Google Search Revenue Jumped 17%. Your Clicks? No Data.

News: On 22 July 2026, Alphabet reported its second-quarter results: the "Google Search & other" line reached $63.271 billion, against $54.190 billion a year earlier, a 17% increase. The same day, Sundar Pichai said Search is sending "billions of clicks to websites every week", without providing any basis for comparison (Alphabet, Q2 2026 earnings release, 22 July 2026).

Direct answer: Alphabet's second-quarter 2026 results show that Google Search advertising is doing very well ($63.3B, up 17%) even as AI features take up a growing share of the results page. Google publishes no verifiable metric on the traffic it returns to publishers. For a website, the practical consequence is simple: the link between "ranking well" and "receiving visits" is loosening, and the metric to track becomes presence inside generated answers.

What Alphabet reported on 22 July

The second-quarter 2026 earnings release, issued from Mountain View on 22 July 2026, gives clean numbers. Alphabet's total revenues stand at $119.796 billion, up 24% year over year. Google Services gains 15% to $94.5 billion. Google Cloud accelerates sharply to $24.768 billion, up 82%. YouTube advertising reaches $11.055 billion, up 13%.

And the line that matters to anyone running SEO, "Google Search & other", moves from $54.190 to $63.271 billion. In the release, Sundar Pichai sums up the dynamic in one sentence: "Our popular AI features are driving Search query growth."

The number Google gives, and the one it does not

In his remarks published on Google's official blog the same day, Sundar Pichai writes that the company is now sending "billions of clicks to websites every week" through AI features in Search.

Compare how the two statements are built. Search revenue comes with an exact amount, a reference quarter, a year-over-year change and a segment breakdown. Traffic returned to publishers comes as an absolute number, with no comparison year, no trend, no split between AI Overviews, AI Mode and classic links, and no indication of what it represents per site. "Billions of clicks" stays true whether the figure rises or halves. It is a statement no public data can contradict, and therefore a statement no public data can verify.

The only independent measurement available points the other way. The revised working paper by Saharsh Agarwal and Ananya Sen (SSRN, 2026) puts the drop in outbound clicks at 38% to 39.8% when an AI Overview is displayed, and notes that the lost clicks were not of lower quality. Both claims can coexist without contradiction: Pichai says queries are growing, so a total click volume can rise while clicks per query fall. That is exactly why the absolute total tells you nothing about your own situation. We flagged the same disconnect when AI citations started pointing to pages outside the top 10.

The real signal of the quarter: Google's AI got cheaper

One sentence in Pichai's remarks deserves more attention than the rest: "this quarter we reduced the cost of AI Mode responses to its lowest level since launch".

The unit cost of a generative answer is the factor that caps its deployment. As long as an AI answer is expensive, Google reserves it for queries where it pays off. Once it becomes cheap, nothing prevents extending it into the long tail, which is precisely the set of niche queries small and mid-sized businesses have built their organic visibility on for fifteen years. Worth recalling that AI Mode and AI Overviews were merged into a single experience at Google I/O on 19 May 2026, and that AI Mode had already passed one billion monthly active users by that date.

The consequence is structural, not cyclical. Every drop in cost per answer widens the surface where visibility depends on a citation rather than a ranking. It is the same shift observed in the concentration of AI referral traffic.

What to do now

1. Change your steering metric. Rankings and sessions are no longer enough. Measure your citation rate inside generated answers, drawing on the Search Console AI visibility report.
2. Produce what a model cannot generate on its own. Proprietary data, dated field findings, numbers only you hold. Synthesis content is already absorbed into the AI answer.
3. Structure for citation. Direct answer at the top of the page, named entities, verifiable sources, consistent markup. A generative engine cites what it can extract cleanly.
4. Cover your sector's news. Dated and sourced content holds up better, as shown by the Top Stories carousel inside AI Overviews.
5. Track share of voice, not just rank. In a market where the answer precedes the click, being cited three times across ten queries beats ranking first on one, as illustrated by share-of-voice measurement in AI Mode.

What this article does not cover

Alphabet's results are consolidated globally: they allow no reading by country, by sector, or by site size. Nothing in the release splits Search revenue between queries served with or without an AI answer, and Google does not publish the number of queries handled either. The "billions of clicks" claim is neither verifiable nor falsifiable as it stands, so we treat it as neither true nor false, only as unmeasurable. Finally, the 950 million figure covers the Gemini app, not the search engine: these are two distinct surfaces. We will update this article if Google publishes a numbered series on outbound traffic.

Our take

A $63.3 billion quarter settles one debate: AI in Search is not cannibalising Google's business model, it is reinforcing it. So the silence sits elsewhere, on the one variable that concerns you. When a company measures to the dollar what it collects and by order of magnitude what it redistributes, it is telling you where its priority lies. Build yours accordingly: produce content that AI needs to cite, rather than content it is content to summarise.

Frequently asked questions

Alexis Dollé

CEO & Founder

Growth and SEO content strategist, I founded Cicéro to help businesses build lasting organic visibility, on Google and in AI-generated answers alike. Every piece of content we produce is designed to convert, not just to exist.

LinkedIn