If you want one number: the Gemini app reached 950 million monthly active users, a figure Sundar Pichai stated on Alphabet’s second-quarter 2026 earnings call on 22 July 2026. That is the answer to the question most people are asking when they search for Google Gemini statistics, and it comes from the company rather than from an aggregator quoting an aggregator.
The reason to care about provenance here is that this subject has become a content farm. Search the phrase and the first page fills with pages carrying confident figures and no source, often contradicting each other by hundreds of millions of users. Every number below is attributed to a dated, primary disclosure, and where a figure is survey data rather than company data, that is stated.
The 950 Million Number Is the Gemini App, Not Gemini
The single most common error in circulation is treating one figure as the total size of Google’s AI business. It is not. It is the monthly active user count for the Gemini consumer app, disclosed for the quarter ending June 2026.
For context on the trajectory, Pichai had put the same metric at over 750 million users in the prior reporting period. Growth of roughly 200 million monthly users in a quarter is the kind of number that invites skepticism, and the honest caveat is that Google has not published a methodology for how it counts a monthly active user of the app, nor whether use triggered by an Android system integration counts the same as a deliberate visit. That caveat applies to every large consumer AI number from every vendor, which is exactly why cross-company comparisons are weaker than they look.
Checking this yourself takes about four minutes and is worth doing once, because it inoculates you against the next aggregator page. Google publishes the chief executive’s prepared earnings remarks on its own blog each quarter, and Alphabet’s investor relations site carries the call itself. Those two pages contain almost every Gemini figure in circulation, in Pichai’s own words, with the quarter attached. When a statistics roundup gives you a number you cannot find in either place, the number either came from somewhere else that should have been named, or it came from nowhere. The second case is more common than the first.
AI Mode Passed a Billion Monthly Users
The larger figure, and the one with more consequence for anyone who publishes on the web, is AI Mode. Pichai said on the same call that it had surpassed 1 billion monthly active users, measured since its global expansion the previous October.
AI Mode is not the app. It is a search experience inside Google Search, and it works through query fan-out: a single question triggers several related searches behind the scenes, which Google’s own documentation says lets it “display a wider and more diverse set of helpful links” than classic Search. That mechanism is the reason a billion-user surface behaves unlike a chatbot from a visibility standpoint. It is still issuing searches. It is still reading an index.
Alphabet reported revenue growth of 24% year over year that quarter, with Search and Other up 17%. Pichai’s claim has been consistent across several calls: AI Overviews and AI Mode are increasing total query volume rather than cannibalising it. Google has never published the data behind that claim, so treat it as a company position rather than an established fact.
AI Mode is also not the only new surface absorbing queries. Google said more than 140 million users engaged with Ask YouTube on the watch page in June 2026 alone, a feature that answers questions about a video without leaving it. Gemini Enterprise sits inside workplace tools. Antigravity sits inside a development environment. The pattern is that Google is not building one assistant people visit; it is putting model access at the point where a question already occurs. That is strategically duller than a chatbot war and considerably harder to compete with, and it means the number of places your brand can be mentioned or omitted keeps growing without any of them looking like a search engine.
Daily Use Tripled, Which Matters More Than the Monthly Count
Buried under the headline figure is a better one. Daily active users of the Gemini app tripled over the preceding year.
Monthly actives measure curiosity. Daily actives measure habit, and habit is what turns a product into infrastructure. A tripling of daily use against a roughly doubling of monthly use means the ratio between them improved, which is the signature of a product people came back to rather than tried once. Google also said daily users creating videos in the app rose 40% after the Omni launch at I/O in May 2026, which is a narrow metric but a real one.
Token Volume Is the Only Honest Measure of Depth
User counts can be inflated by a default setting. Token throughput cannot, because somebody is paying for the compute.
Google’s model APIs processed approximately 22 billion tokens per minute in Q2 2026, up from 16 billion a quarter earlier. That is a 37% increase in three months in the volume of text actually being run through the models by paying customers and products.
Two enterprise figures sharpen it further. Nearly 500 Google Cloud customers each processed more than 1 trillion tokens over the trailing year, and more than 2,000 enterprises consumed over 100 billion tokens in the same period. Those are not pilot numbers. A trillion tokens is production workload.
If you are assembling a slide on AI adoption and you want one figure that resists the usual objections, use token volume rather than user count. It is harder to game and it is denominated in money.
Two honest limits on the metric. Token volume says nothing about what the tokens were used for, so a surge could be reasoning traces, retrieval context, or batch document processing rather than anything a consumer would recognise as using Gemini. And tokens per minute is a throughput figure, not a cumulative one, so it cannot be annualised into a tidy total without inventing an assumption about load patterns across the day. Use it for what it is good for, which is demonstrating that paid production use is growing fast, and resist the urge to convert it into a per-user figure. Several roundups do exactly that and the arithmetic is meaningless.
The Developer Base Is the Quietest Moat
More than 9 million developers build each month with Google’s models across its APIs and main developer products.
The open-weights side is larger than most people assume. Gemma, Google’s open model family, has passed 900 million total downloads, with Gemma 4 accounting for more than 300 million of those since launching in April 2026. Google’s Agent Development Kit reached nearly 70 million total downloads in the quarter. Antigravity, its agentic development environment, reported more than 2.4 million weekly active users.
None of these numbers appear in consumer coverage because they are boring, and they are the ones that compound. A developer who ships on a model family does not switch casually.
Enterprise Adoption Is Where the Revenue Actually Sits
Google Cloud revenue grew 82% year over year in Q2 2026, with backlog reaching $514 billion. Pichai said nearly 90% of the Fortune 100 use Gemini Enterprise, that customer acquisition velocity had more than doubled year over year, and that existing customers were exceeding their contracted commitments by more than 50%.
Marketplace transactions grew more than sevenfold. On the security side, 90% of the Fortune 100 are named as Google Cloud Security users, and AI workloads scanned by that platform rose more than 45% quarter over quarter.
The $514 billion backlog deserves a note of caution for anyone quoting it. Backlog is contracted future revenue, not recognised revenue, and its conversion depends on data-centre capacity that both Google and its competitors have said is constrained. It is a demand signal, not a bank balance.
The security figures round out the enterprise picture and get almost no coverage. Google says 90% of the Fortune 100 are Google Cloud Security users, that nearly 90% of Wiz customers use AI-powered security features, and that AI workloads scanned and protected by the platform rose more than 45% quarter over quarter. On infrastructure, it claims its Axion processor delivers 30% better performance per dollar than peer offerings, which is the kind of unit-economics claim that decides whether the token growth above is profitable or subsidised. Nobody outside Google can verify that figure, and it is the single number in the set with the most riding on it.
Benchmark Wins and Share Figures Need Different Treatment
Two categories of number get mixed into Gemini statistics roundups and should not be.
The first is benchmark performance. Gemini 3 Pro scored 37.5% on Humanity’s Last Exam, against 26.5% for GPT-5.1 and 13.7% for Claude Sonnet 4.5, as reported by Fast Company and cited by EMARKETER. Benchmarks measure a model on a test. They predict very little about which assistant a given person opens tomorrow, and the gap between benchmark leadership and market leadership has been wide in both directions for three years.
The second is platform preference survey data. One Higher Visibility survey cited by EMARKETER in November 2025 put ChatGPT’s share of US search platform preference at 12.5%, up from 4.1% in 2024, with Google Search at 66.9%. That is self-reported preference from a panel, not measured query share, and it does not isolate Gemini at all. Quoting it as market share is a category error that appears constantly.
There is a third category worth naming, because it is where most readers actually land. The top results for Google Gemini statistics are aggregator pages that cite each other in a loop. While writing this piece, pages on the first page of results disagreed with one another by hundreds of millions of monthly users, attributed figures to quarters in which they were not stated, and in several cases gave no source at all for numbers presented as current. None of them were lying on purpose. They were copying, and a copied number loses its date on the first pass and its denominator on the second.
The usable rule is that any AI adoption statistic needs three attributes before you repeat it in a deck or a pitch: the entity that disclosed it, the date of disclosure, and the exact thing being counted. A figure missing any one of those three is not a statistic. It is a vibe with a decimal point.
The Three-Surface Rule for Reading Any Gemini Statistic
Here is the filter worth keeping. Every Gemini number belongs to exactly one of three surfaces, and conflating them is how bad analysis gets made. Call it the Three-Surface Rule.
The first surface is the app: the 950 million monthly and the tripled daily actives. This is a destination product. People go to it on purpose, and it competes with ChatGPT for the same habit.
The second surface is search: AI Mode at over a billion monthly actives, running query fan-out against Google’s index. This is not a destination. It is the default behaviour of a product people already used, which is why its user count exceeded the app’s inside a year.
The third surface is the platform: 22 billion tokens per minute, 9 million monthly developers, 900 million Gemma downloads, 90% of the Fortune 100 on Gemini Enterprise. This surface generates revenue and almost no headlines.
Most Google Gemini statistics in circulation fail at this first step, because the headline figure gets detached from the product it describes. Before using any figure, name its surface. A statistic about the app tells you nothing about whether your content gets cited. A statistic about AI Mode tells you a great deal.
What These Numbers Mean If You Want to Be Cited
The temptation, looking at a billion monthly AI Mode users, is to go looking for a technical trick. Google’s own documentation closes that door explicitly, and recently.
Its AI features page, last updated 10 December 2025, states that there are no additional requirements to appear in AI Overviews or AI Mode, and no special schema.org structured data to add, beyond being indexed and eligible to be shown as a snippet. Its generative-AI optimization guide, updated 10 July 2026, goes further: you do not need to create machine-readable files, AI text files, or markup, and “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” The guide tells readers in plain terms to check third-party advice on this subject against Google’s own.
What the documentation does endorse is narrow and unglamorous: be crawlable, make structured data match the visible text on the page, and understand that the nosnippet family of controls and the Google-Extended token govern what gets used. Recrawling after changes can take days to months.
The working conclusion from Instant Press client work across both surfaces is that the citation problem is an authority and corroboration problem wearing a technical costume. AI Mode fans a query out across an index and selects from what is there. If the only page in the world describing your expertise is your own homepage, there is nothing for a fan-out to find. If twenty indexed, independent pages describe the same expertise consistently, there is.
That is a less exciting conclusion than a technical fix, and it happens to match where the measurable work sits. Of the thousand-plus outlets in the Instant Press publication inventory, around 96% are Google-indexed, mean domain authority sits at 62.3, and 174 of them score above 80. The reason those figures matter to an AI-visibility question rather than a link-building one is that indexation is the precondition for being available to a query fan-out at all. A mention on a page Google has not indexed is invisible to AI Mode in the same way it is invisible to classic search, and no amount of markup changes that.
The practical sequence, then, runs in the opposite order from how it is usually sold. Get the factual claims about your business consistent across pages you do not control. Make sure those pages are indexed. Confirm your own site is crawlable and that your structured data matches the visible text, which is the one thing Google’s AI features page does name as best practice. Only then worry about which assistant is winning.
The numbers above will all be stale within a quarter, and the thing worth watching in the next set is not the app’s user count but whether token throughput keeps climbing faster than it does.