How Meta Muse shops: what is documented and what is guessed

Meta has said how Muse pays and who fulfils the order. It has said nothing about how Muse decides which product goes on the card. Here is the line between what is documented, what testers saw, and what vendors are guessing.

Last updated: 26 September 2026.

Meta has not published how Muse ranks products. Documented: Muse shops through its browser, searches Shopify Catalog for Shopify stores, keeps conversations out of Meta's ad systems, and has no ad product announced. Everything else is observation or vendor inference.

What Meta has confirmed

Meta ships three things called Muse: Muse the agent, launched 8 September 2026; Muse Spark, the model underneath it; and Muse Image, the July image model. This post is about the agent.

The launch post says Muse "can open a browser, fill out forms, and negotiate", and "doesn't share a person's conversations or the data in their VM with Meta's ad systems" (Meta Newsroom, 8 September 2026). Connectors are apps the user chooses: "Connect your email, calendar, Instagram and more apps you use daily" (Meta, ai.meta.com/muse, read 26 September 2026).

On money: "All orders are fulfilled by the merchant that you purchase from" (Meta Help Center, updated about 23 September 2026). At Connect, Zuckerberg said Meta expects that "over time we will profit by taking a small fee from transactions" (Medianama, 24 September 2026). No primary source says who pays, at what rate, or from when, and he did not use the word advertising (Digiday, 25 September 2026).

Nothing from Meta says how Muse chooses between two products, whether it weights price, reviews, availability or brand, or whether a retailer can pay for position; Medianama filed both under "What Meta did not say" (Medianama, 24 September 2026). On social history: "A user's own Facebook, Instagram, WhatsApp or Threads history is not mentioned anywhere in the announcement" (Medianama, 9 September 2026). We track these gaps on Meta AI and Muse for brands.

The Shopify Catalog path is documented, the ranking is not

For Shopify stores, one more thing is on the record. Muse "searches Shopify Catalog to find the best match" (Shopify on X, 8 September 2026). Catalog carries "title, description, options, images, price, availability, and other key attributes", and relevance "also depends on other signals such as popularity, customer engagement, and brand recognition" (Shopify Help Center, read 26 September 2026).

Then the caveat that matters: AI channels "often re-rank results based on their own logic" (Shopify Help Center, read 26 September 2026). Shopify neither controls nor publishes what Muse does with those results. It also states "There are no fees associated with selling in Meta's direct checkout" (Shopify Help Center, read 26 September 2026). The admin switches and checkout exclusions are in Meta Muse for merchants. Stores off Shopify have no Catalog path; Muse reaches them by browsing.

What first-hand tests saw

Three patterns, none a ranking method. Muse browses and compares across merchant sites, including "OXO's own website and Walmart" once Amazon was closed to it (The Verge, 23 September 2026). It reads visible prices and discounts. It stops where a site stops it. In the four-agent test, Muse completed the Greenies order while another agent picked the wrong size variant (Retailgentic, 15 September 2026). The Amazon side is in Amazon blocks Meta Muse.

No tester reported anything labelled sponsored: "There are no ads in Muse" (aiadeconomy, 10 September 2026), and "nothing coming out of the company suggests otherwise" (Digiday, 25 September 2026).

What vendors are guessing

None of these is sourced to Meta.

Profound, Rankly, Nudge, Wildcard, Catalog and Swap Commerce had published no Muse-specific claim as of 26 September 2026. Agency talk of advertising "to ensure our brand is the one that gets picked by Muse" (Farhad Divecha, AccuraCast, via Digiday, 25 September 2026) describes what Meta might build, not what Muse does today.

How Meta's silence compares with ChatGPT and Google

On fees, OpenAI charged Shopify merchants 4% on ChatGPT checkout (The Information via PYMNTS, 21 January 2026) and retired it in March (CNBC, 24 March 2026); Meta's "small fee" has no payer, rate or date (Medianama, 24 September 2026). ChatGPT has told merchants what it weighs; Google, what it ingests and that it is testing paid placement; Meta, what Muse can do and nothing about weighting. Every Muse-specific optimisation method is therefore inference, whoever sells it.

What to fix regardless

These hold under any ranking, because they decide whether an agent can read and finish.

Fill the fields Catalog reads. Put size, shade and pack count in options, not in a paragraph; the wrong-variant order above is the failure this prevents. The general version is in optimising your product catalog for AI.

Make the visible price the real price. Testers saw stale prices (MBI Deep Dives, 9 September 2026) and a 404 (Retailgentic, 8 September 2026). Muse compares what it can see.

Make checkout finishable by an agent. The failures on record are a licence-number gate and a checkout that blocked autonomous clicks (CNN, 23 September 2026). Meta names two payment paths, a Link one-time card or the shopper's saved card via login; how each lands in your order record is in how Meta Muse pays. Enable agentic checkout is the product we run this on.

Count the orders you cannot see. Shopify attributes Meta-surface orders to Meta and fires no browser pixels in that checkout (Shopify Help Center, read 26 September 2026). They land in the reporting gap we measured for MyMuse; the mechanics are in dark agentic commerce traffic.

Do not buy a ranking nobody has. Treat any Muse-specific ranking method as a hypothesis until Meta documents one.

Where Tru Commerce fits

We have run this on Gray Porter and a group of US DTC brands: we made the product pages and checkout readable and finishable by an agent, then placed a real Muse order on the store. If you want the same run on your store, book a demo.