For twenty years, brands have optimised for a shopper who looks at a page. That shopper is being joined by something else: an agent that reads a page, compares it against every other page, and returns one answer.
Most brand teams are treating this as a marketing problem, and handing it to SEO. It is at least as much a channel control problem, and the reason is simple. An AI agent does not see the price you set. It sees the price that is live.
How do AI shopping agents decide what to recommend?
AI shopping agents rank products on structured data, review depth and clear, consistent pricing, and in ChatGPT's case placement cannot be bought at all — the shopping results are organic. There is no equivalent of a sponsored slot to paper over a weak signal. The brand whose data is complete, parseable and internally consistent wins; the brand whose data is contradictory is either skipped or hedged against.
ChatGPT reports roughly 800 million weekly users and around 50 million shopping-related queries a day. Amazon disclosed on its Q1 2026 earnings call that nearly 20% of shoppers who interact with a sponsored brand prompt in Rufus go on to continue the conversation about that brand.
The mechanics differ by platform — Microsoft launched Copilot Checkout in January 2026, Amazon is monetising Rufus, and Amazon and Perplexity have already been to court over whether an agent may shop on Amazon at all. But the input is the same everywhere. Agents read your product data, your prices and your reviews, and they read them across channels.
Why does channel chaos become an AI visibility problem?
An agent asked to recommend a product does what a diligent shopper would never bother to do: it checks every channel at once. Your Amazon listing, your eBay listings, Google Shopping, the retailer feeds. Then it reconciles them.
If your product sits at £49 on your own site, £41 from an unauthorised seller holding the Buy Box, and £38 in a retailer feed, the agent does not conclude that £49 is the real price. It concludes that the price is unstable — and an unstable price is a weak signal to rank on and an awkward thing to recommend. The hedge writes itself: prices vary widely for this product.
Every argument brands already accept about price erosion — that it pressures partners, destabilises the Buy Box and degrades margin — now has a discovery consequence attached. Inconsistent pricing does not just cost you margin. It costs you the recommendation.
What does an AI agent actually see on your listings?
It sees whoever holds the Buy Box, whatever price is currently live, and whatever reviews have accumulated. It does not know, and cannot know, that the offer it is reading comes from a seller you never authorised.
That is the uncomfortable part. If an unauthorised seller holds your Buy Box, their price becomes your price as far as the agent is concerned. If that seller ships in poor packaging with no warranty support, the reviews they generate become your reviews — and review quality is one of the few signals every agent weighs. A rogue seller is no longer just a margin problem sitting quietly on your listing. They are the version of your brand that the agent describes to the customer.
Amazon monitors pricing across Amazon and external retailers, and external prices can influence Buy Box eligibility and pricing decisions. So a single retailer undercut can lose you the Buy Box, hand the offer to someone else, and change what the agent reports — all without anyone at the brand being told.
What should brands do first?
Fix the inputs before optimising the output. Publishing better schema markup while an unauthorised seller holds your Buy Box at a price you never agreed to is decorating a signal that is already wrong.
In practice that means three things, in order. Know who is holding the Buy Box on every listing, every day. Know where your price is inconsistent across channels, and which seller or retailer is causing it. Then close the gaps — through policy enforcement, distributor conversations, and the ordinary work of channel control.
This is the same discipline brand protection has always required. What is new is the cost of neglecting it. Price inconsistency used to erode margin quietly, over quarters. Now it also determines whether an agent recommends you at all — and that judgement is made in milliseconds, on data you did not curate, about a price you did not set.
You can't protect what you can't see. Increasingly, you can't be recommended for it either.
See what Lupa can find on your brand.
We will put together a sample report showing who is selling your products, at what price, and what impact it is having on your channel.
Request a sample report