A Step-by-Step Guide to Getting Ready for AI Agents

The agents are coming.

For months now tech giants like Google and OpenAI have been introducing AI agents that can carry out actions on a shopper’s behalf and make purchases for them on e-commerce sites. Their ambition is rooted in the growing role AI is playing in shopping, which — while still small — is rapidly growing.

Retailers and e-commerce platforms are rushing to get ready. Companies like Shopify, Etsy and Walmart have been creating alliances with the tech giants to make sure consumers are able to discover, compare and buy products from their sites seamlessly within tools like Gemini and ChatGPT.

“It keeps us moving in line with the fast-changing retail landscape,” said Jetan Chowk, chief technology and transformation officer at sneaker and sportswear retailer JD Sports, which recently announced that it would enable buying through platforms like ChatGPT and Microsoft Copilot. “We want to make sure that we offer the ability to transact across where [our consumers] are searching, where they’re browsing and where the purchase intent is.”

Shopping agents remain a long way from being mainstream. For one, the agents from the big tech players are typically only available in the US at the moment. Shoppers in Europe and elsewhere have to wait a bit longer. Even for US retailers that are fully shoppable, BCG managing director and partner Martin Barthel said he doesn’t expect even a 5 percent sales lift at this point. And agents still make mistakes, like getting product pricing or availability wrong.

That doesn’t mean brands should ignore agentic commerce, however. The landscape is changing fast, and Barthel said brands must start building their agentic readiness, both for their own uses, like customer support, and in preparation for external shopping agents.

“If you do it one month earlier or one month later, you probably don’t care,” he said. “If you say, ‘OK, I’m waiting now for another year,’ then you certainly lose traction.”

For brands that want to make their products available to purchase by AI agents, here are some of the key steps to take.

1) Make sure agents can find your items

Before they can buy your products, agents need to find them. Companies are doing the equivalent of SEO for AI platforms, called generative engine optimisation, or GEO. It entails making sure product pages contain well-structured data that’s easily read by AI, as well as thinking about how users search on AI platforms. Compared to traditional search, AI queries tend to be more conversational and centred on questions users want answered.

Retailers need to think about which questions consumers are asking that could result in AI recommending their product, and which product attributes the AI needs to know about to make that recommendation. Google now allows retailers to add “conversational attributes” to the product data they feed into its merchant centre, which can include answers to common questions shoppers are asking about an item.

“Humans infer and fill gaps; AI agents will only know what is explicit, structured and trustworthy,” said Tarun Chandrasekhar, chief product officer at Syndigo, which helps brands manage and syndicate product data.

Because AI might also draw on what a brand says about its products in blog posts or what people are saying in reviews and on sites like Reddit, companies are creating more content about their items beyond product pages. Chowk said JD Sports’ digital marketing team is working to “drive the right content on the internet” to ensure its products surface in recommendations.

2) Decide which agents you want to let shop your products

Retailers need to consider which AI platforms, if any, they want to let shoppers purchase through. ChatGPT, for example, takes a commission on sales while some others don’t (at least not yet). But Barthel also deemed it a strategic choice, likening it to determining whether you want to make your products available for sale on Amazon or TikTok Shop.

“It’s not a trivial decision,” he said. “If your products are shoppable on those engines, there is a certain risk of being more and more disintermediated by those platforms.”

Brands should also think about whether they want all their products to be available or just select ones. Some items might not qualify. Ashish Gupta, Google’s vice president and general manager of merchant shopping, pointed out that items requiring certain customisations may not be eligible to buy directly with Google’s agent.

3) Connect to the AI platform and share product data

Once you’ve decided where you want to sell, make sure your site permits traffic from those agents (called “allowlisting”) and become a registered seller with the platforms so you can share product data with them directly. Many retailers are already signed up on Google’s merchant centre, but retailers that want to sell through other companies like OpenAI or Perplexity should register with their merchant programmes.

Making sure AI platforms have accurate, up-to-date product information, including pricing and stock availability, is vital, otherwise an agent could try buying the wrong item, give a shopper the wrong price or try to purchase a size that is no longer available. Wrangling all that data and providing it in the right format can be a major bottleneck. The Information reported in January that OpenAI has been slow enabling checkout within ChatGPT because of issues with retailers’ product data not being uniformly structured and standardised.

Brands should ask if their e-commerce providers can help. Shopify has collaborations with Google and OpenAI (and is working with OpenAI on the data challenges mentioned above). JD Sports uses Commercetools’ AI hub, which Chowk said feeds all their product information, including pricing and inventory availability, to the AI platforms.

Companies will want to integrate protocols used by the AI platforms they want to sell through, such as Google’s universal commerce protocol (UCP) or OpenAI’s agentic commerce protocol (ACP), developed in concert with e-commerce and payment providers. These protocols create a standard coding language agents can use to communicate with other software. UCP, for instance, tells retailers which APIs, or application programming interfaces, they need to make accessible to the agent.

“Merchants implement these APIs, and then the agents can call these APIs to carry out various actions, whether it is checking availability for an item, or whether it is completing the checkout,” said Google’s Gupta.

4) Enable checkout

To make purchases, agents have to be able to make payments, which can require some adjustments from retailers and payment services, since many systems were designed to keep out bots used by bad actors. If an agent completes forms too quickly, for example, a site might block it.

Payment companies such as Visa, Mastercard and Stripe are teaming with the AI platforms and working to ensure agents can complete purchases through measures like registering and verifying trusted agents. Retailers should check with their payment processors to see if there’s anything they need to be aware of. Stripe users, for example, can enable payments with one line of code, according to OpenAI.

5) Monitor your visibility on AI tools and stay up on new developments

The amount of time it takes brands to get set up to sell to agents will vary. BCG’s Barthel said if a brand has all its APIs ready, it can have a limited number of products available for sale within a few days of registering with the AI platform. If it needs to build APIs, it will take longer. To reach full scale with all products for sale, including running tests and correcting errors, can be a few weeks.

But once a brand’s catalogue is ready to be shopped by agents, the work doesn’t end. Companies will need to regularly monitor how their products show up in AI recommendations. One 2025 study that involved researchers building a fake website as a testing environment found different platforms often preferred different brands, and factors like where a product appeared in search results influenced an AI’s recommendations. In an interview for BoF’s memo on AI for executive members, Omar Besbes, a professor at Columbia Business School who was one of the study’s authors, imagined a future where marketing teams build test environments and routinely conduct experiments to figure out the best ways to describe products and optimise their marketing.

In the meantime, companies have work to do to make shopping agents a widespread reality.

“There’s a lot that needs to be done across the industry for agentic commerce to be really widely adopted,” said Gupta. “For all of this to work, we need to continue to work with the entire ecosystem to build this future of agentic commerce.”