Twelve days. 1.8 million downloads on iOS across the US and Canada.
That is Meta Muse, and according to Apptopia data reported by TechCrunch, it beats the 1.3 million ChatGPT managed over the same window at its own mobile launch.
Worldwide installs across iOS and Android reached 2.8 million in the same 12 days. Daily active users in the US hit 642,000, against ChatGPT's 231,000 at the equivalent stage. Appfigures had it move from number 2 to number 1 on the US App Store.
Comparing a 2026 launch to a 2023 launch flatters the newer product, obviously. The market is warmed up and the category no longer needs explaining. That caveat does not make the number irrelevant, it just changes what the number means.
The distribution was already in the building
The detail that matters is buried at the end of the data. Over 95 percent of Muse users also have Facebook accounts, and 63 percent use Instagram.
This is not a product winning an audience. It is an audience being handed a product.
Meta has done this before, with stories, with reels, with marketplace. The playbook is not invention, it is placement, and it works often enough that treating it as a fluke is a mistake.
What it means commercially is that assistant usage is about to stop being a behaviour of early adopters and technical buyers. It arrives inside apps that a much broader population already opens forty times a day.
Your least technical customer segment is the one most likely to meet an AI assistant through Meta rather than by downloading something from a company they read about.
One more engine that does not read the same shelf
Every new assistant with real scale is a new answer engine with its own source preferences, its own refusals and its own idea of what your category is.
We have already seen how differently the existing ones behave. The citation mix in Gemini does not match Perplexity's, which does not match ChatGPT's.
Muse arrives inside an ecosystem with an enormous amount of first-party behavioural data and a commercial model built entirely on advertising. Guess which way its answers will eventually lean when there is money attached.
I am not predicting anything sinister. I am predicting the obvious: a company that makes its revenue from ad placement will find a way to make assistant surfaces monetisable, and the timeline for that will be shorter than the one for subscription-funded assistants.
We watched ChatGPT advertising reach a billion dollar run rate in 200 days. Meta starts this race with a fifteen-year head start on ad infrastructure and an existing relationship with every media buyer reading this.
The product lesson underneath the download number
Lenny Rachitsky's review of Muse called it the personal AI agent that gets consumer user experience right. That phrasing is doing a lot of work, and it is worth taking seriously.
Most assistants are still shaped like a developer tool wearing a chat interface. A blank box, infinite capability, no opinion about what you should do with it.
That design assumes the user arrives with intent. A large part of the consumer market does not. They arrive with a vague situation and want something to suggest the next move.
Meta has spent fifteen years building products for people who open an app with no plan. That competence transfers to this category far more directly than model quality does.
The commercial read for the rest of us: the winning assistant in a given segment will not be the most capable one, it will be the one that guesses intent best. That is a product discipline, and it is one most B2B companies are terrible at.
What to actually do in the next 30 days
Nothing dramatic. Three things, in order.
First, add Muse to your answer testing. The same 10 to 20 buyer questions you run through the other assistants, run through this one, and record what it cites. If your category answer there is built out of sources you have never heard of, that is your quarter defined.
Second, look at your Meta properties as source material rather than as campaign placements. If the assistant sits inside an ecosystem where your brand has a page, a catalogue and a review history, then that surface is now feeding answers, not just impressions.
Third, resist the urge to build anything. There will be pressure to launch a Muse presence, whatever that turns out to mean. Wait until there is a documented surface with documented rules.
The people who moved first on every previous Meta format mostly paid for the privilege of being a case study. The people who moved second, with data, generally did better.
The part nobody is pricing yet
A second mass-market assistant is not twice the work. It is a structural change in how much of discovery you can observe.
When there was one dominant answer engine, a brand could plausibly track its position in AI answers as a project. With two at consumer scale, and more coming, the number of surfaces where a buying decision starts and where you have no analytics at all keeps going up.
I wrote about the assistant stopping being a session and becoming a continuous relationship. Muse extends that to a population that never chose to have one.
Downloads are a vanity number. The number that matters is how many of your customers now ask a question to something that is not a search box.
That figure went up sharply this month, and nobody sent you a report about it.