AI agents that shop for your customers

A growing number of AI tools can compare products, read reviews, and suggest a shortlist on a buyer's behalf. Some can go further and complete simple purchases.

This brief explains what is happening, how fast it is moving, and what it means for brands. The short version: the first round of shopping is moving to software, and that changes what it takes to make the shortlist.

The shift

When software does the first round of shopping, the shortlist is decided before a person ever sees your brand. What gets you on it is clear, consistent, checkable information: what you sell, who it is for, what it costs, and why it is different.

The traffic is still small but growing fast. Adobe found that during the 2025 holiday season, traffic to U.S. retail sites from generative AI tools rose 693.4% compared with the year before. Adobe added a fair caution: “While the base of users remains modest,” the jump shows the value AI can bring to shoppers.¹

The audience behind it is large. OpenAI says its products reach “our collective 1.2B weekly users.”² Alphabet says AI Mode in Google Search has passed one billion monthly users.³ Even a small share of those people shopping through AI adds up.

Who is building it

The big platforms have spent the past year building the plumbing. In September 2025, OpenAI launched Instant Checkout, which let people buy from some merchants without leaving ChatGPT, along with an open standard called the Agentic Commerce Protocol.⁴

In January 2026, Google announced the Universal Commerce Protocol, developed with Shopify, Etsy, Wayfair, Target, and Walmart. Google describes it this way: “UCP establishes a common language for agents and systems to operate together across consumer surfaces, businesses and payment providers.”⁵ Google also stresses that “Retailers remain the seller of record.”⁵

Not every early attempt has worked. By March 2026, CNBC reported that OpenAI was ending Instant Checkout in favor of retailer apps inside ChatGPT that send users to the retailer’s own site to pay. Walmart said conversion rates were three times lower for products sold directly in ChatGPT than for those that sent users to its website.⁶ Walmart had made about 200,000 products available inside ChatGPT, while, according to Forrester, only about 30 Shopify merchants were live on Instant Checkout.⁶ Bob Hetu, an analyst at Gartner, told CNBC: “OpenAI underestimated how difficult the enablement of transactions was going to be, which, on the one hand, is a little surprising, but on the other hand, it's not easy for retailers.”⁶ Forrester analyst Emily Pfeiffer summed up the state of play: “Everyone thinks everyone else has this figured out, or is farther ahead of them,” she said. “The fact is that no one has this figured out.”⁶

What shoppers will let agents do

People are happy to let AI do the research but hesitate when money moves. Worldpay’s 2026 research found that among consumers open to AI shopping, 74% would let an agent compare prices on its own and 69% would let it research products. Only 21% would let an agent complete a purchase on its own, while another 59% would allow it if they approve each purchase.⁷

CNBC also cited a Semrush survey of more than 1,000 U.S. consumers: just 22% had bought a product inside an AI tool, while half said they had made a purchase after using AI for research.⁶ For now, the AI shapes the shortlist, and people still finish the purchase.

The long-term stakes are still large. McKinsey estimates that by 2030, AI agents could orchestrate up to $1 trillion in U.S. retail revenue, and $3 trillion to $5 trillion globally.⁸ Treat that as a sense of direction rather than a forecast to plan around.

Research behavior is already there. Bain found that 42% of users of AI chat tools ask them for shopping recommendations.⁹ Adobe’s data shows the same pattern from the retail side: during the 2025 holiday season, U.S. consumers spent a record $257.8 billion online, and AI tools are becoming one of the ways shoppers start.¹

Why it matters

Persuasion still matters, but it now has two audiences. People respond to story. Software responds to structure. Brands that serve both will be easier to find and easier to choose.

Walmart’s numbers are a useful reminder of the second audience. Even when an agent did the finding, shoppers were more likely to buy once they reached Walmart’s own site.⁶ The handover from software to person, the moment someone lands on your page, still needs to work well.

For software, structure means product information that is complete, current, and the same everywhere: names, prices, sizes, stock, delivery times, and returns. As Pfeiffer told CNBC, “Crawling and scraping is inadequate to get the full breadth of product data that you need to do a good job of commerce.”⁶ Brands that supply clean product data directly through feeds and partner platforms will be described more accurately than those that leave agents to guess.

For people, story still decides the final choice. An agent can narrow ten options to three. A person picks the one they trust. Reviews, clear proof, and a recognizable brand do that work.

There is a risk side too. If an agent repeats an outdated price or a wrong spec, the customer blames the brand, not the software. Keeping your own information right is now part of customer service.

How to prepare without overreacting

Nobody needs to rebuild their store for agents this year. The protocols are new, and as the Instant Checkout story shows, the platforms are still learning what works. But a few steps pay off whatever happens:

• Clean product data. Make sure titles, descriptions, prices, and stock are correct in every feed you send to marketplaces and search engines.
• Answer the real questions. Put sizes, compatibility, delivery times, and return terms on the page in plain text, not only in images.
• Ask your platform. If you sell through Shopify or a similar service, ask what it is doing to support agent shopping, so you don’t build what it will provide.

Coming soon: Get your files AI-ready: naming, versions, and one source of truth →