.

The Seer recovery is worth noticing too. Click-through on AI Overview queries rose from its low point, which suggests part of the earlier drop was not permanent, but the gap against queries without an AI Overview stayed open. Seer describes its own data as directional rather than causal. If the vocabulary is new, our entry on zero-click searches explains how the term is used.

How both datasets can be true at once

Google says queries are rising and total click volume is roughly stable. Independent studies say each search that shows an AI answer produces fewer organic clicks. These statements do not contradict each other. More searches multiplied by fewer clicks per search can leave the total roughly flat, while an individual site loses traffic on exactly the queries where an AI answer appears. The publisher complaints and Google’s stability claim may be the same phenomenon measured at different altitudes.

How a stable total and fewer clicks per search fit together More searches AI Mode adds queries overall, says Google (earnings remarks) × Fewer clicks per search that shows an AI answer (Pew, Ahrefs, Seer, Agarwal and Sen) = A flat total Organic clicks “relatively stable” year over year (Google, August 2025) What one site sees: fewer visits on exactly the queries with an AI answer
Figure 3. A schematic, not a chart: more searches multiplied by fewer clicks per search can leave the total roughly flat while an individual site loses visits. Sources: Google’s Q2 2026 earnings remarks and Liz Reid’s August 2025 post for the totals; Pew, Ahrefs, Seer and Agarwal and Sen for clicks per search.

Revenue is a separate question again. Search & other revenue grew 17% in the quarter, and the randomized experiment found sponsored clicks unaffected by AI Overviews. Google’s financial results and the traffic that reaches independent sites are not measures of the same thing, and one should not be read as evidence about the other.

Search Console now shows AI search impressions, but not clicks

The most practical thing Google shipped this year is a report. On 3 June 2026 Google Search Central announced Search Generative AI performance reports in Search Console. They show how often your URLs appeared in generative AI features, including AI Overviews, AI Mode and generative features in Discover, broken down by page, country, device and date. The rollout began with a subset of sites and, according to Google’s note on the post, reached all websites worldwide as of August 31, 2026.

What the announcement does not list is a click metric. Google says it may add metrics over time. Until it does, you can see where you appear inside AI features but not what that appearance is worth in visits. Impressions make AI features look generous; clicks are the metric in dispute.

An opt-out that removes you entirely

Alongside the report, Google added a Search Console control that lets a site opt out of appearing in, and helping ground, AI Overviews, AI Mode and AI Overviews in Discover. Google says sites that opt out receive no traffic or impressions from those features, and that the control is not used as a ranking signal for results outside them. It started with a subset of site owners in the UK. Do not confuse it with Google-Extended, which is a separate control covered in our note on Gemini and Google-Extended.

Preferred Sources now reach AI answers

In August 2026 Google said people had already selected more than 600,000 unique sources as Preferred Sources, and that those sources are now surfaced in Top Stories, AI Overviews and AI Mode. That is a count of sources selected, and on its own it says nothing about how much visibility a selection brings. It is still the one lever in this article that a site’s own readers pull, rather than Google’s systems.

One experience, two citation sets

Google now presents AI Overviews and AI Mode as one experience, but they have not behaved as one source of citations. Across 540,000 query pairs, Ahrefs found the two cited the same URLs only 13.7% of the time, while their answers reached an average semantic similarity of 86%. Same conclusions, mostly different sources. That analysis compares single generations of each answer, so the overlap across repeated runs may be higher, but the safe working assumption is that appearing in one does not carry over to the other.

Google has also described AI Mode as using a query fan-out technique: it issues multiple related searches across subtopics and data sources at once before composing the answer. The practical consequence is that a page built around one broad head term competes with pages that answer a specific sub-question cleanly. Answer first, name things in full, and use headings that state the question. Writing for extraction shows what that looks like in practice.

What to do with Google’s AI search numbers

  1. Open the new Search Console report and sort by page. It is the only first-party view of where Google’s AI features surface your URLs. Keep a monthly export by page and country.
  2. Measure the answer side per surface and per engine. Search Console covers Google only, and only impressions. For other assistants, run a fixed set of buyer questions and record who is named and who is cited. Report each engine separately before any average, because engines differ enough that a blended number can describe none of them. The GEO measurement framework sets out the rest.
  3. Do not count being cited as traffic. Pew found clicks on links inside AI summaries in 1% of visits. Seer found that, on informational queries, being cited in an AI Overview brought +120% more organic clicks per impression than not being cited, yet still trailed queries with no AI Overview by -38%. A citation is worth having, but it is a visibility outcome first and a traffic outcome second.
  4. Decide on the opt-out deliberately. Opting out removes you from these features entirely. It suits a publisher that has decided the exchange no longer works for it; it is not a fix for falling clicks.
  5. Write for sub-questions. If AI Mode fans a question out into related searches, the unit you compete on is the sub-question, not the head term. How to appear in Google AI Overviews has the page-level checklist.

If you want those answer-side checks run on a fixed question set across engines, that is what InsightWonder’s AI visibility tracking is built to do.

Sources

An earlier version quoted stat-aggregator figures we could not trace to their producer; they were removed. Every number above, charts included, was checked against these primary sources.