AI Shopping Agents Are Back on Amazon. Your Listing Is the Part That Matters.

On August 4, 2026, the Ninth Circuit vacated the injunction that had kept Perplexity's Comet browser off Amazon since March. The court held that when an AI agent shops on your behalf, you are the one accessing Amazon's servers, not the company that built the agent. Comet can operate on Amazon.com again while the case continues in district court.

The ruling settles less than the headlines suggest. The panel decided a narrow question about the Computer Fraud and Abuse Act and said outright that it was not building a legal framework for agentic AI. Amazon's other claims survive, and a footnote preserves Amazon's ability to control agent access through its own terms of service.

For sellers, the practical answer is that nothing on your listing needs restructuring this week. The exposure sits in your advertising measurement, not your organic rank, and the listing work that pays off here is work you should already be doing for Amazon's own AI layer.

Here is what changed, what did not, and the one number worth watching.

What the Ninth Circuit Held, and What It Left Open

The panel ruled that Perplexity does not "access" Amazon's computers under the CFAA because Perplexity's servers never touch Amazon's servers directly. Comet runs on the shopper's machine, takes screenshots of what the browser already loaded, and sends those screenshots to Perplexity for instructions on what to click next. The court called the Assistant a tool rather than a person, and the statute punishes a person who intentionally accesses a protected computer.

That mechanical detail carries the whole ruling. Amazon's theory required treating the agent's autonomous behavior as Perplexity's own conduct, and the court declined, partly on the rule of lenity and partly for a reason worth sitting with: Amazon's reading could have exposed ordinary shoppers to criminal liability for aiding unauthorized access to a store they were logged into.

What the panel left open is broad. It expressly did not rule on tort claims, did not decide whether a different factual record might produce a different answer, and reversed only the preliminary injunction. Amazon's trademark and state-law claims go back to the district court intact.

Amazon Can Still Block Agents. It Just Cannot Call It Hacking.

Footnote 5 of the opinion is the part sellers should read. The court preserved Amazon's ability to regulate access to its store through private terms of service, and said only that Amazon is unlikely to win by invoking federal and California computer-fraud statutes to do it.

That distinction matters because the underlying dispute was never really about the law. It was about a user-agent string, the small identifier that tells a website an AI agent is driving. Perplexity chose not to send one, which meant Amazon could not selectively block agent traffic the way it blocks a crawler.

Amazon has blocked dozens of external agents, including OpenAI's, and it will keep doing so. The tools available to it are contractual and technical rather than statutory. Expect the blocking to continue, and expect the fight to move to terms of service and detection rather than the courtroom.

Why Is Amazon Fighting Agents While Buying Ads Inside ChatGPT?

Because the two behaviors serve the same goal. Amazon wants the shopper to arrive at Amazon.com and complete the purchase inside Amazon's own environment, where its ad inventory lives. An agent that browses, compares, and checks out on the shopper's behalf skips that environment. A sponsored placement inside ChatGPT that sends the shopper to a product page does not.

During Prime Day 2026, Amazon bought sponsored placements inside ChatGPT to promote the event, and Adobe found that shoppers arriving at retail sites from AI chatbots were 40% more likely to buy than shoppers arriving from search, email, or social. That was the first Prime Day where AI-referred traffic converted best of any channel.

So the position is consistent even if it looks contradictory. Referral traffic from an assistant is welcome. An agent operating inside the store is not. Every seller strategy question here follows from understanding which side of that line a given AI surface sits on.

The Real Exposure Is in Your Advertising Data, Not Your Organic Rank

Amazon argued in its complaint that agent activity forces it to build detection systems to identify and exclude automated traffic before advertisers get billed, because advertisers pay for human impressions. That is Amazon's own framing of the risk, and it points directly at the metric sellers should watch.

If agent traffic returns at any volume, the noise lands in impression counts before it lands anywhere else. You will not see a line item labeled "agent" in your reports. You will see impressions that behave oddly relative to clicks, and a click-through rate that drifts for reasons your bid changes do not explain.

The practical move is to baseline now. Pull a clean 30-day impression, CTR, and conversion snapshot per campaign before agent volume changes, so that if something shifts in Q4 you can tell whether it came from the market or from measurement. Our guide to reading Amazon business reports without drowning in them covers which four metrics carry that diagnosis.

What Should Sellers Change Right Now?

Nothing structural, and that is a deliberate position rather than a shrug. The listing content that an external agent parses well is the same content that Amazon's own assistant parses well, and Amazon's assistant is the surface with actual volume behind it today.

That means the work is already on your list. Product attributes filled out completely rather than partially. Bullets that answer a use-case question instead of stacking adjectives. A title that survives the 75-character limit while keeping the entity and the primary use case intact, which our breakdown of how to write an Amazon title that ranks and converts walks through field by field.

The one genuine addition is query-shape research. Agents and assistants receive full sentences, not two-word searches, and the phrasing shoppers use with an assistant differs from what they type into the Amazon search bar. Mining those question-shaped queries is now part of finding the terms that drive Amazon sales, and it is the cheapest change on this list.

If you want the full picture of how AI discovery works across both Amazon's internal layer and external assistants, our Amazon AI optimization guide is the deeper resource, and our breakdown of how Alexa for Shopping reshaped listing discovery covers the surface that already sits in every Amazon shopper's search bar.

The Uncomfortable Part

Agentic commerce on Amazon is a story about legal posture and platform control, and the seller-side impact is currently small. Anyone selling you an urgent agentic-readiness program right now is selling ahead of the evidence.

What is not small is the direction. AI-referred shoppers went from the worst-converting cohort to the best-converting cohort inside a year. The brands that show up in those conversations are the ones with listings a machine can parse and third-party coverage a machine can cite. That was true before August 4, and the ruling did not change it.

Watch the impression data. Fix the listing fields. Skip the panic.

---

FAQ

Will agent traffic show up in my Search Query Performance data?

Not as a labeled segment. Amazon does not break out automated traffic in seller-facing reports, and its stated approach is to filter automated impressions out of advertiser billing rather than surface them to sellers. If agent volume grows, the effect appears as unexplained movement in impression-to-click ratios, which is why a clean pre-Q4 baseline is worth pulling now.

Does the ruling change anything about Alexa for Shopping?

No. Alexa for Shopping runs inside Amazon's own store and was never part of the dispute. The case concerned external agents operating on Amazon from outside, and Amazon's internal assistant sits on the opposite side of that line.

Should I add schema markup to my Amazon listing so agents can read it?

You cannot. Amazon controls the markup on the product detail page, so the fields you fill in Seller Central are your only input. Structured data is a lever on your own brand site and on the third-party pages that assistants cite, and both feed AI visibility indirectly.

If agents compare prices across retailers, does my Amazon price need to match my DTC site?

Not necessarily, but the gap becomes visible in a way it was not before. Assistants surface cross-retailer comparisons in a single response, so a meaningfully higher Amazon price now competes against your own storefront inside the recommendation itself. Model the comparison before you set Q4 promotional pricing.

Is the Perplexity case over?

No. The Ninth Circuit vacated a preliminary injunction and sent the case back to the district court. Amazon's trademark and state-law claims were not decided on appeal and remain active, so the litigation continues on separate grounds.