How AI shopping agents like Meta's Muse choose brands

How do AI shopping agents choose brands?

When a shopper names a brand, the agent’s job is to find it and buy it. When the request is generic, the agent picks. That second case is where brands win or lose, and we found nothing public that explains how Muse makes the pick.

Compare two instructions. “Reorder the dishwasher tablets I always buy, same brand” leaves the agent to find a store and a price. “Get me some dishwasher tablets” leaves it to choose the brand, the pack size and the retailer, from the stores it can check out at.

A shopper who names you gives the agent an errand. A shopper who doesn’t gives it a decision.

There is early evidence on how agents decide when they get a decision. L.E.K. Consulting points out that a search engine shows a list of results, where every brand gets a chance, while an AI tool returns a short list or a single recommendation. A 2025 study by Amine Allouah and colleagues studied AI agents acting as shoppers and found that they concentrated demand on a few products and showed strong position biases. Model updates could reshuffle market shares, and in its tests agents penalised sponsored tags.

Which AI agents can shop for you?

Meta’s Muse is the one built for consumers, and it is already buying at scale. Three always-on agents launched within seven weeks of each other in 2026: xAI’s Grok Bot on 11 August, which Reworked reported pauses for approval before consequential actions such as buying; Muse on 8 September; and OpenAI’s Dots on 29 September, whose launch examples are work tasks across apps.

Meta’s announcement says Muse can open a browser, fill in forms and check out with Link, built by Stripe. It “checks with the person before sensitive actions like sending an email or making a purchase.” The shopper still approves the basket, but the agent may have filled it.

Six weeks in, Meta put numbers on it. On its 21 October earnings call, Fortune reported, Mark Zuckerberg said more than 10 million people in the US and Canada had set up Muse agents, a quarter of them use it daily, the most common use is shopping, and the average purchase is over $200. Meta “had to work with Stripe to raise credit limits since the volumes were above what any of us expected,” he said. Meta plans to charge businesses a fee when Muse helps them complete a sale.

Which retailers let Muse buy, and which block it?

Shopify stores and Meta’s named partners say yes; Amazon says no. Amazon began blocking Muse on the night of Sunday 20 September, Bloomberg reported. Amazon spokesperson Lara Hendrickson said third-party applications should “operate openly and respect service provider decisions about whether or not to participate.” Amazon asked Meta to remove Amazon from the experience and, according to Amazon, Meta declined, MediaPost reported.

Shopify took the opposite position. Chief executive Tobias Lütke posted on X that Shopify was “partnering deeply with Muse to enable agentic checkout with Shop Pay on all Shopify stores”, Retail Insight Network reported. On 24 September Shopify posted that “Shop Pay is now available as a wallet in @Muse”, according to TruCommerce, which notes that Meta’s own payments page had not yet listed it. At Meta Connect, Alexandr Wang said Muse can access the entire Shopify product catalogue, and Meta named retail partners including Best Buy, Gap, Sephora, Walmart and Wayfair, according to TechCrunch.

So the shelves an agent can buy from are being settled one retailer at a time. If most of your sales go through a retailer that blocks an agent, that agent’s users can’t buy you there.

How many shoppers use AI to buy?

Before Muse, plenty used AI to research and few let it buy. The surveys below were all fielded before Muse launched, which is why Meta’s 10 million figure matters.

Survey	Who and when	Finding
L.E.K. Consulting	US consumers, April 2026 (2 650 consumers surveyed)	About 30% used AI to inform a purchase; 8% of AI shoppers had completed one entirely through an agent
L.E.K. Consulting	Same survey	46% of AI shoppers started research on a stand-alone AI tool, up from 25% two years earlier
Gartner	Two US surveys: 846 people (late 2025) and 322 (January 2026)	Willingness to let AI make purchase decisions topped out at 11%, in lower-stakes categories
Global Payments and The Lantern	8 027 consumers in seven markets, May 2026	Shoppers expect AI to make 15% of their purchases within five years, up from 9% a year earlier

The research step had already moved. L.E.K. found that 31% of AI shoppers had largely made their decision before reaching a brand or retailer site, up from 26% two years earlier.

“Consumers are not looking to outsource shopping decisions to AI,” said Gartner’s Kate Muhl in May. Five months later, a quarter of 10 million Muse users were using it daily. The buying step has started to move, and we don’t yet know how far it goes.

Why does unaided brand awareness matter for AI agents?

Because the shopper’s instruction comes from memory. Byron Sharp of the Ehrenberg-Bass Institute defines mental availability as the probability that a buyer will notice, recognise or think of a brand in a buying situation.

In a shop, a brand a buyer only recognises can still win at the shelf. When an agent shops, the buyer may never see the alternatives. A brand that has to be noticed has fewer chances. A brand that comes to mind gets written into the instruction.

The survey measure closest to “will they ask for us by name” is unaided awareness: the share of category buyers who name you without seeing a list. If your aided awareness is high and unaided is low, you have the “recognised but not remembered” leak from finding where your funnel leaks. We expect agents to make that leak more expensive.

Being named only pays off if the agent can buy you where it shops, so distribution matters as much as memory.

How should brands prepare for AI shopping agents?

Run the agent-ready check. Six steps, most of them repeatable monthly:

1. Baseline unaided awareness now. Measure how many category buyers name you unprompted, next to your closest rivals, while agent purchases are still new.
2. Make the name easy to ask for. Say and show the brand name clearly in advertising, and keep product names identical everywhere, so a buyer can repeat them to an agent.
3. Map where agents can check out. List every retailer that sells you and mark which have partnered with agents and which block them. For Muse today, that means Shopify stores and Meta’s named partners on one side, Amazon on the other.
4. Make product data consistent. L.E.K. says the shift raises the importance of structured, consistent and machine-readable data. Names, sizes, prices and stock status should match in every store.
5. Put generic requests to the agents. Each month, ask the main agents for your category without naming a brand, and log the brands and retailers that come back. Model updates can reshuffle the answers, so one test tells you little.
6. Keep your reviews and search results clean. L.E.K. found that 94% of AI shoppers checked the AI’s output before buying, most often through search (48%) and reviews (47%).

If the board wants to know why brand spend still matters when software picks, start with investing in your brand.

What don’t we know about AI shopping agents yet?

These are the open questions we’d watch:

• How much agents will buy. Shoppers in the Global Payments survey expect 15% of purchases within five years, a forecast by shoppers about their own behaviour. Meta’s early numbers are bigger than those surveys suggested, in our reading.
• How agents rank brands on a generic request. The Allouah study found model updates can reshuffle market shares. We found nothing published on how Muse picks.
• How placement will be sold. Meta says it will charge businesses when Muse completes a sale, and says it won’t share details from Muse conversations with its advertising systems, though Engadget reported that shopping could indirectly influence the ads people see.
• Which retailers open up and which block. Amazon and Shopify chose opposite sides within days of each other.
• Whether unaided awareness predicts agent purchases. We couldn’t find any public data linking the two. It’s a reasonable bet, and it still needs measuring.

Where does Wink fit, and where doesn’t it?

Wink, the brand tracking platform, measures unaided awareness: how many real category buyers name your brand without a prompt, each month, next to up to five competitors. It tracks in South Africa, the US, the UK and Australia.

The survey asks unprompted awareness before aided, then consideration, usage and preference. Results arrive monthly as a three-month rolling average, every number comes with its margin of error, and Wink says when a change is too early to call. That covers step 1 of the agent-ready check.

Wink doesn’t track what AI agents recommend or buy, which retailers they can reach or how much they sell. For step 5, query the agents yourself, and read sales by channel for the rest.

Take your unaided awareness baseline now, while agent shopping is new. Book a demo to see it measured next to your closest rivals.