Two announcements landed on Tuesday morning, and read together they describe the shape of retail's next decade better than either does alone. OpenAI introduced Sponsored Agents, a test in which a relevant ad in ChatGPT opens a clearly labeled conversation with a business's own agent, plus a ChatGPT Ads app in the Shopify App Store and a HubSpot integration so a marketer can create and run campaigns without leaving the CRM. Google unveiled AI Performance Insights in Merchant Center, a "share of voice" measure of how a brand is discovered in AI Mode and AI Overviews, alongside cart transfer and checkout-flow testing in its Universal Commerce Protocol hub.

Put those next to a number from PYMNTS Intelligence the same day: before a shopper reaches a retailer, an AI assistant may already have narrowed the field to three options. One company is selling a way into that conversation. The other is selling a way to measure whether you are in it. The shelf has moved inside the agent, and it has a sponsored tier.

For a decade, merchants fought for the top of a results page. The page is now a conversation, the ranking is a recommendation, and the recommendation has a price.

What shipped

OpenAI's piece has three parts. Sponsored Agents is the new format: when an ad is relevant, the user can choose to step into a labeled conversation with the advertiser's agent, ask follow-up questions, spell out what matters to them, and click through when ready. It is being tested with select advertisers in the United States; home-services marketplace Angi is among the first. The Shopify app lets any US merchant on the platform create and manage ChatGPT campaigns from the store admin. The HubSpot integration makes HubSpot the first CRM partner, so leads generated in a ChatGPT conversation land in the pipeline a sales team already runs.

Google's piece is aimed at the same merchant from the other side. Performance Insights shows how products and brands surface in conversational results, framed as share of voice, available in the US, Canada, Australia, India, and New Zealand. The UCP hub updates are the practical bit: cart transfer to a merchant's own site, tools to test the checkout handoff, and analytics to follow. Google's own line on why: "To succeed in this new era of commerce, businesses need to know how customers are discovering them through conversational experiences."

Note what each company did not do. OpenAI did not put the checkout inside the sponsored conversation; the agent hands the shopper to the business's website. Google built cart transfer, which is a way of admitting the handoff is where the value leaks. Both stopped one step before money, which is exactly where the shopping agents stopped two weeks ago.

The numbers behind the announcements

The demand side is real and still small. Google's own survey work, cited in the PYMNTS report, puts 39 million US adults, 15 percent of the population, using AI for product discovery, and 139 million Americans having made a purchase with AI assistance at some point. Only 24 percent say they would let an agent shop and buy for them, and 14 percent would let an AI app hold their card. PYMNTS Intelligence's September survey finds 23 percent of US consumers used AI for product discovery in the past year, and among AI adopters who research purchases, 93 percent used it for discovery. The report's own summary: "Being recommended by an AI could fast become as important as ranking on a search page."

The supply side, measured at the till, is more interesting than the survey. Digital Commerce 360 went back to four retailers who reported less than 2 percent of referrals from AI at the start of the year. Bero is still under 5 percent and "not really growing quickly." Povison is at about 5 percent and expects 10 by early 2027. La Joya Jewelry went from 2 to 3 percent to 5 to 6 percent, and its AI-referred visitors convert at 13 percent or better against a normal 3 percent. Edible Brands, a top-1,000 retailer, is under 10 percent, ChatGPT is about 95 percent of it, and those visitors convert at 5 percent against a site average of 10, with an order value $5.50 higher.

So the traffic is a few percent and climbing, the intent is high, and whether it converts depends on the category. That is a channel worth a budget line and not yet a channel worth a strategy. Which is precisely the moment an ad product arrives.

The Destination Economy, turned inside out

I have argued in the MM Destination Economy Thesis that the merchants who win the agent era are the ones that remain a destination, a place the shopper or the agent chooses to go, rather than an interchangeable line in someone else's results. Tuesday shows what the alternative looks like at scale. The result is no longer a page of ten blue links; it is a shortlist of three, produced by a model, and the model's owner now sells a labeled seat at the table.

Ben Thompson put the commercial logic plainly in Monday's Stratechery: "ChatGPT ads are working, and solve Amazon's biggest problem with chatbots." The chatbot's problem was never demand. It was that a conversation has no shelf, and a retailer cannot pay for a shelf that does not exist. Sponsored Agents build the shelf. On Tuesday, writing about Salesforce, he went further: the company is "abandoning UI as a moat, which is a very smart move because it's disappearing for everyone." I would extend that to retail. If the storefront is disappearing into the agent, the store's own agent is the storefront, and the question for a merchant becomes whether it owns that agent or rents it inside someone else's conversation.

That is why Google's share-of-voice metric may matter more than OpenAI's ad format. A metric becomes a target. The first time a merchant's board sees "share of voice in AI answers: 4 percent," a budget appears to move it, and the only two ways to move it are to be recommended or to be sponsored. Google has shipped the scoreboard; OpenAI has shipped the bid.

The checkout is still where it breaks

None of this fixes the step where money moves, and the same week's data says that step is already leaking. PYMNTS Intelligence's checkout study finds 21 percent of US consumers, 56 million people, abandoned a cart because their preferred payment method was missing, and 47 percent of those abandoners wanted a digital wallet. Among wallet users, 52 percent say they would link the wallet to an AI agent within two years, and 39 percent of non-users say the same.

Read that against Tuesday. A shopper who found the product through a sponsored conversation, was handed to the merchant's site, and then cannot pay the way they want is a sponsored conversation the merchant paid for and lost. The retailers with the strongest AI conversion in Digital Commerce 360's sample are the ones whose sites finish the job. The merchant's bot problem, telling a buying agent from a scraping one, and the court's ruling that the agent is the user both land at this step too. Discovery is being industrialized. The handoff is still artisanal.

What to do, if you sell things

Turn on Google's Performance Insights now, before the holidays, and record the baseline; it is free and it is the number your competitors will be shown. Treat the Shopify ChatGPT Ads app as a test with a fixed budget and a separate conversion measure, because La Joya's 13 percent and Edible's 5 percent came from the same channel. Fix the payment-method gap before buying the sponsored seat, since a missing wallet at checkout is the cheapest reason to lose the sale you just paid to start. And decide, deliberately, whether your own agent is the product or whether you are content to rent one inside a platform's conversation. The AI shoppers who convert better are loyal to no one, and a sponsored seat does not change that.

What to watch

Whether Sponsored Agents gain a checkout, so the sponsored conversation can finish the sale without the handoff. Whether Google's share-of-voice figure becomes a bid target with a price attached, which is the moment the AI shelf becomes an auction like the last one. Whether any top-1,000 retailer crosses 10 percent of referrals from AI before the holidays, which none of the four in the sample has. And whether the labeled sponsored conversation stays separate from the organic recommendation, which is the promise the whole format rests on and the first thing anyone should test.

Sources

If the shortlist has three seats and one of them is sold, how long before the other two are?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.