Meta’s Muse checks a shopper’s saved preferences, searches retailer websites and Meta’s product catalog, then filters the results into recommendations. A small study by Profound found that websites supplied most final choices, despite the catalog returning many more candidates.
Sensor Tower estimates put Muse above 10 million US downloads in 40 days, TechCrunch reported on October 20.
From dozens of products to a handful of recommendations
Profound’s October 2 research covered 50 prompts in one retail category, using a single Muse account. It reported these typical results:
| Source | Products found | Products recommended |
|---|---|---|
| Meta’s catalog | About 60 across 30 merchants | Two |
| Retailer websites | Four across two merchants, after visiting three domains | Four |
Catalog listings dominated retrieval, but website finds took more places in the shortlist, showing how those stages can diverge.
Profound, which sells AI marketing software, calls the sample exploratory; it does not establish how Muse weighs individual ranking factors.
The customer still reviews the purchase
Meta’s account of Muse’s design explains how the agent uses a full browser to navigate websites, fill in forms and complete transactions. The company built explicit approval controls for consequential actions, including purchases.
Its published checkout example shows a merchant’s product page inside an approval card, alongside a payment method and estimated total. The shopper can review the order and choose “Deny” or “Allow.” The card asks them to check the merchant’s details and terms before approving.
Access to the retailer can interrupt the process too. SEW previously reported on Amazon blocking Muse from buying on customers’ behalf, citing objections to how the agent accessed accounts. A shopper’s permission to proceed does not override the merchant’s access restrictions.
The shopper has an activity log that retailers may lack
Profound also reports that Muse’s browser traffic can resemble ordinary human visits, making agent activity difficult to identify on a merchant’s website.
On the customer’s side, Meta describes a more explicit record: tapping the Muse avatar opens an activity log and the permissions the user has approved. Those records offer a way to investigate a shopping attempt from the account that initiated it.
A retailer testing its own products could save the exact shopping request, recommendations and activity log, then match any approved order to its transaction records. That would let the retailer investigate specific failures, such as a product missing from the shortlist, a website access problem or a proposed order the shopper declined.
Start the conversation by posting the first comment