Will sellers game Google's new agentic commerce protocol?

Cory Doctorow argues that Google's "universal commerce protocol" (UCP) — described as "a way for AI 'agents' to retrieve prices and product descriptions and make purchases" — will run into the same wall that sank the "semantic web": sellers have a standing financial incentive to lie to machines about their own prices, and that incentive does not go away because the machine reading the page is now a chatbot instead of a search crawler.

His mechanism is the junk-fee playbook. He describes an online travel listing where a hotel that discloses its true all-in price "ends up at the bottom of the price-sorted list, rents no rooms, and goes out of business (or pivots to lying about its prices, too)," while a hotel that hides fees until checkout or arrival "will seem like the cheapest, best option." Because "crooked offers drive out honest ones," Doctorow concludes that once lying is profitable and computers are easier to lie to than people, "everyone has to lie, or be pushed out of the market" — the same logic, he says, that already killed decades of attempts to build a semantic web of accurate machine-readable listings. He adds a second, narrower failure mode specific to UCP: Google already cannot stop search-engine-optimization firms from gaming its crawler results, so "Google somehow thinks that the companies that spend millions of dollars trying to trick its crawler won't also spend millions of dollars trying to trick its chatbot" — in his words, Google is "providing the internet with a tool to inject lies straight into the chatbot's input hopper."

This connects directly to Can AI seller agents shift buyer preferences through description edits?, which measures exactly the adversarial dynamic Doctorow predicts: sellers already edit product descriptions to shift an AI shopping agent's choices, in five of six models tested. Doctorow's argument supplies the economic reason that finding should generalize rather than stay a lab curiosity — gaming the agent is cheaper and more effective than disclosing honestly, by his account, so it should scale with adoption rather than get engineered away. It also bears on When do pricing cues actually mislead AI shopping agents?: that note finds shopping agents fall back on marketing heuristics mainly when looking up real attributes is costly, which is precisely the condition Doctorow says sellers will keep manufacturing by refusing to expose accurate, cheap-to-check rate cards.

Doctorow's piece is commentary built on analogy — to the semantic web's history and to search-engine SEO — not a direct study or test of UCP, which he describes as newly announced rather than yet measured in deployment. He does not report any data on how often UCP-connected agents are currently deceived, nor any response from Google to the obfuscation problem beyond the claim that sites will "expose accurate rate cards." The implication, at the strength an analogical argument allows, is that any evaluation of UCP's reliability should treat price and description accuracy as an adversarial, continuously contested property of the ecosystem rather than a solved input format.