When an AI agent shops, the brand site comes last: what the PYMNTS study says, and what to change

Section 1 · The study

Four consumers in ten would send their AI agent to a marketplace

PYMNTS Intelligence, the research arm of the US payments media outlet, publishes a monthly benchmark on how consumers use AI. The October 2026 edition is based on a survey run August 10–13, 2026 among 2,191 US consumers, weighted to represent the country's adult population.

One question sums it up. Imagine your AI assistant buys an item for you at the same price, with the same shipping and the same returns everywhere. Where do you want it to buy?

Why “all else equal” matters so much The question neutralizes price, shipping and returns. What remains is the spontaneous preference for a place to buy. And in that tie, the brand site comes last, behind an option that amounts to letting the AI decide.

Section 2 · Habits

AI-assisted purchases already flow mostly to the big platforms

The survey doesn't only measure intent. It also asks where the most recent AI-assisted purchase happened. For 59.7% of respondents, it was Amazon. Big-box and department stores follow with 19.5%. Brand sites or apps account for only 6.3%.

Why those places? Respondents first cite the best price or a discount (27.5%), then free or fast shipping (17.3%), trust in payment (9.1%) and habit (8.1%). A direct link from the AI tool is cited by only 2.5%.

Yet AI is already moving some purchases. 44.7% of respondents say it led them to buy from online stores they had never used, and 39.9% that it changed where they buy. The door is open for brands, provided they pass the filter.

Section 3 · What wins them back

Price, shipping, returns: nothing about design, nothing about the homepage

PYMNTS then asked what would make these shoppers choose the brand's site over a marketplace. Every answer is a sales term.

None of these reasons is about the site's look and feel. An agent doesn't see a banner or a carousel. It reads a product page and compares a price, a delivery time, a return policy, a stock level. If the information is vague, contradictory or buried in the terms and conditions, it moves on to the next merchant.

“An AI agent doesn't compare storefronts. It compares terms.”

Section 4 · Walmart

Walmart opens a dedicated entry point for agents, with rules

On October 21, 2026, Walmart announced three building blocks to let AI agents shop on behalf of its customers while keeping the customer in control.

Walmart specifies that agents see the same prices as human customers, with offers and fees shown before purchase. In other words, big retail is getting its terms ready to be read by machines.

What is MCP? The Model Context Protocol is an open standard that lets an AI agent plug into a service (a catalog, a cart, a tool) and use it in a structured way, without having to “read” web pages like a human.

Section 5 · To do

Make your terms readable before polishing the storefront

For a brand, the takeaway is not to run away from marketplaces. It is to treat sales terms as a product in their own right, readable by a human and by an agent, and tested.

1. Price and offers: are the final price, discounts and site-only perks visible on the product page, without a click or a login?
2. Shipping: are delivery time and cost shown clearly at product level, and as competitive as on the marketplaces you sell on?
3. Returns: does the policy fit in one clear sentence on the product page, rather than in the terms and conditions?
4. Stock: is availability accurate and readable in the page code, not only in an image?
5. Structured data: are price, availability and shipping described in the page markup that machines can read?
6. Tests: test these elements like the rest of the site. Showing the delivery time or return policy higher on the product page is a simple A/B test to launch.

A caveat on these numbers This is a US survey, weighted but self-reported, about a situation that is still partly hypothetical. It doesn't predict your traffic. The ranking and the weight of sales terms, however, are signals to take seriously.