Perplexity Comet vs Amazon: The May 15 Ninth Circuit Hearing and Your Merchant Playbook

What the lawsuit is really about

Amazon filed in the Northern District of California in November 2025. The complaint targets Comet, the agentic browser from Perplexity, for letting AI agents sign into a user’s Amazon account with the user’s own password and complete purchases on their behalf. Amazon’s core federal claim sits under the Computer Fraud and Abuse Act, the CFAA.

On March 10, Judge Maxine Chesney granted Amazon a preliminary injunction. Comet can no longer use a password-authenticated session to drive an agent through Amazon product pages and checkout. Perplexity appealed immediately, and the Ninth Circuit set oral argument for May 15.

The reason this case matters beyond Amazon is simple. The ruling will decide whether a user authorizing an AI agent to act on third-party sites with their own credentials counts as unauthorized access under the CFAA. If the Ninth Circuit sides with Amazon, every buyer-side agent path into a logged-in platform becomes legally risky. That includes Comet, ChatGPT Agent, and Claude computer use. If the court sides with Perplexity, federal precedent effectively green-lights agentic commerce.

For context on the agent itself: Comet Agent runs Claude Sonnet 4.6 by default, with Opus 4.6 available on Max subscriptions. Comet iOS pre-orders just opened. By the time a ruling lands, this is not a niche product. Millions of shoppers will be routing buying intent through it.

Two outcomes, two merchant scenarios

Whichever way the court goes, sellers should be ready. Here are the three realistic paths and what each means.

RulingImpact on Amazon sellersImpact on DTC sellersAction window
Injunction upheld (Amazon wins)Agent traffic stays off Amazon listings, internal SEO weights stableDTC sites lack a password gate, agent traffic flows to DTCStart agent-readiness build now
Injunction reversed (Perplexity wins)Comet can place orders, listings must become agent-friendlyHeavy competition, agents compare Amazon and DTC prices side by sideRework product feed and schema
Remanded (most likely)Uncertainty period of at least 6 to 9 monthsBoth sides must prepHedge in both directions

Remand is the highest-probability outcome. The Ninth Circuit has trended toward narrow CFAA readings over the past two years, and hiQ v. LinkedIn still controls. Amazon’s fact pattern is different though. Comet uses credentials the user voluntarily provides, not public scraping. The panel is likely to find the factual record thin and send it back to the district court.

Remand is also the hardest case for merchants because there is no final answer but you cannot afford to wait. The checklists below assume hedging rather than betting on either side.

DTC site agent-readiness checklist

This work matters regardless of the ruling. ChatGPT Agent and Claude computer use are not constrained by the Amazon case, and they will road-test the DTC path first.

First, your product feed has to be machine-parseable. Every product page needs complete schema.org Product markup with price, inventory, SKU, variants, shipping, and return policy as structured data. Do not render price only through client-side JavaScript. Many agent runners use headless modes where JS execution is uncertain, so the more information that lives in plain HTML, the better.

Second, checkout must support guest flow and autofill-friendly forms. An agent running visual operation through Claude Sonnet 4.6 handles standard labels like “Email” and “First Name” cleanly. Forms that use placeholders as labels, iframe-wrapped fields, or forced account creation will tank agent conversion rates.

Third, expose an agent-friendly endpoint. A simple read-only /api/products.json or an llms.txt index lets agents grab your catalog without rendering pages. Shopify added agent permission toggles in admin recently; self-hosted stores need to build this themselves.

Fourth, move returns and shipping policies into short, fixed-URL pages. Agents check policy before checkout. Policies buried in footer PDFs are invisible to them.

Hedging strategy for Amazon sellers

For Amazon sellers, the logic inverts. In the near term the injunction holds, so PPC and organic ranking weights on platform stay stable. But you have to assume a reversal is possible within six months, so hedging starts now.

The core hedge is diversifying traffic sources. Pull your traffic report. If more than 90 percent comes from Amazon internal search, you depend entirely on an entry point that agents could rewrite. Aim to push off-Amazon sources to at least 20 percent across Google Ads, Meta, TikTok, and owned email.

The second hedge is standing up a DTC presence, even a thin one. Do not wait for the ruling. A Shopify or self-hosted mirror with your hero SKUs gives you a landing surface if Comet wins and agents start price-comparing away from Amazon at scale.

Third, accelerate Brand Registry enrollment and A+ Content. If the ruling allows agent-driven orders, agents will rank by brand identity rather than individual listing. Sellers with a Brand Store outperform generic listings in that scenario.

Last, watch Amazon’s own moves. Amazon is testing Rufus, its first-party shopping agent, in parallel with the Comet lawsuit. The intent is clear: keep agent traffic on platform. If Amazon opens a Rufus API to sellers, integrate immediately.

What to do before May 15

May 15 falls on a Thursday. No ruling drops that day. Ninth Circuit opinions after oral argument typically take two to six months. Treat May 15 as a signal day, not a deadline. A few items should still be done before the hearing.

First, run your DTC site through Google Rich Results Test and the Schema.org validator. Fix every error. Switch product pages to server-side rendered price and inventory.

Second, brief support and ops. If Comet or another agent places an order, the request may carry an unusual user-agent or routing pattern. Make sure those are whitelisted and not flagged as fraud.

Third, register for Anthropic, OpenAI, and Perplexity partner programs. Onboarding bars are low now. They get higher once the ruling lands and merchants pile in.

Fourth, listen to the oral argument on May 15 through Court Listener or PACER. The judges’ questions reveal direction two to three months before the formal opinion. That early signal is worth more than any analyst write-up.

This case is the first federal definition of property boundaries in agentic commerce. The legal gap between “a user manually clicking through a browser” and “an AI agent doing it for them” gets defined starting May 15. Merchants who prep both sides come out ahead of merchants who bet on a direction.

Related Articles