Buying agents don't read your brand. They read your feed.

For most of ecommerce history, the product feed has been a back-office file nobody wanted to own.

It lived with the paid media team. It existed to keep Google Shopping happy. It was a spreadsheet of titles, prices, GTINs and availability that got fixed when a campaign broke and ignored the rest of the time. Marketing wrote the lovely brand copy on the website. Someone else, quietly, kept the feed alive.

That arrangement is about to cost people money.

Because the feed is no longer just a PPC admin file. It is becoming the machine-readable version of what your business actually sells. And the machines reading it are multiplying. Google Shopping reads it. Free product listings read it. AI Overviews and AI Mode read it. Gemini reads it. And the thing everyone keeps waving at on stage, the buying agent that shops on a customer’s behalf, will read it too.

Here is the part that should worry you.

Most businesses are running what Search Engine Journal nicely called an unholy trinity. There are three separate sources of truth about your products, and they are owned by three different teams who rarely talk. The product feed in Merchant Center. The structured data on the page. And the actual rendered product page a human sees. When those three disagree, and they almost always disagree, the machine does not throw an error and ask you to clarify. It picks a winner.

It might pick the feed that says out of stock when the page says in stock. It might pick the schema that says one price when the page shows another. It might decide your product is in the wrong category, or that a competitor with cleaner data is the safer thing to recommend. You do not get a vote. You are not even in the room. A system reconciled your contradictions and moved on.

This is the bit the AEO crowd keeps missing while they sell magic files.

The same week this product-feed point surfaced, Google published clearer guidance on optimising for generative AI in Search. The headline, in Google’s own framing, is blunt: from Search’s perspective, AEO and GEO are still SEO. You do not need a special llms.txt. You do not need a markdown copy of your site. You do not need artificial content chunking or a rewritten “AI version” of your pages. The generative surfaces run on the same core Search systems as everything else: indexed content, retrieval, query fan-out, quality systems, anti-spam.

In other words, the lever everyone is trying to sell you, the AI hack, the GEO file, the schema package with a clever name, is mostly theatre. The real work is older and more boring. Make the thing legible. Make it true. Make it consistent across every place a machine can read it.

And nowhere is that more concrete than the product feed.

Think about what a buying agent actually does when a customer says “find me a cordless impact driver under a hundred quid that’s in stock and ships this week.” It does not admire your homepage. It does not feel your brand values. It queries structured data. It compares titles, specs, prices, availability and categories across a set of candidates. It cross-checks the feed against the page to see if the source is trustworthy. Then it shortlists. The whole transaction happens in a layer your marketing team has historically treated as someone else’s job.

If your feed is wrong, you are invisible in exactly the moment a sale was about to happen. Not because your product is worse. Because your data was harder to trust than the next one along.

One of our ecommerce conversations right now is a tools retailer with around eighty thousand products, heavy dependence on paid Shopping, thin technical SEO and a backlog of product content that reads like it was written to win a vague-poetry competition. The instinct, the one the market keeps selling, is “write better product descriptions” or “add an AI SEO layer.” Both miss the point. The valuable job is not nicer adjectives. It is making the product data consistent enough that Google, Shopping, AI answers and a future buying agent can all trust it and all tell the same story.

That is closer to infrastructure than copywriting. Which is exactly where the work is going.

So if you sell anything online, the question to sit with is not “are we doing AEO.” It is “do our three sources of truth agree.” Pull twenty of your products and check the boring fields side by side. Feed title against page title. GTIN and MPN. Price in the feed, price in the schema, price on the page. VAT treatment. Availability. Category. Image. Merchant Center status. I would bet money you find contradictions in the first ten, and that nobody currently owns fixing them end to end.

Co-owning that file is the unglamorous move that actually pays. Not PPC alone. SEO, paid, ecommerce and dev all looking at the same product truth, because every one of those channels now depends on it and the AI surfaces depend on it twice over.

None of this is the exciting version of AI that gets keynote applause. There is no agent with a personality. There is no dashboard that makes you look clever in a meeting. There is just the quiet, commercial fact that the systems deciding what to recommend to your customers are reading a file you have been neglecting, and they will believe it over your marketing every single time.

The brands that win the agentic-commerce shift will not be the ones with the boldest AI claims. They will be the ones whose product data is clean, consistent and true enough that a machine can recommend them without taking a risk.

Fix the truth layer. The adjectives can wait.

---

Jason Sibley is the founder of Cleo, a post-agency marketing and AI company. JasonVsTheNoise is where he writes about what is actually happening with AI, marketing, and how businesses should be thinking about both.