One week after it launched, Meta’s personal AI assistant Muse has climbed to the top of the U.S. App Store free chart, knocking ChatGPT into second place. The ranking, posted during the week of September 18, puts Google’s Gemini at No. 8, Anthropic’s Claude at No. 13, xAI’s Grok at No. 32, and Microsoft’s Copilot at No. 64.
Muse arrived on September 8 with a different premise from the chatbots that preceded it. Where ChatGPT and its rivals answer questions, Muse is built to act: it sends emails, books travel, fills out forms, and compares prices before placing an order on the user’s behalf. Meta’s bet is that the next phase of consumer AI belongs to agents that do things, not to assistants that talk.
The chart position is a striking reversal of the usual ordering in the AI race. On benchmark tests that measure raw model ability, Meta’s frontier models have consistently trailed those from OpenAI and Anthropic, according to the published evaluations the companies cite. On the consumer download chart, Meta now leads, a result that suggests distribution can beat raw capability when the product is built into platforms people already use.
Meta’s reach is the obvious explanation. The company can push Muse through Instagram, Facebook, and WhatsApp, three of the most downloaded and most used applications in the world, a channel that none of its rivals can match. OpenAI and Anthropic built their audiences from a standing start; Meta simply told its existing users the new assistant was there.
The company has spent the past two years preparing this ground. Meta open-sourced its Llama models and built its own AI assistant directly into its apps, and Mark Zuckerberg has argued publicly that Meta’s scale would make it the distribution leader in AI even if its models never won a benchmark. Muse is the clearest test of that argument yet, a standalone product that carries the strategy into direct competition with the chatbots that defined the category.
The launch has not come without friction. The Verge, reviewing Muse the same week, described the product as “creepy,” then added that the reason might not be what readers expected. The discomfort, the review suggested, lay less in what the assistant said than in how much it was willing to do without being asked, a line that captures the broader unease around agentic software.
That unease is the backdrop against which Muse’s early success will be judged. An assistant that books flights and fills forms needs access to accounts, credentials, and payment details, and every mistake it makes lands on the user who granted that access. The same capability that makes the product useful is the thing that makes its failures expensive.
Meta has framed the assistant as a step toward a broader agent platform, one in which users delegate routine tasks and the software handles the mechanics. The company has said little in public about how it will handle the liability questions that come with an agent that acts in a user’s name, questions that will sharpen as the product moves beyond the early adopters who dominate the download chart.
The App Store ranking measures downloads, not usage, and downloads are a trailing indicator of curiosity rather than habit. Whether Muse holds its position will depend on whether the people who installed it keep opening it, a test that the first-week chart position does not answer. ChatGPT’s own rise was followed by a long stretch in which most users who downloaded it drifted away before returning when the product improved.
What the ranking does establish is that Meta can move users faster than its rivals can move models. The company spent years training its own frontier models and open-sourcing them, a strategy that kept it in the conversation without ever clearly leading it. With Muse, it has chosen to compete on the surface where users actually are, and for one week that surface belonged to Meta.
The agentic turn that Muse represents is spreading across the industry. OpenAI has shipped an operator-style tool that can browse and act on the web, and Anthropic has built computer-use capabilities into its models, both aimed at the same shift from answering to doing. Meta’s advantage is not that it invented the idea, but that it can hand it to billions of people through apps they already open every day, a distribution edge that the chart position has now made visible.
The contest is far from settled. OpenAI’s ChatGPT remains the most widely used AI product by most measures, and its lead in paying subscribers and enterprise deals is substantial. But the chart position shows that the consumer market for AI is still being decided week by week, and that the company with the best model has not yet locked in the company with the most users.
