Does Google's AI pricing tool coordinate prices across rival merchants?

Matt Stoller's newsletter piece argues that Google's newly announced Gemini-powered ad and commerce system is becoming more than an advertising platform — it is starting to coordinate pricing across competing retailers. He reads this off three Google announcements made the same week: Gemini gaining access to users' Gmail, YouTube, Photos and Search data; a deal to power Apple's Siri; and a "new Gemini-powered ad service and open protocol to create personalized surveillance pricing for merchants," with partners including Walmart, Visa, Mastercard, Shopify, Gap, Kroger, Macy's, Stripe, Home Depot, Lowes and American Express. The tell, per Stoller, is that CEO Sundar Pichai said Google will sell "not only marketing, but price coordinating services," and that the documentation for its "universal commerce protocol" lists "dynamic pricing" as a tool for merchants. Kroger, a Google partner, has already said it will use Gemini enriched with its own proprietary data for consumer pricing. Google is also piloting "Direct Offers," where, in Stoller's words, "businesses will pay to allow Google set prices when it makes recommendations to users through Gemini," rather than simply buying ad placement.

Stoller's reasoning turns on a structural point about Google's existing ad business: "both Nike and Reebok advertise on Google, and it's a little weird, because it means that each of them teaches Google how to sell sneakers, then rival sneaker companies can also hire Google to also sell sneakers. That's not illegal." But, he argues, "if both of them give Google the authority to do pricing, then all of a sudden Google is coordinating pricing for sneakers, which looks much more like an automated form of price-fixing." The same intermediary that already mediates attention between rivals would also mediate their prices, collapsing the line between serving competing clients and coordinating among them. He situates this in a longer pattern of Google using acquisitions — Applied Semantics in 2003, Keyhole in 2004, Android in 2005, YouTube in 2006, DoubleClick in 2007 — to extend control over "the world's information," with antitrust enforcement (he cites the 2006 Foundem complaint) as the check that historically failed to keep pace.

This sits alongside Can an AI agent serve both merchant and user interests fairly? but reframes the loyalty problem at the market level rather than the single-agent level: Azhar's concern is one agent's conflicted loyalty to one user, while Stoller's concern is one platform simultaneously setting prices for many competing sellers, which raises collusion rather than just bias. It also extends Can EU consumer law protect AI agents as algorithmic buyers? by supplying a concrete commercial mechanism — an ad platform turned pricing coordinator — that existing consumer-protection and antitrust frameworks were not built to catch.

The excerpt does not establish that Direct Offers or the dynamic-pricing protocol has actually produced coordinated or elevated prices in practice; Stoller's argument is an inference from the structure of the rollout ("there are many unanswered questions about how this new system will work"), offered in an opinion newsletter rather than an empirical study or regulatory finding. If the structural reading holds, though, it implies that antitrust review of agentic-commerce platforms needs to examine pricing-coordination features specifically, not just advertising and data practices, since the conflict-of-interest problem scales from a single agent's loyalty to an entire market's price-setting.