AI assistants have changed where people research a purchase. A new survey says they have barely changed where people buy. When the price, shipping and returns are identical, most shoppers would still point their AI agent at a large marketplace, not at the brand that makes the product.
The numbers come from "From Prompt to Purchase: Who Wins the AI Shopper?", a PYMNTS Intelligence report published on October 23. It surveyed 2,191 US consumers in August 2026.
The question it asked
The survey set up a clean test. Imagine an AI agent buying for you, and every seller offers the same product at the same price, with the same shipping and the same return terms. Where should the agent buy?
A large online marketplace won with 40.2%. A department store or big retailer got 21.8%, and the brand's own website got 16.1%. The remaining 21.8% said they would let the AI decide.
That last group is the one to watch. More than one shopper in five would hand the choice of seller to the agent. For those buyers, no brand site, marketplace or retailer wins on habit. It wins on whatever the agent can find and trust.
Why the marketplace keeps winning
Respondents who picked a seller type were asked why. Easier or more reliable returns came first at 36.6%. Better selection followed at 34.9%, and plain habit at 32.7%.
Behaviour matches the stated preference. Among people who had made an AI-assisted purchase, 59.7% completed the most recent one on Amazon. Only 6.3% finished it on the brand's own website or app. And 57.9% of consumers who shop with AI said they usually still buy on Amazon, even when the AI found the product for them.
So the AI did the research and the marketplace took the sale. PYMNTS CEO Karen Webster put it in one line: "Recommendation is the easy part."
What would move a shopper
The report also asked 1,286 people who make most of their purchases on marketplaces what would get them to buy direct. A lower price or an exclusive discount moved 53.9%. Free shipping as fast as the marketplace's moved 49.9%. Checkout handled by the AI moved only 12.1%.
That is a useful correction to a lot of agentic-commerce talk. OpenAI launched Instant Checkout in ChatGPT in September 2025, then said in March that it was moving purchases into retailers' own apps inside ChatGPT. The survey suggests why the first version struggled. A smoother checkout is not what shoppers were missing. Trust in the seller was.
Where this does not carry over
These are consumer answers about consumer goods, and they are stated preferences, not observed behaviour. A software buyer comparing three vendors does not have an Amazon to default to. B2B purchases also run through more people and more steps than a single cart.
But the shape of the finding should travel. The agent narrows the field, and the final pick goes to the option that answers the questions buyers worry about. For a shopper, that means returns and selection. For a B2B buyer, it means pricing, integrations, security and who else uses the product. If those answers are not on pages an AI can read, the assistant fills the gap with a review site, a directory or a competitor's comparison page. In B2B, those are the marketplace.
The bigger picture
The 21.8% who would let the AI decide are the new middle of the funnel. They have outsourced the shortlist, and the agent builds it from what it can retrieve and cite. A brand that is absent from those sources is not in the running, whatever its product is like.
The practical lesson is to publish the facts that settle a purchase, in plain text, on your own site: price, terms, limits, and proof. Then check what the assistants actually say when a buyer asks about your category. Their answer may name you, a reseller or a rival, and you cannot fix that until you know which.
ClawWorld runs AI visibility work for B2B startups. Start with a free baseline report: 10 buyer questions, who ChatGPT and Claude name instead of you, and the pages they cite.