Google is reporting a Gemini figure that would have sounded exceptional not long ago: in July, CEO Sundar Pichai put the Gemini app at 950 million monthly active users. At the same time, market studies circulate that place Gemini well behind ChatGPT or, depending on the method, show declining shares. That is not a contradiction another leaderboard can resolve. It shows how imprecise the word “user” has become in the AI race. To understand whether Gemini is catching up, the first questions are: user of what, over what period, and on which surface?
Key takeaways
- Google said the Gemini app had 950 million monthly active users in July 2026, while daily active users had tripled over the prior year.
- Market-share studies measure different things, such as app use, web traffic, or the distribution within a defined category.
- Large reach can come from product integration and by itself says little about daily use, willingness to pay, or satisfaction.
- For users and companies, the more important question is which assistant completes tasks reliably and how data, costs, and switching options are handled.
950 million is a statement, not a simple comparison
In its second-quarter earnings presentation, Google said the Gemini app had reached 950 million monthly active users. The company also pointed to sharply higher daily usage and Gemini’s deep integration across Cloud, Workspace, Android, and other products. The number matters: Google has a distribution machine that no standalone chatbot provider can replicate. A feature can become visible through an existing account, phone, and work suite without asking people to form an entirely new habit.
That is exactly why the number should not be read as a direct score. “Monthly active” does not mean someone works with the assistant every day for a long time. It also does not tell us whether that person opens the standalone Gemini app, uses a feature in another Google product, or interacts with the service once during the measurement period. Google reports different AI products in different places. AI Mode in Search, Gemini Enterprise, and the Gemini app are not interchangeable measures. Adding them together would create an impressive but editorially empty super-number.
The same problem applies when comparing ChatGPT and Claude. OpenAI, Google, and Anthropic do not always publish the same metric or report it on the same schedule. Weekly active users, monthly active users, installed apps, web visits, paid accounts, and processed tokens answer different questions. One high figure can mean reach, while another can indicate intensive work. Both matter, but they are not the same thing.
Market share depends on the measurement method
The Sensor Tower study summarized by TechCrunch described the worldwide AI-assistant market differently at the end of May: ChatGPT held 46.4 percent, Gemini 27.7 percent, and Claude 10.3 percent. The study measures a specific market and a specific kind of use, not the full influence of a foundation model across every Google service. It is useful as a snapshot, not as a final map of the AI economy.
Other datasets can reach very different results. Web traffic captures people visiting a website, but not necessarily usage in a mobile app or operating system. App analytics track downloads and sessions, yet may treat enterprise access and preinstalled features differently. Platforms that estimate which model generated text from writing patterns measure traces on the web rather than users. A recent report of a Gemini decline draws on such heterogeneous sources. Treating that as clear evidence of an exodus would put too much certainty on too little comparability.
It helps to look at incentives. Google can offer Gemini at many touchpoints, from Search to Workspace. Anthropic can appeal to specialists through productivity and coding. ChatGPT benefits from a strong, distinct brand and its own recognizable use moment. Those strengths look different in different measurements. Share of app time is not automatically share of business value, and a billion monthly actives are not automatically a billion loyal regulars.
How distribution changes the competition
Google does not need Gemini to win solely as an isolated destination. It can place the AI capability inside products people already use each day. That lowers the barrier to trying it and generates feedback at enormous scale. In the earnings presentation, Google also cited roughly nine million developers building monthly with its models in APIs and products, plus strong Gemini Enterprise usage among large companies. Reach, developer access, and infrastructure reinforce one another.
For competition, that means the race is not decided only by the best answer. It is about standards, data flows, accounts, file formats, and which assistant is already where work happens. That can be convenient, but it also creates lock-in. Connecting Gmail, Drive, Calendar, and documents can provide context and automation. At the same time, it makes clear permissions and the ability to change providers more important. Reporting on security risks around reasoning traces in major models is a reminder that reach increases the duty to build sound safeguards; it does not reduce it.
Users do not have to select one winner from this competition. Different tools can make sense for research, writing, programming, and personal organization. Anyone handling work data should check what a service may retain and which company settings apply before running a convenience test. Using one model only because it is prominently suggested on a phone turns the platform operator’s product decision into your own.
Outlook: Repeat use is what matters
Gemini’s growth is a strategic success for Google. The figure shows that the company can convert enormous reach into AI contacts. It does not prove that users have already made a final choice or that one model is better for every task. The Sensor Tower data also points to a market where people move between assistants and use multiple products side by side.
The better question, then, is not whether Gemini has “beaten” ChatGPT. It is which assistants become reliable habits and why. Anyone tracking that shift should keep four measures separate: reach, repeat use, willingness to pay, and success at concrete tasks. Only together do they say something about the real strength of an AI product. Everything else is a race for numbers in which each provider draws the finish line.