Amazon vs Perplexity decides who your visitors are

On June 11, the Ninth Circuit hears oral arguments in *Amazon v. Perplexity*. The narrow question is whether Perplexity's Comet browser violated the Computer Fraud and Abuse Act by logging into Amazon on a user's behalf and shopping for them. The actual question is bigger than that, and the SEO industry has barely registered it.

The question is whether an AI agent acting under a user's authorisation counts as an authorised visitor to your website. If the Ninth Circuit says no, the agentic future Google demoed at I/O has a legal problem that nobody is pricing in. If it says yes, every retailer, marketplace, and booking platform is about to lose the ability to decide who gets to access their logged-in surfaces.

Either outcome rewires the assumptions the industry has been quietly building strategy on for the last twelve months.

What the case actually turns on

Amazon's argument leans on the CFAA, a 1986 statute written to prosecute hacking-style intrusion. Amazon says its terms of service govern who is authorised to access logged-in areas of amazon.com. A human is authorised. An agent acting for that human is not, because the terms don't extend permission to delegated software. Under this reading, Comet logging into your Amazon account on your behalf is functionally the same as someone you've never met logging into your Amazon account on your behalf — even if you told them to.

Perplexity's appellate brief calls this "a fundamental misfit." Their argument is that the user is the authorised party at all times, that Comet acts under delegated authority, and that contractual terms can't manufacture federal criminal-law violations out of an agent doing what its user told it to do. Mozilla and the EFF filed amicus briefs supporting that position.

The Northern District of California sided with Amazon at the preliminary injunction stage in March. Roughly a week later, the Ninth Circuit paused that injunction pending appeal. Appellate stays of preliminary injunctions are not routine. They are the first signal a higher court has doubts.

That's where we are. A district judge accepted Amazon's theory. The appellate court is sceptical enough to have paused the injunction. Oral arguments are in ten days.

The Van Buren problem Amazon has to get around

In 2021, the Supreme Court decided *Van Buren v. United States* and narrowed the CFAA in a way that matters here. The Court held that a person with permission to access a system does not violate the CFAA by accessing it for the wrong reason. The decision was a deliberate pushback against years of CFAA stretch, where the statute had been used to prosecute behaviour — scraping, account sharing, terms-of-service breaches — that bears no resemblance to 1986-era hacking.

The real question isn't whether agents can shop on Amazon. It's whether the laws written for human visitors apply to software acting on their behalf.

The real question isn't whether agents can shop on Amazon. It's whether the laws written for human visitors apply to software acting on their behalf.

Amazon's theory has to survive *Van Buren*. The argument is that the user has permission to access their Amazon account, but the agent doesn't, because the agent is a separate entity that the terms never authorised. This is a clean legal argument and also a strange one, because in practical terms the agent is a tool the user is using to do something the user is allowed to do.

If the Ninth Circuit holds that delegated software access is fine when the human is authorised, *Van Buren* extends naturally to agents. If it holds the other way, every website's terms of service becomes a tool for blocking agentic traffic at the federal-law level, and the entire agentic commerce architecture Google rolled out at I/O has a legal exposure problem nobody has briefed clients on.

Why this lands on SEO, not just legal

The instinct is to file this as an interesting legal case that doesn't affect anyone's day job. That instinct is wrong, for three reasons.

First, Google spent I/O demonstrating Universal Cart, agentic booking, and information agents that monitor listings in the background. The Universal Commerce Protocol launched earlier this year as an open standard for this kind of activity. Sundar Pichai told Patrick Collison that search is becoming an "agent manager." All of this assumes agents can access merchant sites under user authorisation. If the Ninth Circuit says they can't — at least not where terms of service object — Google's roadmap collides with American federal law for any merchant who chooses to invoke it.

Second, the case is the first one. The Ninth Circuit's ruling will be cited in every subsequent dispute over agent access for years. Retailers watching closely already understand that the precedent is what they're buying. A favourable ruling for Amazon gives every site a CFAA-shaped weapon to deploy against unwanted agent traffic. A favourable ruling for Perplexity strips that weapon away and forces sites to compete on whether they're worth the agent's attention rather than whether they can keep it out.

Third, the question of who counts as a visitor is the question SEO has always answered. We've been answering it for humans and Googlebot for two decades. We're about to need an answer for ChatGPT agents, Comet, Gemini, Claude, and a long tail of agents nobody has named yet. The CFAA case decides whether that answer is a technical decision or a legal one.

What this means for the next twelve months

If you advise clients on technical SEO, you're about to need a position on agent access. The defensive instinct — block everything, write terms that explicitly prohibit agentic visitors — is the Amazon strategy. It might win in court. It might also lose the customers who delegate their shopping to an agent and never see your site again because the agent routed around you.

The offensive instinct — open up to agents, structure your data so they can transact cleanly, treat them as a discovery channel — is the Perplexity-aligned strategy. It assumes the Ninth Circuit either rules for Perplexity or, if it rules for Amazon, that the practical reality of agentic commerce overrides the legal exposure. That's a bet.

Most clients don't yet know they need a position. The pattern I've seen across consulting work for the last eighteen years is that legal questions become marketing questions about six months after a court rules, and by then everyone is reacting rather than planning. This is one of those moments where the planning window is open.

A few practical implications worth thinking through now.

If you run an e-commerce site, decide whether your terms of service should explicitly address agent access, and decide it before a court tells you what your terms mean. If you run a content site, your exposure is smaller but not zero — the same theory could apply to agents accessing paywalled or member-only content. If you advise on schema, structured data, and merchant feeds, the Universal Commerce Protocol is going to either expand rapidly or stall on legal uncertainty, and the difference is roughly ten days away.

The attribution crisis from I/O assumed the agent could get to your site. The CFAA case is whether that assumption holds.

The honest limits of this argument

I don't know how the Ninth Circuit is going to rule, and anyone telling you they do is guessing. Appellate stays of preliminary injunctions are a signal but not a verdict. The panel could land on a narrow ruling that decides this case without setting broad agent-access precedent. They could also remand on procedural grounds and we end up exactly where we started, eighteen months from now, with a different bench.

It's also possible the case settles before oral argument. Amazon and Perplexity both have reasons to want this resolved on their own terms rather than at the panel's discretion. A settlement leaves the question open, which is the worst outcome for clarity but a real possibility.

What I'm confident about is that this case — or one structurally identical to it — decides the agentic commerce question in the United States within the next eighteen months. The question doesn't go away. The only variable is which court answers it and how broadly.

The one to watch

The SEO industry has spent the year arguing about citations, attribution, and whether AI Overviews count as a click. Those are real questions. They are also downstream of a more basic one: whether the agents doing the citing and the clicking are allowed to be on your site in the first place.

On June 11, three judges in Seattle start answering that question. Whatever they say will shape the next decade of how websites think about who their visitors are, and whether the distinction between a human and a piece of software acting on a human's behalf is one the law is willing to make.

That's the case. It's the most important thing happening in search this month, and almost nobody outside the legal press is writing about it.