More than a quarter of Muse’s engaged shopping users had an average purchase size above $200, Meta CEO Mark Zuckerberg said during the company’s October 21 earnings call, according to TechCrunch’s earnings coverage. That is a potentially important commerce signal for a personal AI agent, although it describes a subset of shoppers rather than spending across the entire service.
Meta also said more than 10 million users had adopted Muse and that about 25% of people who set it up now use it daily. Shopping emerged as one of the prominent activities Zuckerberg discussed, alongside productivity tasks involving email and calendars. The disclosures do not establish that shopping is the largest use case by users, sessions or time spent.
These company-reported figures add substance to the adoption story. They also require separation from earlier third-party download estimates, and from Meta’s forecasts for revenue, privacy protections and eventual profitability.
Ten Million Users Is Not Ten Million Daily Users
The earlier milestone concerned installations. In its October 20 adoption report, TechCrunch cited Sensor Tower estimates that Muse had accumulated 10.9 million downloads, including 10.2 million in the United States, during its first 40 days.
Meta’s subsequent disclosure concerned more than 10 million “users.” The supplied earnings coverage does not establish whether that means registered accounts, completed setups, people active during a particular period or another definition. Nor does it specify whether the count uses the same geography as Sensor Tower’s estimate.
The figures therefore support different conclusions. Sensor Tower estimated the scale of distribution; Meta reported adoption from its own perspective. Their similar size does not make them interchangeable, and the company’s announcement does not independently validate every part of the download estimate.
The daily-use statement has another denominator: people who set up Muse. Zuckerberg said about 25% now use it daily. That suggests recurring engagement, but it is not a conventional retention curve showing how a fixed group behaves a week or month after signup. It also should not be mechanically multiplied by the headline adoption count to produce an official daily-active-user total.
Sensor Tower separately estimated that Muse reached 2.5 million daily active users on October 16. Its estimates also put U.S. retention after 14 days at 23%, compared with an average of 16% for ChatGPT, Claude and Grok. Those are useful external signals, but they remain third-party estimates measuring different things from Meta’s disclosures.
Together, the reports suggest more than launch curiosity. They do not yet supply a single, consistently defined account of acquisition, activation and sustained use.
The Shopping Figures Describe a Subset, Not Total Sales
The purchase-size disclosure is commercially interesting because it concerns spending associated with agent-assisted shopping, rather than downloads or conversations alone.
Its wording matters. Zuckerberg’s claim was that more than a quarter of engaged shopping users had an average purchase size above $200. It was not that the average Muse purchase exceeded $200, that a quarter of all Muse users bought something, or that every transaction made by those shoppers crossed that threshold.
The cited coverage does not provide the number of purchasing users, transaction frequency or total merchandise value. Without those denominators, a high purchase average cannot establish the size of the commerce business. A relatively small group making occasional expensive purchases could produce an encouraging statistic without generating substantial recurring revenue.
Zuckerberg also said Plaid had told Meta it had not seen comparable activity through its connector on another service. That is an executive recounting a partner’s observation, not a separately published Plaid transaction analysis.
Meta’s Muse launch announcement provides context for how shopping is supposed to work. The company says Muse can browse and complete tasks on a person’s behalf, requests approval before making a purchase, and supports checkout through Stripe’s Link, including one-time-use cards that conceal the underlying card details.
Those features describe a route from recommendation to payment. The earnings disclosure adds an early spending signal, but not enough information to calculate conversion, repeat purchasing or commission revenue.
Free Access Makes Commerce Central to Monetization
Meta’s stated business model connects those shopping signals to a potential revenue stream: keep Muse free for most users, retain a percentage of commerce conducted through the service, and charge power users for advanced models or more tokens.
That creates two distinct economic relationships. Commerce revenue would depend on transactions and whatever share Meta can collect. Paid access would depend on people valuing additional capabilities or usage enough to subscribe. The cited earnings coverage gives neither a commission rate nor specific paid-tier pricing, so it cannot support a revenue estimate.
This structure also changes what competitive success would look like. A personal agent could have fewer users than a general-purpose assistant yet become commercially valuable if those users regularly complete purchases through it. Conversely, a large installation base could be expensive to support if people rarely transact or pay.
Zuckerberg projected that Muse would eventually generate more revenue per user than Meta’s family of apps does today. That is a company forecast, not a demonstrated result. A meaningful comparison would need consistent user definitions, time periods and geographic treatment, along with a distinction between revenue and profit.
For readers assessing the AI competition, the useful disclosure is the proposed revenue mechanism. Meta is describing an agent that participates in transactions and sells additional computing access, rather than relying solely on the advertising model associated with its social platforms.
Current Privacy Controls Are Separate From Future Confidentiality
Zuckerberg presented users’ willingness to connect email and calendars as evidence of trust in Muse’s security. That is Meta’s interpretation of behavior. Connecting an account does not, by itself, establish that users understand the protections or that the system has passed an independent security assessment.
The launch announcement describes Muse Secure VM as a dedicated virtual computer containing the agent and a person’s connected data. Meta says a separate Sentinel agent checks actions before they reach the internet, while users control connected services and permissions. The company also says Muse requests approval for sensitive actions and provides an audit trail.
These controls address practical concerns about an agent taking actions through personal accounts. Their effectiveness still depends on implementation, and the adoption figures are not a security evaluation.
Meta makes two additional data-policy commitments in the launch post: Muse does not share conversations or data inside the virtual machine with Meta’s advertising systems, and users can opt out of having interactions used to train its AI models. Those are specific promises, not a blanket statement that no processing or retention occurs.
Sources
- TechCrunch’s earnings coveragetechcrunch.com
- October 20 adoption reporttechcrunch.com
- Muse launch announcementabout.fb.com
