Gemini hits 1 billion users: 63% now talk instead of type

News — On 11 August 2026, Google announced that its Gemini app has passed one billion monthly users, and that 63% of them now speak to the model rather than type, with a growing share using voice only (Google official blog, 11 August 2026).

The number that matters to you is not the billion, it is the 63%. A billion monthly users puts Gemini among the major answer surfaces, alongside Search and ChatGPT. But the shift that changes your editorial work is the move to the spoken query. When someone dictates a question instead of typing it, they do not produce the same query, and the model does not look for the same thing inside your pages.

What Google actually announced

Google adds that the app has more than 100 million active users on iOS, that 38% of school-related requests include an attachment, and that Gemini now connects to more than 40 popular apps. On 12 August the company expanded that list of third-party services connected to the app.

This milestone extends a trajectory already visible in referral data: we covered the point at which Gemini overtook Perplexity on AI referral traffic.

Why "63% by voice" is not a product footnote

A typed query is frugal. People write "emergency plumber Lyon", three words, because typing costs effort. A dictated query is a full sentence, with a verb, a context and usually a constraint: "I need a plumber open this Sunday near Croix-Rousse, I have a leak under the sink".

That is not a difference of style, it is a difference of raw material. The second phrasing states explicitly the qualifying criteria the first left implicit: availability, neighbourhood, nature of the fault. The model therefore has more elements to match, and it looks for them in written passages, not in keyword density.

The second consequence is harsher. An answer read aloud cannot display ten blue links. The format forces a far narrower selection than a results page, and the gap between being cited first and being cited third moves closer to the gap between existing and not existing. This is the same movement we saw when Search Live rolled out across more than 200 countries, and when Gemini landed behind Siri, which we analysed at WWDC 2026.

What this changes for a small business

First, read your pages out loud. Ask a real customer question, in full, then look for the sentence on your page that answers it. If you cannot find it in ten seconds, a model will not find it either. This test costs ten minutes per page and tells you more than most automated audits.

Second, write the constraints, not just the promises. "Two-hour response across Lyon 1 and 4, seven days a week, excluding gas boilers" is a quotable sentence. "Your trusted plumbing expert" is not. The criteria you avoid writing down are exactly the ones the user just dictated into their question.

Third, instrument before you conclude. Assistant-driven traffic does not land in the right place in default reports. We set out how to isolate ChatGPT, Claude and Gemini traffic in GA4: without that step, you have no way of knowing whether this shift is costing you or paying you.

The limits of this reading

Four caveats, to stay honest. First, 63% of users does not mean 63% of queries. Google is reporting a share of users who talk to the model, not a share of voice interactions in total volume. Someone can dictate one question a week and type the other twenty.

Second, the Gemini app is not Google Search. A large part of that billion uses it to write, translate, code or sort photos, none of which ever touches your website. Treating those users as information seekers would be a shortcut.

Third, Google published no country breakdown, no paid subscriber count and no measurement methodology. Nothing in the announcement says what share of that billion is European, or how the threshold is computed. These are company figures, reported by the company itself.

Finally, this article does not cover voice assistant optimisation in the older sense of the term, nor speakable markup. The subject here is how you phrase your answers, not how you tag them.

The Cicero take

The usual reflex on this kind of announcement is to treat it as one more channel to optimise. The opposite is happening. The spoken query does not create a new playing field, it simply exposes an old flaw: most company pages have never answered a question, they have only described themselves.

While users typed three keywords, a vague page could still capture the click and let the visitor do the sorting. When the question is asked in full and the answer is read back, that margin disappears. You no longer win by being findable. You win by being answerable.

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