Build5 publishers2 min readPublished
Meta gave Muse up to half its daily house-ad impressions on the way to 5 million US downloads
Sensor Tower estimates Meta's Muse agent hit 5 million US downloads in 22 days with up to half of Meta's daily house-ad impressions behind it. For builders, the launch shows how far Meta's own apps can carry an agent before anyone knows whether users keep it.
The Engineer · Build desk
What happened
- Muse launched on September 8 as what Meta calls a "personal AI agent", software built to act for a user instead of only replying to prompts.
- Muse held the top spot on the US App Store for 12 straight days.
- Meta also promoted Muse outside its own apps, and Sensor Tower ranked it among the top 15 US brands by ad impressions in the week of September 21.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint Download speed stops working as a measure of a consumer AI product's appeal when one entrant can fill its own house-ad slots with its launch.
- contradiction Three trackers disagree on the install count, so any estimate of how many downloads Meta's ads produced is built on a number nobody has audited.
- decision Agent builders without social apps of their own have a better yardstick than install curves: how many users connect accounts and authorize purchases after the first month.
House-ad impressions and downloads measure different things. One counts how often Meta chose to show Muse in its own house-ad slots [5]. The other counts one response to that exposure [1]. The reporting does not include a channel split showing what share of installs came from a Meta ad. Without it, the claim that most of the 5 million came from the push is an inference from timing.
The timing is close. The off-platform ad ranking in the week of September 21 came 13 days after the September 8 launch, and the 22-day run ends on September 30 [1].
The install count is itself a model output. Sensor Tower, Apptopia and Appfigures each produced a different cumulative estimate during the early launch, using different methods [11]. Taken at face value, 5 million in 22 days averages about 227,000 US installs a day [2].
The comparison with ChatGPT, Claude and Grok is a benchmark run on other people's launches. For it to say anything about product pull, the launches would need matching platforms, matching category familiarity and comparable promotion. ChatGPT arrived before consumer generative AI was a familiar category. Claude and Grok were iOS-only for parts of their early mobile launches, while Muse shipped on iOS and Android with Meta's promotion already running [12]. Sensor Tower noted those differences itself [13]. The apps that have matched Muse's pace are mostly games, by Sensor Tower's count of 23 that passed 5 million US downloads within 22 days [4].
The design Meta describes gets one default right. Meta says Muse can open websites, fill out forms, send emails, book travel and keep working after the app is closed [8]. It already supports payment workflows [10]. Before sending an email or making a purchase, it is designed to stop and ask for approval [9]. That gate is the right default for software that can spend a user's money.
I think calling distribution the moat is premature on this evidence. The numbers show Meta can put an agent in front of people who never went looking for one, through Facebook, Instagram, WhatsApp and its ad network [15]. The Neuron wrote: "The real test is whether Meta can turn a burst of attention into repeated delegation." [16]
What to watch
- Thirty- and ninety-day retention estimates for Muse from Sensor Tower, Apptopia or Appfigures.
- Muse's daily download rate once its share of Meta's house-ad impressions falls from the September peak.
- Whether the three trackers' cumulative estimates for Muse converge as the launch ages.