Agentic Commerce SEO: How Brands Get Products Into AI Shopping Agents

Agentic commerce SEO is the work of making your products findable and buyable by AI shopping agents, not just human shoppers. It matters now because a federal appeals court just ruled that retailers cannot easily wall those agents out. On August 4, 2026, the Ninth Circuit let Perplexity's Comet agent keep shopping on Amazon, and the reasoning applies to every retail site.

The ruling settled a legal question. It did not settle the commercial one. If a shopping agent can walk into any store and buy on a user's behalf, the store that structures its catalog for the agent gets bought from, and the store that assumes a human is reading the page gets skipped. That is the shift, and it is worth understanding before the practical work.

What the Amazon v. Perplexity ruling actually decided

Amazon sued Perplexity in November 2025, arguing that its Comet browser agent "trespassed" on Amazon's servers by shopping and buying on users' behalf without Amazon's authorization. In March 2026, District Judge Maxine Chesney in the Northern District of California agreed it was likely a Computer Fraud and Abuse Act violation, issued a preliminary injunction banning Comet's shopping agent from Amazon.com, and ordered Perplexity to destroy data the agent had collected.

The Ninth Circuit vacated that injunction. In published opinion 26-1444, a three-judge panel held that Perplexity's assistant is "a tool, not a person" for CFAA purposes, and that "it was the user who 'accessed' Amazon's computers, with the help of Perplexity's AI agent." The key mechanical fact the court leaned on: Perplexity itself does not directly communicate with Amazon's servers. The traffic routes through the user's own computer, so the user is the one accessing Amazon, the same way a person using a web browser is. The court also warned that an injunction "would impair consumer choice and needlessly limit development of a nascent technology."

This is the first federal appeals ruling on whether AI agents can browse and transact on retail sites without the site's permission, and the Electronic Frontier Foundation, which filed in support of Perplexity, noted the court acknowledged "little to no existing caselaw" on ascribing responsibility for AI agents under the CFAA.

Read the holding narrowly, because the court did. It decided a CFAA question, nothing more. As Cooley's analysis put it, the court "left open the possibility that the same conduct could be the basis for other types of claims, such as claims sounding in tort or contract," and terms-of-service enforcement remains available to site operators. It also flagged that different facts, like an agent with more autonomy or one that talks to servers directly, could change the outcome. And the fight is not over: Amazon has petitioned the Ninth Circuit to reconsider, arguing the ruling "degraded website owners' ability to set the terms on which powerful, fast-evolving, and potentially destructive AI agents may enter their secure systems."

Why this matters for ecommerce brands

The legal detail is not the point for a merchant. The direction is. The most-cited retail site in the country tried to block a shopping agent using the strongest hacking statute available, and lost. If Amazon cannot easily keep agents out, a mid-size brand's robots.txt and login wall will not either. The realistic planning assumption for 2026 is that user-directed shopping agents will reach your catalog whether you court them or not.

That changes who your product page is written for. A shopping agent does not scroll your hero image, react to your brand story, or get nudged by a countdown timer. It reads structured data, filters on constraints the user gave it ("waterproof, under $80, ships free, in stock"), and returns the item whose machine-readable attributes match. If your price and availability live only in JavaScript-rendered widgets, or your feed says one thing and your page says another, the agent has nothing clean to match against and moves on. This is answer engine optimization applied to a cart: the answer is a product, and the citation is a purchase.

It also runs parallel to what already happened inside Google. When AI Mode started answering product questions with Google-hosted panels, the control point moved from the page to the Merchant Center feed, a shift we covered in how Google cites itself in AI Mode. Agentic commerce extends the same logic across every agent, not just Google's.

How AI shopping agents find and buy products

There are two jobs an agent has to do, and they map to two different pieces of your setup. First it has to find and evaluate your product against a query. Then it has to actually complete a purchase. Discovery runs on structured data. Checkout runs on a protocol.

For discovery, the agent reads machine-readable product data from three main places: on-page schema.org/Product markup, a product feed you send to a platform, and identifiers that let it match your item to the wider catalog. The schema.org Product type carries the fields an agent filters on: offers with price and availability, brand, gtin, aggregateRating, and review. Whether schema markup lifts your visibility is a genuinely contested question for informational content, which we dug into in does schema markup help AI citations, but for products the case is stronger, because the agent needs those exact fields to compare items and cannot invent them.

For checkout, the clearest public example is ChatGPT's Instant Checkout, built on the Agentic Commerce Protocol, an open standard OpenAI co-developed with Stripe. Products enter through a feed: per OpenAI's commerce documentation, merchants send structured product data as CSV or JSON, starting with a sample feed for validation and then daily snapshots covering identifiers, descriptions, pricing, inventory, media, and fulfillment. Some fields are required so ChatGPT displays price and availability correctly; others are recommended, adding rich media, reviews, and performance signals that help ranking and buyer confidence. The merchant stays in control throughout: per Stripe, merchants "can accept or decline the order, charge the payment method, calculate and remit sales tax, and handle fulfillment and returns, as they normally would." At launch, US Etsy sellers were live, with over a million Shopify merchants named as coming soon.

Here is how the surfaces line up.

The through-line: agents read structured data, not marketing copy, and each one has a control point you feed. Compare that to how text-answer engines choose sources, which run on undocumented retrieval pipelines rather than a feed you submit, a mechanic we decoded in how ChatGPT picks its sources. Commerce is more legible than that, because a purchase needs exact, current data and the platforms tell you what they want.

What ecommerce brands do now

None of this requires a rebuild. It requires making the data an agent needs correct, current, and consistent across every place it lives. In rough priority order:

• Put price and availability in plain, structured data. Mark up every product with schema.org/Product, including offers (price, priceCurrency, availability) and brand. If those values only exist inside a JavaScript widget or an image, an agent reading the raw page cannot match your item to a constrained query. Static, parseable facts are the entry ticket.
• Submit and validate a product feed. For ChatGPT commerce, that means a CSV or JSON feed to OpenAI, validated first as a sample, then refreshed on the daily snapshot cadence the docs describe. If you already run a Google Merchant Center feed for Shopping, you have most of the raw material; the work is meeting each platform's spec, not starting over.
• Make your identifiers airtight. GTIN, MPN, and brand are how an agent knows your "Model X running shoe" is the same product a review site rated and a competitor also lists. Missing or wrong identifiers make your item unmatchable, which is worse than a lower rank because the agent never considers it at all.
• Keep feed, schema, and page in agreement. Price, availability, and specs should match to the character across your feed, your on-page schema, and the visible page. Mismatches get products disapproved in feed programs and make an agent distrust the data it did parse. This is the same discipline that governs Google's product panels, and it is unglamorous, ongoing work.
• Feed the attributes agents filter on. Users shop agents with constraints: size, material, shipping speed, returns policy, compatibility. The more of those you express as clean attributes rather than prose buried in a description, the more queries your product can satisfy.
• Earn reviews and mentions where agents can read them. Structured aggregateRating and review data, plus third-party coverage, give an agent corroboration beyond your own claims. That third-party signal has become one of the stronger inputs to AI visibility generally, a pattern we laid out in brand mentions vs backlinks.

The unifying idea: an agent trusts data it can retrieve, parse, and cross-check. Everything above is a way of handing it more of that.

The caveats worth stating plainly

Do not over-read the ruling. It is one circuit, it is narrow, and Amazon is fighting it. The court decided a CFAA question and explicitly left contract and terms-of-service theories on the table, so the legal landscape for agentic shopping is still forming. As of August 2026, treat "agents can reach my catalog" as the working assumption, not a permanent settlement.

Do not over-read the tooling either. Instant Checkout and the Agentic Commerce Protocol are real and shipping, but coverage is early: a subset of merchants, mostly US, with more named as coming. Perplexity's Comet works differently, driving a browser rather than consuming a feed. The specs and participants will move over the next few quarters, so re-verify field requirements and merchant lists before you build to them.

What will not change is the underlying requirement. Whether an agent consumes a feed or drives a browser, it needs your product data to be machine-readable, accurate, and consistent. That work pays off across every agent and every surface, which makes it the safest bet in a space where almost everything else is still in motion. For the broader question of which AI channels actually send buyers your way, see our breakdown of AI referral traffic.

The one-line version

A federal appeals court just made it hard to keep AI shopping agents out of your store, so stop trying, and make your products legible to them: clean schema, a validated feed, airtight identifiers, and price and availability that match everywhere an agent might read them.