The week the shopper stopped being just a person
For two years we have heard that AI agents would end up shopping for us. This week it stopped being a prediction.
On 21 September, Shopify and Meta announced that the Muse assistant can pay with Shop Pay in any Shopify store. A week later, on 28 September, Shopify went further: it opened checkout to any browser-based AI agent, not just Meta's. On the same day, Amazon was sticking with its decision to block agents that try to buy on its platform.
Two e-commerce giants, two opposite strategies. And in the middle, your store.
What exactly Shopify announced
It is worth understanding properly, because the small print matters.
- The agent no longer needs screenshots. Shopify has extended WebMCP to checkout with three tools: read the cart, update it and submit the order. The agent talks to the store in its own language instead of guessing where each button is.
- The buyer has to authorize the purchase. The agent prepares the order, but the person signs off before paying.
- It is switched on by default for eligible US merchants. Anyone who does not want to take part has to turn it off manually in the Shopify admin, under Sales channels, Agentic. In Spain and Latin America it is not yet the default, but the direction is clear.
The numbers that change everything
On 28 September Adobe published its holiday season forecast, built on data from more than a trillion visits to US online stores. It is US data, but what happens in American e-commerce usually reaches Europe and Latin America a year or two later.
- Traffic from AI assistants will grow 130% on last year. On Thanksgiving, 159%.
- Those visitors add to cart 32% more often than visitors arriving any other way.
- They generate 43% more revenue per visit. Eighteen months ago it was exactly the other way round: regular traffic was worth far more than traffic from AI.
- 77% of people who shop with AI help feel more confident in their purchase, and 69% say they are less likely to return it.
Why this is not just another trend
What struck us most is how much the quality of those visits has changed. A year and a half ago, the visitor arriving from ChatGPT browsed and left. Today they buy more and return less than someone arriving from Google.
It makes sense. When someone asks an assistant to find a coffee machine under a hundred euros with good reviews and fast delivery, the assistant has already done the comparing. Whoever reaches your store from there is not coming to browse: they are coming to confirm.
As one customer experience consultant put it this week, brands have to start designing for a world with two customers: the person and the person's agent. The second one does not see your photos or your sale banner. It reads data.
What to review in your store before November
You rarely need to rebuild the store. It is usually enough to make what you already have readable by a machine. This is the list we are going through with our e-commerce clients:
- Product pages with structured data. Price, availability, variants, brand, ratings and product code marked up with Schema.org. An agent that cannot find the price in a format it understands moves on to the next store.
- Shipping and returns written out, not hidden in an image. Delivery time, cost and return conditions in plain text. It is the first thing an agent compares when two stores sell the same thing.
- Real, synchronized stock. If the agent recommends a product that turns out to be sold out, the store loses the sale and the agent learns not to recommend it again.
- Your robots.txt, reviewed with judgment. Blocking AI crawlers may make sense for Amazon. For a mid-sized store it usually means disappearing from the conversation where the purchase is decided.
- Measure traffic from AI. In your analytics, filter visits coming from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. If you do not measure it, you will not know whether it is growing or how much it sells.
- A checkout without surprises. Costs that appear at the last step, mandatory sign-up or unnecessary steps slow a person down and stop an agent.
And if your store is not on Shopify
The news comes from Shopify, but it affects any online store. WooCommerce, PrestaShop, Magento or a custom-built store get the same kind of visit and are judged by the same standard: if the data is well structured, the agent can recommend you; if not, it skips you.
The difference is that on Shopify part of the work comes done by the platform, and on your own store it has to be done by hand. It takes a bit more work, that is all. And it is exactly the same work that AI search visibility already required, which we covered in our article on how to get ChatGPT to recommend your company.
What we would do this week
If you run an online store and can only spare one afternoon, this is what we would prioritize:
First, measure. Create the segment of traffic arriving from AI assistants in your analytics. It takes twenty minutes and tells you whether this already affects you or not yet.
Second, check your ten best-selling products with Google's rich results test. If price and availability do not show up, an agent is not seeing them either.
Third, write your shipping and returns into the product page, not just in a legal page linked from the footer.
That covers eighty percent of what an agent needs to recommend you.
If you want us to look at it with you
We build and optimize online stores on Shopify and custom platforms, and we work on visibility in search engines and AI assistants through SEO and GEO. That is exactly where this news lands.
Book a thirty-minute call. We review your store with you the way an AI agent would see it: what data it reads, what is missing and what is costing you sales. You leave with a concrete list of changes ranked by impact before Black Friday. No strings attached.
Sources
This article draws on the following publications, consulted on 29 September 2026:
Frequently asked questions
It is a purchase made by an AI agent on a person's behalf: it finds the product, compares stores, fills the cart and prepares payment, and the person only authorizes the final order. Since 28 September 2026, Shopify lets any browser-based agent complete checkout on its stores with the buyer's authorization.
Not in Shopify's model. The agent reads the checkout, fills it in and leaves it ready, but the purchase needs the buyer's explicit authorization before it is submitted. What does change is who does the searching and comparing: it is no longer the person.
For now, Shopify's default activation applies to eligible US merchants, and Adobe's figures are for the US market. But traffic from assistants like ChatGPT, Gemini or Perplexity already reaches stores in every country, and what settles in US e-commerce usually arrives elsewhere within a year or two. Preparing now costs little.
For most stores, no. Amazon can afford it because it is the destination people go to directly. A mid-sized store depends on being found, and blocking agents usually means being left out of the comparison where the purchase is decided. The sensible move is to allow access and take care of what data they find.
In your analytics tool, create a segment of visits whose source is chatgpt.com, perplexity.ai, gemini.google.com or copilot.microsoft.com. In Google Analytics 4 you do it with a session source filter. That shows how many visits arrive, which pages they view and how much they buy compared with the rest.