Meta launched Muse, its personal AI agent, in September. On its Q3 earnings call on October 21, Meta said more than 10 million people in the US and Canada had set up a Muse agent in roughly six weeks, and that shopping is already its number one use case (Retailgentic has the full rundown). The detail that matters for brands and retailers sits underneath those numbers: the AI shopping agents Muse sends out are already retrieving product data, building shortlists, and fetching product and search pages at volume.
The sequence matters. An AI shopping agent retrieves product data first, builds a shortlist from what it can interpret, and only then fetches the product pages of the products that made the cut. A product whose data it could not read never gets the visit. For anyone who owns a product catalog, that makes this a mechanism to prepare for, not a trend to watch: product data has to be complete and structured at the point of retrieval.
The clearest signal comes from HUMAN Security, whose September 28 analysis of agent traffic is cited in Scot's post. In that sample, Muse accounted for 72% of observed agent requests, roughly five times the request volume of ChatGPT's agent.
Where those requests land shows why Agentic Commerce Optimization (ACO) is now critical work for any brand and retailer. HUMAN found that 84% of Meta's agent requests target shopping-oriented pages, meaning product pages and on-site search, and that traffic driven by Muse was 25% more likely to reach checkout than traffic from other agents. Reaching checkout is not the same as completing an order. But an agent that reads a product page and then proceeds to checkout has done exactly what a brand or retailer wants a shopper to do, and it did it on the strength of the product data it retrieved and read along the way.
Six weeks in, the volume is still early and the sample is one security vendor's view, but the direction is not ambiguous. Muse's agents are reading product pages at a scale no other open answer engine has produced this quickly, and a meaningful share of those reads end at a cart.
Meta is building Muse to transact, not to answer. On the Q3 call, Zuckerberg said Meta will “build commerce tools so we get paid a fee based on the transactions that we help facilitate,” with businesses paying that fee, and that more than a quarter of Muse’s weekly active users already shop through it at an average purchase above $200, travel included. Other answer engines can reach checkout too. The difference is that Meta’s revenue depends on the sale completing, so Meta has every incentive to make sure Muse is designed to carry the shopper to checkout rather than stop at an answer. HUMAN Security’s finding that Muse reaches checkout 25% more often than other agents fits that picture, and so does the strain on payments partners: Stripe raised credit limits in week one and Plaid said it had never seen anything like it. Purchase-ready product data matters on every surface with a checkout. What is different on Muse is how often sessions get that far, so price, availability, and variant data decide the outcome of more sessions here than anywhere else.
Muse is also bringing a different shopper. Gen X and Boomers lead Muse adoption, the reverse of every other answer engine, and Muse users prompt more often than ChatGPT users do, 9.2 times a day against 5.9, according to Gener8's panel data. The likely cause is distribution: Muse launched into Facebook's existing audience rather than the Gen Z and Millennial base that answer engines built first. The effect is that an older, value-oriented shopper asking direct questions about price, specifications, and fit is now arriving through an agent, in categories like consumer electronics and CPG where specification completeness decides inclusion.
Meta’s own numbers have now caught up with the traffic data, and more is coming: Shopify reports on November 3, Muse has not launched outside the US and Canada yet, and the take rate does not have a number. The brands and retailers that benefit from what comes next are the ones whose product data was ready before the agent went looking. Two things are specific to Muse and worth doing now.
Muse is not yet two months old, it is already the largest source of agent traffic on the open web, and Meta has said in plain terms that it will charge a fee on the transactions Muse facilitates. One more line from the call is worth sitting with. Meta’s CFO, Susan Li, said the company expects agents to move away from navigating websites the way a human does and toward authenticated APIs and approved partnerships. When that happens, the question for brands and retailers changes from whether an agent can read your product data to whether your product data is ready to be transacted with, under a protocol, with a take rate attached. The products Muse recommends today are the ones whose data was ready before it went looking. The products it sells next year will be the ones whose data is ready for that.