On 10 August 2026 the Ninth Circuit Court of Appeals lifted the preliminary injunction Amazon had obtained against Perplexity, the AI search company whose Comet assistant can browse and transact on a shopper's behalf. Source: Global Cosmetics News at https://www.globalcosmeticsnews.com/amazon-loses-bid-to-block-perplexitys-ai-shopping-tools/, with same-week coverage from MLex, ChannelNews and Boston Brand Media.
The reasoning matters more than the outcome. Amazon argued that Perplexity's agent violated federal and California computer-fraud law by accessing its site without authorisation. The court found Amazon unlikely to succeed on that claim, because when the agent logs in and browses, it is the user who is accessing Amazon. The agent is acting for them, with their credentials, at their instruction.
That is the first US federal appeals ruling on whether AI shopping agents are allowed to operate. It says, in effect, that a marketplace cannot treat a customer's chosen assistant as an intruder.
Why this is a beauty story and not a tech story
It is easy to file this under legal news and move on. The reason not to is that it settles a question founders have been quietly hoping would resolve the other way.
Until now, the reasonable assumption was that agent traffic would stay small because platforms would block it. Amazon tried, won at first instance, and has now lost that protection. Every other marketplace watching this case just learned the same thing. Agent traffic is not a pilot that platforms can switch off if it hurts their ad revenue. It is a channel that arrives whether or not anyone in the supply chain wants it.
For a brand at £500k to £5m, that means a growing share of your discovery is being done by software that has never seen your packaging, does not care about your founder story, and will not scroll.
What an agent actually does with your listing
An agent shopping for a customer is doing something narrow and unromantic. It takes a stated need, gathers candidate products, compares them on whatever attributes it can extract reliably, and returns a short list with reasons.
The work happens in that middle step. The agent can only compare on facts it can find and trust. Size, price per unit, full ingredient list, format, claim, and whether the same product says the same thing in more than one place.
This is where most indie beauty listings fall down, and not because they are badly written. They fall down because they were written for a human who is already interested. A person landing on your page has usually arrived via a creator, an ad or a friend. They already have context. They will forgive a listing that leads with mood and buries the actives, because the mood is what brought them.
An agent has no context and no patience. If your 30ml serum lists its size in the fifth image and not in the product attributes, the agent cannot compare your price per millilitre. If your ingredient list is a JPEG, it cannot read it. If your hero claim is "skin that feels awake", it has nothing to match against a shopper who asked for something for dullness.
You do not get filtered out for being expensive. You get filtered out for being unreadable.
The consistency problem is bigger than the content problem
Most brands can fix a thin product page in an afternoon. The harder issue is that agents compare across sources, and beauty brands are unusually inconsistent across sources.
The same serum is 30ml on your site, 1 fl oz on Amazon, and listed under a slightly different product name at a retailer. The price is £42 direct, £38 on the marketplace during a promotion nobody told you about, and £45 at a stockist who has not updated since last year. The claim is "clinically tested" in one place and "dermatologist tested" in another, because two different people wrote the two listings eighteen months apart.
A human shopper never notices this. An agent notices immediately, because comparing sources is the whole job. Inconsistency reads as unreliability, and unreliable products do not get recommended.
This is the part that is genuinely worth a founder's attention this month, because it is not a marketing task. It is a data hygiene task, and it usually has an owner problem rather than a difficulty problem. Nobody at a growing brand is responsible for the fact that the same SKU says three different things in three places.
What to do about it without hiring anyone
Start smaller than feels satisfying. Take your three best-selling SKUs, not the whole range.
For each one, write down the size, the price, the full INCI list, the format, and the single claim you most want to be found for. That is your source of truth. Then open every place the product appears: your own site, Amazon, each retailer that stocks you, and any marketplace listing you did not create yourself. Compare each against the source of truth and fix what does not match.
You will find things. Most brands find at least one listing they had forgotten existed, and at least one price that is wrong.
Then make the facts machine-readable where you control the page. Ingredients as text, not an image. Size in the product attributes, not only in the description. Structured product data on your own site so the size, price and availability are stated in the markup rather than implied by the layout. If you already have product schema in place, check it actually reflects the current price, because stale structured data is worse than none.
The wider shift this ruling confirms
There is a version of this story that ends with founders panicking about being replaced by robots, and it is not the useful version.
The useful version is that discovery is splitting into two motions. One is still emotional and human. It runs on creators, on packaging, on the reason your brand exists, and it is not going anywhere. The other is comparative and mechanical, and it is growing, and it rewards a completely different kind of preparation.
Brands that do well over the next two years will be the ones that stop treating those as the same job. The story sells the product to a person. The data gets the product in front of the person in the first place.
The court has now told us the second motion cannot be blocked at the platform level. That removes the last reason to wait and see.