What a Muse Agent Sees When It Reads Your Store
Every review of the Muse agent is written for the person installing it. Is it safe, what can it reach, should you let it near your messages.
Nobody is writing the other half. Your customers are the ones installing it, and the software they install goes and reads your store.
The useful question is not whether Muse is impressive. It is whether an agent can buy from you, and that turns out to be decided by your platform rather than by anything you do this quarter.
What a Muse agent actually reads on your product page
“Muse is a personal AI agent. It doesn’t just answer questions, it actually does the work. It helps people stay on top of things, takes tasks and projects off their plate, and turns long‑term goals into action plans.”
Meta's own page for Muse says it browses the web, completes tasks and makes purchases. Put that against your product page and the question is what it has to work with when it lands.
Not your photography. Not your reviews, your badges, your carousel or the trust signals you paid a designer for. An agent reads the machine-readable layer and acts on it.
That layer is smaller than the page you designed. Your product data, your policy pages, your robots rules. As far as a Muse agent is concerned, that is the store.
How a Muse AI agent checks your price, stock and returns
A shopper forgives a stale price. They see 39 on the listing, 42 at checkout, shrug and carry on. An agent takes the first number as fact and acts on it.
Stock behaves the same way. An agent working from a feed row that says in stock keeps going until something stops it, and the thing that stops it is your checkout, after it has already told its owner the job is done.
Returns are the field least likely to exist as data at all. A shopper asks support or reads the policy page. An agent needs the term in a field it can compare, and a paragraph is not that.
This stopped being a matter of opinion in September. Shopify made Shop Pay a wallet inside Muse, and the eligibility list for agentic checkout requires catalogue inclusion and completed Terms, Privacy and Return policy pages. A missing returns policy is not a soft signal there. It is the difference between being purchasable and not.
“AI is playing an increasingly central role in that effort, including helping shoppers source information about a brand’s returns policy, leveraging AI-powered customer support, and personalized AI‑generated recommendations to streamline their return.”
What Muse from Meta changes for a store owner
The question got answered in about a fortnight, and not by a forecast. Meta lined up partners - Shopify, Walmart, GameStop, Sephora, Expedia, OpenTable. Amazon went the other way.
| Position | Who took it | What it means for a store like yours |
|---|---|---|
| Opted in | Shopify, plus named partners including Walmart, Sephora and GameStop | Purchasable today if you meet the eligibility list |
| Browsable, not buyable | Retailers that allow the crawl but block automated clicks | Found, then handed back to a human at checkout |
| Blocked | Amazon, from 20 September, unresolved | Not reachable, and the reasons are worth reading |
Notice what is not in the right-hand column. No integration, no vendor, no Muse-specific build. Your position was decided by your platform, and in Shopify's case it was decided by a default you did not set.
CNN tested it in September and got as far as a pickup cart and an approval prompt. On a retailer that blocks automated clicks the reviewer had to type in personal details by hand. So the picture is not an agent buying freely across the web. It is an agent that completes a purchase where the rails exist and stalls where they do not.
On Shopify, Muse checkout lives under Sales channels, then Agentic, and it is on by default for eligible stores in the US, Canada and Mexico. Whichever side of that you want to be on, it is worth being there deliberately.
Amazon gave three reasons, and the middle one should interest you more than the other two. It got no advance notice. The agent does not identify itself when it browses. And it appears to capture and store customer credentials.
Read the middle one again from your side of the transaction. If an agent does not announce itself, it is not in your logs as an agent, and every question about whether agents are reaching you becomes unanswerable from your own data. That is a measurement problem before it is a policy one.
Shopify answered the credentials question differently, by building a wallet so the agent never handles a card at all. Two companies, the same facts, opposite conclusions - which tells you this is a commercial negotiation about who owns the customer, dressed as a security review.
There is no side door either, and it is worth knowing before somebody sells you one. Shopify's Shop app sells from Shopify merchants. It will link an Amazon account, but only to pull in order records for delivery tracking, and you cannot buy an Amazon listing inside it. An agent reaching Shop Pay gets no closer to Amazon's catalogue, because the products are not there to buy.
Which leaves a split worth sitting with if you sell in both places. Amazon's own sellers are wiring themselves into Shopify in the meantime, through Marketplace Connect in one direction and Multichannel Fulfillment and Buy with Prime in the other. So the plumbing between the two companies is busier than ever, and none of it carries an agent.
Your Shopify storefront is agent-reachable today. The same products, listed on Amazon, are not. That is one catalogue, two answers, and you did not choose either of them.
That is the split, and it is the one your platform has already picked for you. Either the rails exist and the agent uses them, or they do not and the agent is left driving a checkout form built for hands.
Where the Muse agent gets your catalog wrong
Five fields decide it, and each fails in its own way.
- Price disagreement. Feed, page and markup have to say the same number at the same second, and they are updated by different systems on different schedules.
- Stock that is a guess. A feed refreshed nightly against inventory that moves hourly means the agent acts on last night.
- Shipping written for a human. "Usually ships in 2-3 days" cannot be compared. In a structured field it is a filter you pass.
- Return terms as prose. Named in Shopify's own eligibility requirements, which makes this the cheapest item on the list to get wrong.
- Attributes that do not match the question. Size, material, compatibility. A missing attribute is not a rejection, it is never being considered, and that leaves no trace you can go and look at.
The feed requirements differ per platform, and the page-level work is a separate job from the feed.
There is a second reason to care about the fields above, and it arrives at the worst possible moment. Before finalising a purchase, Muse shows its owner the merchant total, the shipping terms and the item details, and waits for approval.
So your shipping window and your return terms are not read quietly by a machine and forgotten. They are put in front of a buyer at the instant they decide, summarised by software you do not control. A wrong number reaches the customer as your answer, stated with confidence, and you never get to correct it.
Which also answers the question I would otherwise not be able to answer. You mostly cannot tell whether an agent read your page. Agent traffic arrives without a distinguishing referrer, and a product that was passed over generates no request at all. You check whether the page is legible rather than waiting for evidence that it was read.
How to audit your store for the Muse agent this week
An afternoon, no vendor call, no budget.
- On Shopify, open Sales channels then Agentic and read what is switched on. Then check your catalogue against the eligibility conditions: Terms, Privacy and Return policy pages complete, product types not subscription, bundle, customised or B2B-only.
- Open the feed file and read twenty rows. Not the dashboard summary, the file. Price, stock, shipping estimate and return window populated and current on every row.
- Run three product URLs through Google Rich Results Test at https://search.google.com/test/rich-results. A missing Product schema, or a price in the markup that disagrees with the page, gives an agent two answers.
- Read your robots.txt line by line. Write down every agent you block and why. A rule written when AI crawlers first arrived stays in force until somebody removes it.
- Buy from your own store in a private window, on a phone, not logged in. Count the steps that need a human hand. Each one is a place an agent stops, which is exactly what the CNN reviewer ran into.
What to fix first before Muse AI agents reach your category
Price agreement, then stock accuracy, then returns as a structured field. In that order, because each one is worthless while the one above it is wrong.
The reason is not Muse. Those three are the fields Google Shopping reads, the ones AI Overviews quote when they quote you, and the ones a marketplace listing needs. Fix them and you have improved four surfaces, one of which happens to be agent checkout.
And if you are on Shopify, the rails were switched on for you in September. The only question left is whether your catalogue is in a state to use them.
Getting products picked by shopping agents covers the catalogue side in more depth.
Run the agent-readiness checker on your store for a one-click read of what an agent sees, then work the audit above.
FAQ
What is a Muse agent?
Personal software that takes a goal from its owner and works across apps and the web to finish it. Meta's own page for Muse says it browses the web, completes tasks and makes purchases. For a store owner the useful framing is not what it is but what it does to you: it arrives at your product page with intent, reads the machine-readable layer and decides from that.
Is Muse AI safe to use?
That is the consumer question, and the merchant version is different. Your exposure is not the agent touching your site, it is your store being described wrongly to a buyer who never sees your page. A stale price or an empty return field reaches the customer as your answer, stated with confidence, and you get no chance to correct it.
Can a Muse agent buy from my store?
Test it rather than guess. An agent completes a purchase the way a determined shopper with no patience would, so buy from your own store in a private window, on a phone, not logged in, and count the steps that need a human hand. Each one is a place an agent stops.
How do I know if a Muse AI agent read my product page?
Mostly you cannot, and that is the awkward part. Agent traffic tends to arrive without a distinguishing referrer, and a product that was passed over generates no request at all. This is why the audit is worth more than the analytics here: you check whether the page is legible rather than waiting for evidence that it was read.
What does Muse from Meta see that a shopper does not?
Less, not more. It does not see your photography, your reviews or your trust badges. It reads your Product schema, your feed rows, your page text, your robots rules and your policy pages, and where two of those disagree it has no way to pick the right one.