Meta launched Muse, an AI agent that shops for you, on 8 September 2026. Amazon started blocking it on 20 September, twelve days later.
Meanwhile only 9% of shoppers say they would let an AI agent buy on its own. Yet across 412 companies tracked by Neil Patel's agency, AI already sends about 6.2% of traffic and about 11.4% of sales.
I watched Neil's breakdown of the Muse fight because those numbers point in opposite directions, and if you sell online, how they fit together decides whether AI agents ever put your products in front of a buyer. Here is the story, the data, and the two jobs I think every store should start on now.
Twelve days from launch to ban
Muse is Meta's first standalone personal agent. According to Neil, it reads your email, books travel, fills in forms and shops for you, and it keeps working after you close the app. When it buys something, it pays through Stripe's Link, which issues a single-use card number so the store never sees the real card. Neil says there is a free plan, a $20 a month plan and a $100 a month plan, and that Mark Zuckerberg described the long-term model as taking a small cut of the purchases Muse makes, possibly paid by the merchant.
On 20 September, Amazon began blocking Muse from its store. Amazon's complaint, as reported in the press: Meta never asked for permission, Muse did not identify itself as an AI agent, and it could reach customer account details. Meta disputes parts of that. Later in September, Meta named shopping partners including Walmart, Best Buy, Sephora and Wayfair, with Shopify stores reachable through their catalog.
→ One retailer shut the door and several others opened theirs within the same month.
The real fight is over who stands at the checkout
Neil skips the question of who is right and goes to the money. Amazon made more than $68 billion from advertising last year. That income depends on people browsing: scrolling past sponsored products, seeing "customers also bought", comparing three options before picking one. An agent skips all of it. It finds the item, buys it and leaves. Nobody browsing means nobody seeing ads.
Muse is not the first agent Amazon has pushed back on. Amazon took Perplexity to court over its shopping browser, and has moved to block agents from Google and OpenAI. At the same time, Amazon's own Alexa can shop for you. So Amazon is fine with AI shopping as long as it is Amazon's AI.
The question underneath is who sits between the customer and the purchase: the store, the AI company or the payment company.
→ For a store owner the useful question is narrower: can these agents find you, and will they recommend you?
Helping you shop and buying for you are different
Before rebuilding anything around robots buying things, Neil looks at how many people want that. Accenture surveyed 25,590 consumers in 16 countries for its "Talk to My AI Agent" report, published in June 2026:
- 74% are open to an agent handling routine shopping tasks, like researching, comparing and watching for price drops, as long as they stay in control.
- 32% would let an agent choose for them within limits they set, provided they approve the payment.
- 9% would let an agent buy on its own.
Finding, choosing and paying are three levels of trust, and most people are still at the first one.
→ People want help. They are not handing over the card.
AI already moves money, mostly before the click
This is where it gets interesting. Almost nobody lets an AI buy, yet AI already drives sales. NP Digital looked at traffic and sales data from 412 companies: AI sent about 6.2% of traffic and about 11.4% of sales. A visitor who arrives from an AI tool is nearly twice as likely to buy as the average visitor.
The reason is simple. By the time that visitor lands, the AI has done the comparing. They read the summary, saw the alternatives and arrive ready to buy. And it is not only ChatGPT. Google's AI Overviews do the same job inside search.
Neil's quick check, which I'd run today: open your analytics and look for referrals from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. If they barely show up, that traffic is probably being counted as direct, and you are undercounting your best converting visitors.
→ AI is not buying yet. It is already deciding what gets bought.
Toilet paper and the white couch
Neil's way of explaining the 9% is two products. Toilet paper is the easy one: same brand, same softness, best price, reorder. Nobody misses browsing for it, and an agent can do it today.
A couch is different. An agent can do a lot of the work: check your room dimensions, delivery terms and durability reviews, and narrow hundreds of options to six in minutes instead of two evenings. But choosing the white couch you will sit on every night for years is a matter of taste, and people keep that. Neil said he tried to buy a mattress without trying it and it did not work out, so he went to a store and lay on a few.
Accenture's interviews found the same split. People were happy to hand off detergent and wanted to pick the details of their own hotel room. The purchases tied to how people see themselves are the ones they hold on to.
→ Agents take over the specs. Humans keep the taste.
The shortlist is the new shelf
This is the part that matters most for a store. AI will not make every purchase, but it will decide which products a customer ever considers. The customer picks, but only from the shortlist the AI builds. If you are not on it, you never know the sale existed.
Brand is what gets you onto that list. Accenture found that 56% of people would tell their agent which brands to consider. If people know your name, they will ask the agent to look for you.
The other number is less comfortable. Among loyal shoppers, people who buy the same one to three brands in a category, 37% would let an agent switch them to another brand if it found a better fit.
→ Your brand gets you into the room. It does not keep you there.
What marketers think gets you named
NP Digital asked 500 marketers which factors matter most for showing up in AI answers. Brand mentions came first at 94%, reviews and sentiment second at 91%, and backlinks came in at 3%.
Neil is careful to say this is what marketers told them, not a measurement of how the systems work, though it matches what his team sees in client data. His explanation: backlinks were built for a crawler to count, while an AI reads the web more like a person, and a person cares what others say about you more than how many sites link to you.
I sell backlinks as part of Distribb, so weigh my view accordingly. I read the 3% as a survey of opinions, and I don't think links and mentions compete. AI tools pull heavily from search results, and links still help pages rank there. Mentions and reviews decide what the AI says once it finds you. You need both.
In a second survey of 600 marketers, 59% said machine recommendations already have a significant influence on buying decisions, and 3% said none. When they rated trust signals across six AI platforms, third-party citations scored highest on five of them. On ChatGPT, brand mentions edged ahead.
→ What other people say about you counts for more than what you say about yourself.
Why ChatGPT backed off its own checkout
One more piece of the story. OpenAI launched a checkout button inside ChatGPT, and on 24 March 2026 it scaled that back to focus on product discovery. People were using ChatGPT to research products, not to buy them there.
Walmart's numbers show why. Purchases made inside ChatGPT converted at about a third of the rate of shoppers who clicked through to Walmart's own site. Neil adds that bad product data was a big part of the problem: wrong inventory, wrong shipping costs. The agent could not trust the feed, so the customer could not trust the agent.
Neil also notes that OpenAI says ChatGPT's shopping results draw on product data from brands and third parties, including review summaries, and that they are not ads. You cannot buy your way onto that list.
→ AI is taking over comparing before it takes over buying, so the comparison is where you win or lose right now.
Job one: product data an agent can read
Neil boils the response down to two jobs. The first is your own data. An agent can only recommend what it can evaluate, so every product page needs:
- Clear specs, written as text
- Every variant
- The current price and availability
- Shipping cost and delivery time
- The return policy
If you sell online, publish a product feed, because that is what ChatGPT and other agents read directly. And keep it honest. If the feed says in stock when you are not, the agent learns not to trust you.
His test takes two minutes. Open your best selling product page, ignore the photos and the design, and read what is left. If you cannot tell the size, the price, when it ships and what happens if it comes back, neither can the agent. If your descriptions need work, our guide to product description SEO walks through how to write them.
Job two: other people talking about you
The second job is outside sources: reviews on sites your customers already trust, and real customers talking about you in communities and on social. That is what gives both the shopper and the agent a reason to believe you.
Neil's starting point is small enough to do this week. Pick one review site your customers check before they buy, then ask your last 20 customers to leave a review there.
His observation about who gets left out stuck with me. Most companies only do one job. E-commerce brands often have great product data and nobody talking about them. Brands with a great reputation often have product pages an AI cannot read. The ones that show up in AI recommendations do both.
To see where you stand, our free AI visibility checker runs five buyer-style prompts for your category across ChatGPT, Perplexity, Gemini and Google AI Overviews, and shows where you are cited, where you are only mentioned, and which competitors got named instead. I wrote about a related pattern earlier, on why ChatGPT skips big chains for local recommendations.
How Distribb handles both jobs for a store
The two jobs are clear. The problem is time. A store owner already runs inventory, ads, customer service and shipping, and "publish content that gets you mentioned" sits at the bottom of the list every week. That gap is why I built Distribb to run on its own once the store is connected.
On Shopify, here is what it does:
- It reads your catalog. Distribb syncs your products and prices when you connect, so articles mention what you actually sell, in your currency, and link to your real product and collection pages.
- It writes for the comparison phase. It finds the searches buyers make before they purchase and publishes buying guides and comparisons to your Shopify blog on a schedule. Those are the pages an AI reads when it builds a shortlist.
- It builds authority. Distribb's backlink exchange places contextual links from real businesses, which helps your pages rank in the search results AI tools draw from.
- It spreads your name beyond your site. On Accelerator, your best articles become done-for-you videos, Instagram carousels and Medium posts, so you are talked about in more places.
- It tracks AI citations. AI citation tracking is included on every plan, so you see where AI engines name your store.
Pro is $97 a month for 30 articles. Accelerator is $495 a month for 90 articles plus the done-for-you distribution. You connect the store and it runs. The details are on the Distribb Shopify page.
What I'm taking from Neil
- Measure the AI traffic you already have. If AI referrals are invisible in your analytics, you are making decisions about your best converting visitors without seeing them.
- Plan for agents that compare, not agents that buy. With 9% willing to let an agent pay, the near-term battle is the shortlist, not the checkout.
- Your product page is now read by a machine first. Specs, variants, price, stock, shipping and returns in plain text decide whether you get evaluated at all.
- Loyalty is weaker than it looks. If 37% of loyal shoppers would let an agent switch them, being the familiar brand is not enough. You have to win the comparison too.
- Do both jobs. Good data with no reputation, or reputation with unreadable pages, leaves you off the list. The stores that get named do both.
FAQ
What is an AI shopping agent?
It is an AI assistant that shops on your behalf: it researches products, compares them, watches prices and, in some cases, completes the purchase. Meta's Muse, launched in September 2026, is one example. It pays through a single-use card number, so the store never sees your real card details.
Why did Amazon block Meta's Muse?
Amazon said Meta never asked permission, that Muse did not identify itself as an AI agent, and that it could access customer account details. Meta disputes parts of that. The money behind it is Amazon's advertising business, which depends on shoppers browsing, and an agent that buys directly skips the sponsored listings.
Do people trust AI agents to buy for them?
Mostly not yet. In Accenture's survey of 25,590 consumers, 74% were open to agents helping with shopping tasks while they stay in control, 32% would let an agent choose within limits before they pay, and only 9% would let an agent buy entirely on its own.
How do I get my products recommended by AI?
Do two jobs. First, make your product data readable: specs, variants, price, stock, shipping and returns in plain text, plus an accurate product feed. Second, get other people talking about you, starting with reviews on a site your customers already check. Brands that do both are the ones AI tools tend to name.
How can I see if AI tools send me traffic?
Look at referral sources in your analytics for chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. If they barely appear, part of that traffic is probably counted as direct. To see whether AI answers name your brand, run buyer-style prompts for your category or use an AI visibility checker.
Start with your best seller
Read your best seller's page with the photos ignored, fix what an agent could not work out, and ask your last 20 customers for a review on one site. Then check where AI answers name you today. If you want the comparison content, the backlinks and the AI tracking running for your store without adding to your week, connect it on Distribb for Shopify and start the free trial. For a wider view of the tools that measure this, see our roundup of AI visibility tools.