Amazon v. Perplexity: The Fight Over Who Controls AI Shopping

In Amazon.com Services LLC v. Perplexity AI, Inc., Amazon sought to prevent Perplexity from using its infrastructure to power an AI shopping agent that would stand between Amazon's commercial surface and the end user. The district court issued a preliminary injunction against Perplexity, but the U.S. Court of Appeals for the Ninth Circuit stayed that ruling pending appeal. The merits remain unresolved.

The structural question the case raises is whether a dominant commercial surface can restrict or resist a new AI‑mediated layer trying to insert itself between merchant infrastructure and the consumer. That question will shape how every AI shopping agent, booking bot, and autonomous purchasing tool operates going forward.

The core issue is control over the customer relationship. When a consumer delegates grocery purchasing to an AI agent with instructions to find the lowest price, the agent queries multiple surfaces, selects the optimal combination, routes the transaction, and notifies the consumer only after the order is placed. The merchant whose products were selected never controlled the search, the comparison, the checkout presentation, or the moment of decision. As JDSupra explains, "everything upstream of that order — the context in which its products were evaluated, the criteria by which they were chosen, and the customer relationship that might have been built — belonged entirely to the agent."

That is a fundamentally different commercial relationship than the one merchants built e‑commerce infrastructure to support. It's exactly why platforms are trying to shape the conditions under which agents can participate before that shift becomes structural.

OpenAI's commerce materials provide a clear example of controlled access. Product‑feed onboarding in ChatGPT is currently available only to approved partners. Merchants are directed into a documented integration path, required to provide structured catalog data, and informed that OpenAI may remove products or ban sellers from being surfaced if platform policies are violated. These are participation rules that determine who may appear inside the shopping experience, through what integration path, and under what continuing conditions.

The same pattern extends to payment infrastructure. Visa's Token Service and Mastercard's Delegated Authentication framework both embed credentialing and authorization conditions at the network level before any individual merchant sees the request. A merchant might be willing to list products but unwilling to permit fully autonomous checkout. A platform might allow discovery but only through an approved commerce schema. The effect is cumulative: the market is not building one gate, it is building a stack of them.

The Ninth Circuit's decision to stay the district court's preliminary injunction means Perplexity can continue operating its AI shopping agent while the appeal is pending. That's a significant procedural win for the AI‑agent side: the court apparently wasn't convinced that Amazon would suffer irreparable harm from Perplexity's continued operation during the appeal.

But the stay is not a ruling on the merits. The underlying legal questions — whether Perplexity's use of Amazon's infrastructure constitutes trespass to chattels, whether it violates terms of service, whether Amazon can enforce restrictions on automated access — remain unresolved. The case could establish precedent for how AI agents interact with commercial platforms, or it could settle quietly and leave the question open.

Amazon v. Perplexity is the most visible legal battle, but it's part of a broader shift. JDSupra identifies multiple layers where participation can be conditioned: commerce protocols that govern how products and orders are structured, payment and trust protocols that condition how delegated credentials are recognized, and infrastructure protocols that determine whether agent‑to‑system interactions are recognized as legitimate at all.

Perplexity CEO Aravind Srinivas has publicly argued that AI‑driven commerce should be open and that restricting AI agents harms consumers. Amazon's position is that its infrastructure is private property and it gets to set the terms of access. Both arguments have merit, and the outcome will likely land somewhere in the middle — with some level of mandated access for AI agents but with conditions that protect merchants from wholesale disintermediation.

If you're building AI agents that interact with commercial platforms, the Amazon v. Perplexity case is your canary in the coal mine. The era of unrestricted scraping and automated access to commercial surfaces is ending. Expect more platforms to implement API‑based access with participation rules, more payment networks to add agent‑specific credentialing requirements, and more legal challenges as AI agents push deeper into commercial workflows.

The practical takeaway: build your agents to work within approved integration paths where they exist. Agents that respect platform rules and add value to the merchant relationship (rather than just extracting the cheapest price) will face fewer legal obstacles. The builders who win in agentic commerce won't be the ones who bypass gates — they'll be the ones who design agents that gatekeepers want to let through.