A multimodal model, an always-on personal agent, and a redesign push the Gemini app past 900M users.
The Gemini app crossed 900 million monthly users on the back of Omni, a stronger multimodal model, and Spark, an always-on personal agent — alongside a redesign that makes asking the assistant the default. This is not just a usage milestone; it is a signal that a huge share of everyday questions now resolve inside an AI assistant rather than a search box, and increasingly get handled by an agent acting on the user’s behalf. For brands, answer-and-agent visibility has quietly become a real distribution channel, and 900 million users is the number that makes it impossible to ignore.
The headline is scale: the Gemini app now serves roughly 900 million monthly users, a figure that puts it among the most-used software on earth. But the more important story is what pushed it there. Omni brings stronger multimodal reasoning — the ability to understand and work across text, images, and more — while Spark introduces an always-on personal agent that can carry out tasks rather than merely answer questions. A redesign ties it together, making the assistant the natural first stop for a growing range of everyday needs.
Read together, these tell you the app is not just growing but changing what people use it for. A more capable multimodal model handles a wider variety of questions well, and an always-on agent shifts the app from a place you ask things to a place that does things. That combination — broad capability plus agentic action, at 900 million users — is what turns the Gemini app from a chatbot into a mainstream front door for both answering and acting, which is exactly why it matters for anyone thinking about visibility.
The trajectory here mirrors the whole industry’s. Assistants began as conversational tools that answered questions, useful but bounded, and they are steadily becoming agents that plan, retrieve, and execute multi-step tasks on a user’s behalf. Spark is Google’s expression of that shift within the Gemini app, and it reflects a broader bet that the future of the interface is an assistant that does not just tell you things but handles them — booking, comparing, gathering, deciding — with the user supervising rather than clicking through each step themselves.
This matters because an agent changes who is reading your content. When a user researches manually, they read your pages; when an agent does it for them, the agent reads your pages and decides what to surface or act on. That is a different audience with different needs — it wants clean, structured, unambiguous information it can parse and trust — and being legible to it is an extension of being citable. The Gemini app reaching 900 million users on an agentic redesign means this shift is arriving at mass scale, not in some distant future.
In practice, the Gemini app now blends answering and acting. For informational questions, it draws on Gemini’s reasoning and, where needed, live retrieval to compose an answer, often with sources. For tasks, Spark can take a goal and carry it out across steps, pulling in whatever information and actions it needs. The multimodal capability from Omni widens the inputs it can handle, so a photo, a screenshot, or a document can be the starting point of a query, not just typed text.
The mechanics matter because they define where a brand can appear. In the answering path, being one of the sources the assistant cites is the visibility that counts. In the agentic path, being the information or option the agent selects when carrying out a task is the equivalent — and it depends on your content being machine-readable and trustworthy enough for the agent to use confidently. A brand that is clear, structured, and credible is eligible in both paths; one that is not is skipped in both.
An agentic assistant compresses the discovery journey. Where a user once searched, compared several sources, and chose, an agent can do much of that internally and present a result — which means the moments where a brand could once have been encountered collapse into whatever the agent decides to surface. This is the zero-click dynamic taken a step further: not just an answer in place of links, but an agent that resolves the whole task, potentially without the user ever seeing the sources it consulted along the way.
For brands, that raises the stakes of being the source an assistant trusts, because the agent’s choice increasingly is the discovery. If Spark carries out a research or comparison task and never surfaces you, you were absent from that customer’s decision entirely, with even less visibility into why than a zero-click search offers. Being the citable, machine-legible, trustworthy source is how a brand stays present in an experience where an agent, not the user, does the looking — and at 900 million users, that experience is mainstream.
A huge share of everyday questions now resolve inside the Gemini app rather than a search box, and Spark means many are handled by an agent acting for the user. Being the source it answers with — or acts on — is a real distribution channel.
The core implication is that a mass-scale surface for both answering and acting now sits between your customers and their decisions, and much of the research that once happened on the open web happens inside it. Visibility in the Gemini app — being the source it cites in answers and the option it selects in agentic tasks — is a genuine channel, not a fringe experiment, precisely because 900 million users is not a niche. The research that shapes purchases is increasingly mediated by the assistant, and being absent from it is being absent from that research.
This makes the Gemini app a surface to manage deliberately, and because it draws on Google’s intelligence and retrieval, the fundamentals that win in Google’s answers largely apply. Auditing whether the assistant surfaces you for your priority questions, understanding who it surfaces instead, and closing the gap is the work. A 900-million-user milestone on an agentic redesign is a prompt to treat that work as real, because the assistant is where a growing share of your customers now go to ask — and to delegate.
For brands, the assistant’s growth rewards two things at once: being citable in answers and being legible to agents. The first is the familiar answer-layer discipline — answer-first, evidenced, structured, entity-clear content the assistant can quote and trust. The second is newer and increasingly important: content and data structured cleanly enough that an agent carrying out a task can parse, compare, and act on it confidently. Structured data, clear specifications, and unambiguous facts are what make a brand usable by an agent, not just readable by a person.
This dual requirement is an opportunity for brands willing to invest in genuine clarity, because both citability and agent-legibility reward the same underlying discipline: making your information clean, structured, evidenced, and trustworthy. The brands that do this are eligible across both the answering and acting paths of a 900-million-user assistant, while those with messy, thin, or unstructured content are skipped in both. As assistants become agents, being machine-usable is becoming as important as being human-readable.
Understanding how the assistant selects what to surface sharpens the approach. In the answering path, it favors sources that are relevant, authoritative, clearly structured, and safe to quote, weighting freshness for time-sensitive queries — the same signals that govern Google’s AI answers. In the agentic path, it needs information it can act on reliably, which puts an even higher premium on structure, accuracy, and machine-readability, because an agent making decisions on a user’s behalf cannot afford to act on ambiguous or unreliable data.
Evidence and entity clarity matter across both. Concrete, well-sourced content is easier for the assistant to trust and quote, and a clearly-defined entity is easier for it to attribute facts to and select confidently. For brands, this argues for content that is simultaneously answer-first and cleanly structured — readable as a quotable answer and parseable as reliable data. On an agentic assistant at mass scale, that combination is what keeps you present whether the assistant is answering a question or carrying out a task.
An agentic assistant at 900 million users redistributes visibility toward brands that are both citable and machine-legible. The winners are those whose content is clear, evidenced, structured, and trustworthy, because the assistant can both quote them in answers and rely on them in agentic tasks, amplifying their advantage across a mass audience. The losers are brands with thin, messy, or unstructured content and those relying on manual-research traffic that an agent now short-circuits, because an agent that resolves a task internally may never surface them at all.
The determining factor, as always, is not brand size but whether your information is the kind an intelligent assistant would rationally choose to answer with or act on. Brands that invest in genuine clarity and structure are rewarded across both paths; those that do not are increasingly invisible in an experience where an assistant does the looking. As delegation to agents grows, the gap between machine-legible brands and the rest will widen, which makes the investment in clean, trustworthy content more consequential over time.
The Gemini app now mediates a huge share of everyday questions — and increasingly acts on them. DUNkē tracks whether your brand is one of the sources Google’s AI answers cite, across its surfaces and seven other engines, so you know where you stand on the channel your customers ask by default.
The response to the Gemini app’s scale is to treat it as a real channel and run the fundamentals against it, with added attention to machine-legibility for the agentic future. Start by auditing whether Google’s AI answers surface you for your priority questions and who they surface instead — a baseline that tells you where you stand on the surface your customers increasingly use. That audit usually reveals gaps on exactly the high-intent questions where being the assistant’s chosen source matters most.
From there, the work is the familiar discipline plus a structural layer: rewrite key pages to lead with a clean, quotable, evidenced answer; add the structured data and clear specifications an agent can parse and act on; ensure your entity is unambiguously defined; and keep content fresh. Then measure whether your presence in the assistant’s answers moves and double down on what works. Preparing for both answering and acting is how a brand stays visible as the assistant grows and as delegation to agents becomes routine.
Situating the Gemini app among its peers clarifies priorities. Its defining advantages are scale — 900 million users is an enormous audience — and its grounding in Google’s intelligence and index, so classic search authority carries into its answers. ChatGPT reaches its own massive audience through a conversational habit and leans on Bing and earned presence; other assistants and agents are racing to add capability. Each rewards the same core qualities, but the Gemini app pairs mass reach with Google’s retrieval and an aggressive push into agentic action.
For brands, the takeaway is that you optimize for all of them with one set of strong foundations, tuned per surface, and that the shift toward agents makes machine-legibility a shared, growing requirement across all of them. The content an assistant wants to quote and the data an agent wants to act on are both cleaner, more structured, more trustworthy versions of your information. The Gemini app simply brings that requirement to mass scale first, which is one more reason to invest in genuine clarity now.
Assistants and agents pose a real measurement challenge worth planning for, because their answers — and their agentic choices — are even less visible than a zero-click search. When an agent carries out a task and surfaces a result, you may have no direct signal of which sources it consulted or why it chose one option over another. Traditional analytics, built around clicks and sessions, capture almost none of this, which means a brand can be steadily winning or losing in the assistant with no visibility into it at all.
The response is to measure what you can and infer the rest: track your citation share in the assistant’s answers where they are observable, watch branded search and direct demand as downstream signals of assistant-driven discovery, and monitor continuously because the behavior shifts silently. As delegation grows, this measurement gap will widen, which makes establishing whatever visibility you can into the assistant’s choices — and trending it over time — increasingly valuable for understanding your true position.
There are genuine uncertainties here. How reliably agents like Spark perform, how much users actually delegate versus continuing to research themselves, how the assistant selects sources and options in agentic tasks, and how transparent any of this becomes are all open questions. There are also broader concerns about how an agent-mediated web affects the traffic and discovery economics that brands and publishers depend on, since an agent that resolves tasks internally sends even less onward than a zero-click answer.
For brands, though, these uncertainties do not change the fundamental calculus. The Gemini app is where a huge share of customers now ask and increasingly delegate, and being citable and machine-legible pays off regardless of the details, because it rests on qualities every assistant and agent rewards. The concrete risk is not that the app changes; it is remaining invisible in an experience where an assistant, and increasingly an agent, does the looking. That risk is addressed by doing the work now and measuring what you can.
The developments to track are the ones that signal how central the assistant becomes to discovery: how far Spark’s agentic capabilities extend, how much users delegate rather than research manually, how the assistant surfaces sources and options in agentic tasks, and how its answering behavior matures. Each will tell you how much of your customers’ research and decision-making is mediated by the assistant and how the rules of being surfaced are taking shape. The through-line is that a 900-million-user assistant pushing into agentic action makes being the chosen source increasingly consequential.
For your own program, watch your presence in the assistant’s answers over time, per prompt and against competitors, as the best available measure of whether the surface your customers ask by default is naming you. The 900-million-user milestone is a reason to establish that baseline, and any movement in it is a signal to act — because on a mass-scale assistant that increasingly acts on its users’ behalf, being invisible is a cost that compounds as delegation grows.
The deepest way to read this milestone is as a marker on a long migration — from search, to assistant, to agent. Search meant scanning links; the assistant means receiving synthesized, sourced answers; the agent means delegating the whole task to something that acts on your behalf. The Gemini app reaching 900 million users on an agentic redesign is that migration arriving at mass scale, and it reframes what visibility means at each stage: from a rank in a list, to a citation in an answer, to being the source an agent chooses to act on.
That reframe is the strategic takeaway. As the interface moves from search to assistant to agent, the objective shifts from earning a rank, to earning a citation, to being the trusted, machine-legible source an agent selects. The Gemini app is pushing through all three stages at once, at mass scale, which means the brands that build genuinely citable and machine-usable content now are the ones positioned for each stage as it arrives — and 900 million users is the signal that the later stages are no longer hypothetical.
Omni’s multimodal reasoning widens the kinds of questions the assistant handles well, and that changes where brands can appear. A user can now start a query with a photo, a screenshot, or a document rather than typed words — pointing a camera at a product and asking what it is and where to buy it, or uploading an image and asking for comparable options. Each of these is a discovery moment that did not exist in a text-only search box, and being the source the assistant surfaces in response to a visual query is a new dimension of visibility.
For brands, this argues for making your products and information understandable in visual and multimodal contexts, not just textual ones — clear imagery, descriptive metadata, and structured data that connect a visual input to your offering. As multimodal queries grow, the brands whose information is legible across formats are eligible for a wider range of discovery moments, while those optimized purely for text lose ground on the visual and multimodal questions an Omni-powered assistant increasingly handles.
The agentic shift Spark represents makes machine-legibility a first-class concern, because an agent carrying out a task needs to parse and act on your information confidently. That means structured data describing your products and services, clear and unambiguous specifications, consistent and well-defined entity signals, and content organized so a machine can extract exactly what it needs. An agent comparing options or completing a task will favor the source it can read cleanly and trust, and skip the one it has to guess about.
This is an extension of the citability discipline, not a separate project: the same clarity, structure, and evidence that make you quotable in an answer make you usable by an agent. Brands that invest in genuinely clean, structured, trustworthy information are positioned for both the answering and the acting paths, while those with messy or ambiguous content are disadvantaged in both. As delegation to agents grows, preparing your content to be machine-usable becomes as important as making it human-readable.
At 900 million users, the assistant is becoming, for many people, the new homepage of the internet — the default place they start, the way a search engine once was. That reframes the competition for attention: instead of earning a visit to your site, you are earning a presence in the assistant that mediates the start of the journey. It is a more fundamental kind of visibility, because it concerns whether you are encountered at all, before any question of a click arises.
For brands, this means treating presence in the assistant as a strategic priority rather than an afterthought. The brands that are consistently surfaced by the assistant — cited in its answers, selected in its agentic tasks — hold a position at the new starting point of discovery, while those that are not are invisible from the outset. As the assistant becomes the default front door for hundreds of millions, being present within it is increasingly what it means to be found.
The Gemini app crossing 900 million users on Omni, Spark, and a redesign is more than a usage milestone; it is a signal that a huge share of everyday questions now resolve inside an AI assistant, and increasingly get handled by an agent acting on the user’s behalf. Answer-and-agent visibility — being the source the assistant cites and the option the agent selects — has become a real distribution channel, and 900 million users is the number that makes it undeniable.
The right response is to treat the assistant as the real channel it is: audit whether you are surfaced on the questions that matter, do the work of being both citable in answers and legible to agents, and track your presence as rigorously as you once tracked rankings. The brands that prepare for a world where an assistant — and increasingly an agent — does the looking are the ones that stay visible as that world arrives at mass scale. A 900-million-user milestone is simply the clearest signal that it already has.
“When 900 million people ask an assistant instead of a search box — and increasingly let an agent do the looking — being the source it answers with, or acts on, is the whole of being visible.” The Age’X Channel Desk
Get a free GEO audit — the same analysis behind every article here.