How Meta's Muse AI agent will impact ecommerce: seven effects on online stores.

On September 8, 2026, Meta launched Muse, a personal AI agent that can open a browser, find a product and pay for it once its owner says yes. Most coverage asked whether shoppers will trust it. This post asks how Muse will affect ecommerce if they do: what changes for the stores on the other side of the screen.

Updated October 31, 2026: this post now includes what happened in Muse's first month (Amazon's block, Shop Pay across Shopify stores, the launch in Canada and Meta's plan to take a fee on each sale), which we first covered in a separate post on October 9. That post now redirects here.

Facts about Muse are linked to the source we read. Anything marked “our analysis” or “prediction” is ours, and comes with the evidence that would prove it wrong.

What do we actually know about Muse?

We know how Muse is built and how it pays, and very little about how it chooses.

• What it is. Meta's announcement describes a personal agent that runs on a dedicated virtual machine with its own browser and can “open a browser, fill out forms, and negotiate” for its owner.
• Where it runs. It launched in the United States only, for adults, in a Muse app on iOS and Android, on the web and in WhatsApp chats, with a free tier and paid plans at $20 and $100 a month, according to TechCrunch and PBS News. It has since launched in Canada. As of early October those are the only two countries where it is available, The Next Web reports, and Meta has given no dates for others.
• How it reads a page. Meta's security write-up says the browsing sub-agent uses a real Chromium browser and receives “an accessibility tree snapshot of the page — not the raw DOM.”
• How it pays. Through Link, Stripe's wallet, which issues a single-use card number for each new merchant. At stores that do not accept Link, Stripe says it issues “a single-use virtual card scoped to the approved purchase.” Shop Pay, listed as coming soon at launch, is now live, and PayPal has been announced. Every purchase needs human approval “with the exact details of the purchase every time.”
• How it finds products. Shopify merchants are “discoverable and purchasable in Muse by default,” with product data flowing through Shopify Catalog, American Banker reports. Shopify's help center lists Meta as an AI channel that is on by default for eligible stores. A code inspection by Juozas Kaziukėnas, summarized by PPC Land, found two discovery paths (catalog or open web) and three checkout paths (browser, Shopify, Stripe Link). Meta has not confirmed that breakdown.
• Ads. Meta says Muse “doesn't share a person's conversations or the data in their VM with Meta's ad systems,” and it has announced no paid placement in Muse's results.

What happened in Muse's first month?

Four things changed for stores: the audience grew, Amazon said no, Shopify said yes, and Meta said how it plans to get paid.

• People downloaded it, helped by a lot of advertising. Sensor Tower estimates Muse passed 5 million downloads in 22 days across the United States and Canada, The Next Web reports. Meta gave Muse up to half of its own house-ad impressions for two weeks, so treat that as a measure of promotion as much as demand. Internal figures reported by The Information put weekly prompting at more than 3 million people and daily at more than 1 million. Meta has published no numbers of its own.
• Amazon shut the door. On September 20, Amazon blocked Muse from its store. Shoppers who try see a message that continued access by “an unauthorized AI agent” breaks Amazon's conditions of use. Amazon's complaints, in GeekWire's reporting summarized by PPC Land: it never agreed to Muse shopping there, the agent does not identify itself when it browses, and it appears to capture and store customer credentials. Meta says Muse “has no visibility into people's passwords or payment methods.”
• Shopify opened every door. A day later, Shopify's Tobias Lütke announced a partnership “to enable agentic checkout with Shop Pay,” PYMNTS reports. Meta's Alexandr Wang said Muse “can access the entire Shopify product catalog agentically.”
• Meta named its business model. At Connect on September 23, Mark Zuckerberg said Meta “will profit by taking a small fee from transactions,” MediaNama reports. The same report lists Walmart, Best Buy and Sephora as integrated, with Gap, Ulta, Wayfair, DICK'S Sporting Goods and others named on Meta's blog as connected. PPC Land says the fee reportedly comes from merchants, not shoppers.

What nobody has said. We found no merchant guidance from Meta, no published user-agent string or IP range for Muse's browser, and no statement on how it ranks stores or whether partner retailers are favored over everyone else. MediaNama notes Meta did not explain how Muse ranks products, or whether retailers pay to be in its catalog. Meta has not said how large the fee is, when it starts or who pays it. A correction to our own earlier shorthand: we have described Muse as the agent that reads your live store. That holds for the browser path, but Shopify stores and named retail partners now have a catalog path too.

How will Muse affect online stores?

In seven ways, as we read it. Each is explained below, with the evidence that would prove us wrong and what to do about it this quarter.

1. Stores sort into three lanes: retail partner, Shopify catalog or live browsing, and the lane decides how often Muse reaches you.
2. Your accessibility tree becomes your shop window, because that is what Muse reads.
3. The sale is decided on a one-line approval card, not on your product page.
4. Your own bot and fraud rules may turn Muse away before any ranking does.
5. Agent orders disappear into “direct” in your reports.
6. Meta plans a fee on each sale and sells no ads in Muse, so advertising moves upstream, into the feed.
7. The first order matters most, because Muse remembers what worked.

Will Muse treat every store the same way?

No. There are now three ways into Muse, and which one your store is in will matter more than anything on your home page.

Our analysis: there is a partner lane (named retailers such as Walmart and Sephora), a catalog lane (any Shopify store) and a browser lane (everyone else). The first two give the agent structured data and a checkout it knows will work. The third asks it to read a live page and operate a checkout it has never seen: slower and more fragile, but it works almost anywhere.

Prediction: an agent takes the route least likely to fail. When a partner or catalog result satisfies the request, it will rarely go on to browse the open web for an alternative, because every extra page costs time and risks failure. Stores in the browser lane will be reached mainly when the shopper names them.

• What would prove us wrong: repeated tests in which stores outside the catalog and partner list appear as often as stores inside them.
• This quarter: on Shopify, open Sales channels, then Agentic, confirm the Meta channel is on, and treat catalog data as a storefront. On any other platform, your product page is your feed. Make it fast and complete, as set out in what AI shopping agents actually read.

What does Muse see when it opens a product page?

It sees your accessibility tree, the same structure a screen reader uses, and that is a fact from Meta, not an inference. When we wrote why Muse reads your site like a customer, the link between accessibility and agent readiness was our argument. Meta's engineers have now said it outright.

Our analysis: UX debt becomes, in effect, a ranking factor. A price drawn inside an image, a color swatch with no name, an icon-only close button on a newsletter modal: none of these exists for a reader working from the accessibility tree. Call the work that follows conversion rate optimization for robots: the same discipline, for a visitor that cannot squint.

• What would prove us wrong: Meta moving Muse to screenshot-based vision as its main way of reading pages.
• This quarter: open your top product page in your browser's accessibility inspector. If the price, each variant, the stock state and the add-to-cart button are not there as named items, fix that before anything else.

Where does the shopper actually decide?

On the approval card, not on your product page. Meta requires a human yes for every purchase, with the exact details shown. The person approving may never see your photography, your typography or your brand film.

Our analysis: the offer that wins is the one that survives compression. “Waterproof jacket, medium, $184 delivered, arrives Thursday, free returns for 30 days” is easy to approve. “From $159, shipping calculated at checkout” is a question, and a question is a reason to say no. We expect delivered total, delivery date and return terms to decide more approvals than anything visual.

• What would prove us wrong: Meta letting merchants design their own rich approval cards.
• This quarter: write your offer as one sentence, then check that every part of that sentence is on the product page as text, before checkout begins.

Will your store recognize Muse as a customer, or block it?

Possibly not, and we think this is the most expensive near-term risk. Consider what a browser-path purchase looks like from your side: a Chromium browser in a data center, no browsing history, a card number never seen before, details typed by software. Brent W. Peterson, who writes on agentic commerce and is quoted by PPC Land, says it “looks like the spam bots they have been fighting for years.”

Amazon chose to block Muse, and it can afford to, because shoppers will go to Amazon anyway. Almost nobody else can. For a mid-size brand, a blocked agent does not send the shopper to your site. It sends the order to a competitor.

Our analysis: the risk for most stores is not choosing to block. It is blocking without choosing. Amazon says Muse does not identify itself, and we found no published way to identify its traffic, so your bot and fraud rules are deciding by default, with no label to go on. In our audit of twelve DTC stores, one store answered our honest automated visitor with an HTTP 403. A narrow finding, but it shows a default rule refusing a visitor without anyone choosing to.

Prediction: the first Muse problem most stores notice will be declined orders and challenged sessions, not lost rankings.

• What would prove us wrong: Meta publishing verifiable identification for Muse's browser, which would let bot-protection vendors recognize it within weeks.
• This quarter: ask your bot-protection and fraud providers, in writing, how they treat agent-driven sessions and single-use cards. Review declined orders from data-center IP addresses. Do not switch protection off; tune it.

What happens to attribution when the agent makes the last click?

It goes dark. Our analysis: a browser-path order arrives with no ad click ID, no returning-visitor cookie and a fresh card. A loyal customer who orders through Muse can look like a stranger, and last-click reports will file the revenue under “direct.” If Muse grows, some of your best customers will quietly disappear from your attribution.

What to measure instead is share of recommendation: how often agents name you when asked about your category. That is what our quarterly Agent Index tracks, under a published methodology.

• What would prove us wrong: Meta or Stripe giving merchants an order-level flag for agent purchases on every path.
• This quarter: add “an AI assistant” to your post-purchase “how did you hear about us?” survey, and record a monthly baseline by asking two agents your top category questions.

How will Meta make money from Muse, and can you pay to be recommended?

Through a fee on each sale and, as far as anything Meta has published shows, not through ads. Meta says it “will profit by taking a small fee from transactions,” and it has walled its agent's conversations off from the ad systems of a company that earns its living from purchase intent. You cannot pay to be recommended today.

Our analysis: the fee makes Muse a sales channel with a take rate: a marketplace in everything but name, and a new line in your margin next to payment fees and ad spend. It also raises the question Meta has not answered: does the fee affect what Muse recommends? An agent paid on some sales and not on others has a reason to prefer the first kind. We have no evidence that it does, but it is the reason to measure how often you are recommended instead of assuming.

With no ads inside the agent, ad value moves upstream. Meta's launch example is a recipe saved on Instagram becoming a grocery list. A save becomes a deferred task that an agent can carry out later. Content that names the product and the variant is executable. Content that only sets a mood is not. This is the social-to-agent funnel: discovery still happens in the feed, but evaluation and purchase can happen without a click.

Not everyone expects the no-ads position to last. An analysis by MBI Deep Dives, covered by PPC Land, argues Meta can keep ads out of Muse and still earn from what Muse learns. The same article notes a Digiday piece arguing ads in Muse look likely.

Prediction (low confidence): within a year Meta tests a labeled sponsored option inside Muse's results.

• What would prove us wrong: twelve months with no paid placement, or Muse losing the ability to act on saved posts. On the fee, Meta publishing how Muse ranks stores, with the fee shown to play no part.
• This quarter: brief creators to say and write the exact product name. Judge social content by saves as well as clicks. When Meta says how large the fee is and who pays it, put it in your channel margins next to payment fees and ad spend.

Does Muse favor big brands?

We think less than people assume on the first purchase, and more than they assume on the second. An agent has no brand habit. It has evidence: specifications, price, delivery, returns, reviews. A small brand with complete, consistent evidence can win the first order.

But Meta says Muse's memory “persists across your conversations,” according to its design post. Prediction: once an order goes well, “the usual” is stored and the reorder never returns to market. That makes the first agent-mediated order worth more than a human lifetime-value model suggests.

• What would prove us wrong: evidence that agents re-shop every order on price.
• This quarter: keep product names, URLs and SKUs stable, state subscription and reorder terms in plain text, and make sure the first order cannot fail at checkout. Our Audit and Fix services cover this path.

How is Muse different from other shopping agents?

Muse is the only one of the three types below that may operate your live checkout itself. The Muse column rests on the sources above; the other two are our general analysis.

What should online stores do about Muse now?

Four checks this week, then three fixes over the next 30 days. Work in this order, because each step protects the next. Human shoppers benefit from every item, whether or not Muse takes off.

1. This week: know your lane. On Shopify, confirm the Meta channel is switched on under Agentic storefronts. On anything else, your product page is your feed. Run the free check on your best-selling product page.
2. This week: open the door carefully. Ask your bot-protection and fraud providers, in writing, how they treat agent sessions and single-use cards. Do not turn protection off. Tune it.
3. This week: write your offer as one sentence with the delivered total, the delivery date and the return terms. Then check every part of it is on the product page as text.
4. This week: add “an AI assistant” to your “how did you hear about us?” survey. It is the cheapest attribution you will get.
5. Days 7 to 14: fix the accessibility tree. Name the price, variants, stock state and add-to-cart. Cut overlays to one labeled consent banner.
6. Days 14 to 21: reconcile. Check that price and stock match across catalog, schema and page.
7. Days 21 to 30: set a baseline. Record what two agents say about your category, and repeat it monthly. Our guide to getting your products recommended by ChatGPT and Gemini sets out the method and the fixes that follow.

What if Muse stays small?

It might. A month of data is thin. Downloads are not purchases, and nobody outside Meta knows how many orders Muse has placed. The usage figures above are outside estimates and a leak, not Meta's reporting. In a Vogue Business survey cited by Fortune, 31% of respondents said they would hand shopping to an AI agent.

But everything on the list above also helps the humans who shop with you. A clear price, a page that works without a pop-up in the way, and a checkout that does not turn away a real buyer are worth having whether the buyer is a person or their agent.

Written by Harry Harrison, founder of Recommended by Agents, under our editorial standards. Found an error? Tell us. Published September 21, 2026. Updated October 31, 2026. Meta, Muse, Shopify, Stripe, Amazon and other names are trademarks of their owners, and we have no affiliation with them. Spotted something out of date? Tell us and we will fix it.