A courtroom just answered a question every online retailer will face
On 4 August 2026, the Ninth Circuit Court of Appeals in San Francisco ruled that when someone sends an AI shopping agent onto Amazon to browse, compare and buy on their behalf, it is the shopper who is legally “accessing” Amazon’s site, not the company that built the agent. The case was Amazon versus Perplexity, and it is the first time a federal appeals court has directly ruled on who is responsible when an AI agent walks into your shop uninvited.
We have been building AI automations for small businesses for two years now, and this is the ruling we have been waiting for. Not because the legal reasoning is exciting on its own, but because it tells every retailer, tradesperson and service business with a website exactly where they now stand.
What actually happened
Amazon sued Perplexity in March 2026, arguing that its AI browser assistant, Comet, was breaking US computer-hacking law by scraping and buying on Amazon’s site while disguised as an ordinary browser. Amazon won an injunction and Comet was blocked. Perplexity appealed, and five months later the Ninth Circuit overturned that injunction entirely, ruling that Amazon was unlikely to succeed on its core legal claim.
The court’s reasoning has already picked up a name: the browser analogy. Perplexity’s agent takes a screenshot of what the user’s own browser sees, sends it to Perplexity’s servers, then sends navigation instructions back to the user’s machine. Perplexity itself never talks to Amazon’s servers directly. The judges compared this to Safari or Chrome: nobody would say Apple is “accessing” Amazon just because someone uses Safari to shop there, and the same logic now covers an AI agent acting on a user’s instructions.
That is a real narrowing of what a retailer’s terms of service can do to keep AI agents out. If a court will not treat the agent’s operator as the one doing the accessing, blocking agents by lawsuit gets much harder. Blocking them by contract or technical means is still possible, but the legal fallback that Amazon reached for first has just been weakened.
Why this matters even if you have never heard of Perplexity
Most of the commentary on this ruling is written for Amazon-sized businesses and their lawyers. That misses the point for everyone else. Every online storefront, booking page and price list now has to assume that some of its visitors are not people at all, they are agents reading on a person’s behalf, at a speed and scale no human browsing session could match.
That cuts two ways. An agent that can read your site accurately is an agent that can recommend you, and recommendation without a human clicking through your ads is exactly what agentic commerce promises. But an agent that reads your prices, stock levels and delivery terms can also be a competitor watching you in real time, or a scraper feeding a comparison tool you never agreed to appear on.
The UK’s own Small Business Commissioner has already published guidance on getting “agentic ready”, and the advice is refreshingly unglamorous: you do not need a new platform or a big budget, you need your product names, prices, stock status, delivery and returns information to be accurate and consistent everywhere they appear, because that is the data an agent actually reads before it decides whether to recommend you.
The uncomfortable number behind all of this
Here is the statistic that should make every UK retailer pay attention. New PayPal-commissioned research found that 64% of UK consumers say they want to use AI agents for shopping, but separate Forrester consumer research puts the share who actually trust an AI agent to complete a purchase on their behalf at just 24%. That is a forty point gap between appetite and trust, and it is exactly the gap the Amazon-Perplexity case sits inside.
Shoppers want the convenience. They do not yet trust the mechanism. That trust gap is not going to close because a court ruled in Perplexity’s favour, it closes when businesses make their own data trustworthy enough that both the human and the agent reading it get a straight answer. A retailer whose stock status is wrong half the time, or whose delivery promise on the website does not match what happens at the checkout, is going to lose out to a competitor an agent can actually rely on.
What we would actually do about it
We are not telling clients to rush out and build an “AI agent strategy”. Most small businesses do not need one yet. What we are doing is treating this ruling as a prompt to check three things that were already good practice before agentic commerce existed:
First, audit what an AI tool sees when it looks at your business today. Ask ChatGPT, Gemini or Perplexity itself what your prices, opening hours and delivery terms are, and see whether the answer matches reality. Second, make sure your website, Google Business Profile and any marketplace listings say the same thing, because inconsistency is what an agent (and a human) reads as unreliable. Third, read your own website’s terms of use and check they say something sensible about automated access, because after this ruling those terms are doing more legal work than most business owners realise.
None of that requires new software or a large budget. It requires treating your published information as if a machine is reading it literally, because after 4 August 2026, in the eyes of a US appeals court at least, that machine is legally standing in for a real customer.
The bottom line
The Amazon-Perplexity ruling will not change how most UK small businesses trade tomorrow. But it has settled, for the first time, that AI shopping agents are here to stay and that the law currently treats them as an extension of the person using them, not the company that built them. Retailers who get their product data straight now will be the ones an agent can confidently recommend once UK shoppers close their own trust gap. The businesses who wait for a UK ruling of their own to force the issue will be doing this same audit later, under more pressure, with less runway.