AI shopping agents, one year on
OpenAI stepped back from its own checkout, Google announced a cross-store cart, and a court ruled on agent shopping. What the year taught online stores.
Last October we wrote about the moment the shopper became an AI agent. OpenAI had just launched Instant Checkout in ChatGPT, Google had published its Agent Payments Protocol, and it looked as though buying inside a chat would become normal fast.
Some of that happened. Some of it did not. A year on, the picture is clearer, and more useful for a store owner than the launch headlines were.
Checkout inside the chat did not take off
The clearest lesson came from OpenAI itself. In March, it said it would deprioritise Instant Checkout as a standalone feature. The first version, OpenAI said, "did not offer the level of flexibility" it wanted, and reporting at the time suggested users were not using ChatGPT to complete purchases in large numbers. OpenAI now lets merchants use their own checkout instead, either on their website or through an app inside ChatGPT, and is concentrating on product discovery: comparisons of products, prices, features and reviews.
The Agentic Commerce Protocol, the open standard OpenAI built with Stripe, survived the change and still carries merchants' product data into ChatGPT.
The reporting suggested people were happy to research with AI but still preferred to buy from the shop.
Google is building the cart instead
Google took a broader route. In January it launched the Universal Commerce Protocol with Shopify, Etsy, Wayfair, Target and Walmart, and more than twenty other companies signed up behind it, including Visa, Mastercard, Stripe and India's Flipkart. In May, Google announced a Universal Cart: one basket that works across different retailers in Search and Gemini, paid with Google Pay, or handed over to the retailer's own site. The retailer stays the merchant of record either way. It was promised for the US this summer and had not launched by early September.
It is still mostly a US story. Google says checkout built on the protocol will reach Canada and Australia in the coming months and the UK after that. Two pieces from the same announcement do reach further: tools that let retailers worldwide describe products in conversational terms, and a Merchant Center report on how a brand performs in AI results, which Google says is coming to India among other markets.
The courts weighed in on agent access
The year also produced the first big legal test of AI shopping agents. Amazon sued Perplexity last November over its Comet browser's shopping agent, and in March a district court ordered Perplexity to stop. In August, the Ninth Circuit Court of Appeals overturned that order. The judges reasoned that it is the user who accesses Amazon, with the agent's help, much as a person uses a browser. The case continues, but the direction is significant: in the US at least, an agent acting for a customer is being treated more like the customer's tool than an outside bot.
What reached our region
The Gulf and India saw pilots, not products. Mastercard chose the UAE last November as the first market outside the US for its Agent Pay service, starting with Majid Al Futtaim. In India, NPCI and Razorpay began testing UPI payments inside ChatGPT last October with BigBasket. As far as we can see, neither has yet become something a typical store can switch on.
What the year taught
A few lessons stand out.
- Discovery moved faster than buying. Customers do use AI to research and compare. Most still finish the purchase on the shop's own site or app.
- Product data is the storefront. Whether the agent is ChatGPT, Gemini or a browser, it decides what to show from your feed and your product pages: titles, attributes, prices, stock, delivery and returns.
- Your own checkout still matters most. When an agent or a person arrives ready to buy, a fast, guest-friendly checkout on your own site is what turns the visit into a sale.
What to do now
For stores in the UAE and India, the work is the same as we suggested a year ago, with one addition.
Keep your product data complete and accurate in Google Merchant Center and on your own pages. Describe products in the words people use when they ask questions ("good for small apartments", "suitable for sensitive skin"), not only in technical specs. Put shipping areas, delivery times and return terms in plain text. Make sure a guest can buy in a few steps.
The addition: start watching how AI tools describe your products. Ask ChatGPT and Gemini the questions your customers would ask, note where you appear and what they say, and fix the pages behind any mistakes. When agent checkout does reach our markets, the stores that agents already describe correctly will be the ones they buy from.
