eCommerce

    AI-ready product pages: the new SEO for e-commerce

    Francesco Luigi Ziccheddu · November 16, 2026 · 7 min
    Online shop desk at night with a screen showing a tidy product page and structured data

    On Friday 13 November OpenAI launched ChatGPT Wallet, a wallet that lets its agents choose a product and complete the purchase on the user's behalf. For now the release is limited: only some Pro subscribers in the United States, with US cards. But the signal is bigger than the launch. A share of the customers landing on your product page is becoming a program that reads, compares and decides. And if it cannot read your page, it buys from someone else. This is the new SEO for e-commerce: not writing for an engine that ranks you, but for an agent that has to trust your data enough to spend a person's money.

    Why the product page is back at the centre

    For years, SEO for an online shop was a matter of keywords, titles and links. The product page was often where the supplier's text was pasted in a hurry before moving on, because the decision was made by a person looking at photos and scrolling. An agent does the opposite. It does not admire the design and it is not swayed by an adjective: it looks for facts. What the product is, what it is made of, what size it is, what it is compatible with, how much it costs, whether it is available, how fast it arrives, on what terms it can be returned.

    If one of those facts is missing, or if the page says one thing and the feed another, the product is dropped or recommended with less confidence. The signal is coming from several directions. Google has introduced conversational attributes in Merchant Center, announced in May, designed to answer the detailed questions people ask AI Mode. OpenAI asks merchants to submit a catalogue feed so ChatGPT can use it for product discovery. Channels change name every quarter, but the principle stays the same: complete, structured, consistent data.

    OpenAI's Wallet also shows where the final step is heading. Every purchase is approved by the user with a biometric check, the card is replaced by a token that neither the agent nor the shop sees, and the merchant receives what it needs to fulfil the order. In practice the agent chooses, the customer confirms with a gesture and the shop ships. The part where you can make a difference is the first one, where the agent chooses, and it is decided almost entirely by the quality of your product page.

    The first obstacle is often a closed door

    Before improving the content, it is worth asking whether agents can get in at all. A Human Forward study published on 6 November used eight AI assistants to query the sites of 200 US retailers and direct-to-consumer brands. About half, 98 out of 200, turned the crawler away or failed to load, and of the 63 that blocked it, 58 stated in their robots.txt file that AI was welcome. Of the 48 large retailers that could be read, only 2 appeared ready for a checkout handled by an agent.

    Metal warehouse door left ajar with an open padlock and a cyan light spilling through

    The sample is American, not Italian, but the mechanism is universal: a firewall configured years ago, an over-zealous anti-bot protection, a rule copied from an old site. The result is a shop that says it welcomes agents and in practice turns them away. The first check is therefore technical and takes an afternoon: verify what a legitimate crawler sees when it visits the site, and decide deliberately who to let through. A bot that copies your catalogue to resell it is one thing, an agent that brings an order is another.

    What an agent-readable product page contains

    A readable page has, first of all, identifiers: brand, manufacturer code and barcode. They are what allows an agent to understand that your product is the same one it has already seen elsewhere and to compare prices without ambiguity. Then a title that says what the product is, without slogans, and attributes in structured fields rather than buried in a paragraph: material, size, colour, compatibility, recommended age, certifications. If a piece of data is needed to decide, a machine must be able to read it without interpreting it.

    Next come the page's structured data, the kind search engines already know, with price, currency, availability, shipping costs and times, and return conditions. The most frequent mistake here is a page with one price and a feed with another, because an update reached one channel and not the other. For an agent, an inconsistency is reason enough to move on to the next product.

    Then there is the most neglected part, the questions. People talk to an assistant the way they would talk to a shop assistant: is it suitable for a five-year-old, will it fit in a small boot, can it go in the washing machine. A page that already answers these questions, in clear sentences and not in a list of slogans, is more likely to be chosen, both by the agent and by the customer who arrives after it.

    The same goes for images and for the terms of sale. A photo with a well-written description tells a machine what it shows, and a clear shipping and returns page, reachable from the product page, spares the agent from having to guess. These are details that seem secondary to a person, who finds them only at the basket. An agent looks for them before choosing, and a shop that hides them looks like a risky shop.

    One catalogue, and the page is the shop window

    In my view the underlying mistake is thinking of the product page as a text to be written page by page. In a shop with hundreds or thousands of products it does not work: the page, the feed for search engines and the marketplace listings must all come out of a single catalogue that holds the truth about each product, in orderly fields. Those who work this way update a piece of data once and see it change everywhere. Those who work with three versions of the same product, one on the site, one in the spreadsheet and one in the head of whoever follows the orders, discover the problem only when an agent picks the wrong version.

    It is a job of organisation before it is a job of technology, and it is the kind of work in which we at BusinessLabs, with custom software and automation, help those who sell online: put the data in one place and make it travel. The agent comes later, and it is only as good as the data it works on.

    How to start in thirty days without redoing the whole catalogue

    There is no need to fix everything at once, and two weeks before Black Friday it would be unrealistic anyway. In the first week I would do the access check, meaning what a legitimate crawler sees, and pick the fifty products that bring the most margin, not the most revenue, because those are the ones worth being chosen for. In the second I would complete the identifiers and structured attributes for those fifty. In the third I would add answers to the questions customers ask most often, which customer service already knows. In the fourth I would check consistency between page, feed and marketplace.

    Hands ticking a paper calendar next to a laptop showing a grid of products

    At the end there is a test that costs ten minutes: ask an AI assistant the questions a customer would ask and see whether it finds your product and what it says about it. If the product does not appear, or appears with wrong data, you know where to step in without needing any study.

    OpenAI's Wallet is for a few people today, and in a year it will probably no longer be news. My position is that the product page is infrastructure, not a text. Those with complete, consistent data will be ready for whichever agent arrives, while those without will have to redo everything under pressure, at the moment when customers are already elsewhere.

    Francesco Luigi Ziccheddu
    Author
    Francesco Luigi Ziccheddu
    CEO & Co-Founder

    CEO & Co-Founder of BusinessLabs. I have attended major international trade shows since 2015 and share here experiences and insights for online sellers.

    LinkedIn

    Talk to our Amazon agency

    Book a free consultation