Your Product Page Is No Longer the Front Door

For two decades, "can customers find us online?" has largely been answered through SEO and paid search. An entire industry grew up around making businesses findable to search engines that worked the same way. Crawl pages, rank links and then return search results for matching search queries.

A meaningful share of product discovery now happens inside AI assistants and shopping agents. ChatGPT, Perplexity, Gemini, Claude, and a growing number of specialised agents are becoming a real channel. They do not always return the familiar links from your Google searches. They can return one answer, a short list of products, or in some early cases complete a purchase on the user's behalf via emerging protocols like OpenAI's Agentic Commerce Protocol and the Universal Commerce Protocol co-developed by Shopify and Google.

These systems often do not read your website the way Google did. Some still crawl pages. ChatGPT search, Perplexity, and Claude with web search all do, and well-structured pages help them extract your information accurately. That includes JSON-LD product schema, semantic HTML, and clear factual claims. But increasingly, other channels driving AI shopping skip the page entirely and ingest structured product feeds. OpenAI's product feed for ChatGPT is delivered as a file upload to OpenAI's infrastructure rather than crawled from your site. Google's Shopping Graph ingests Merchant Center feeds the same way. In these channels your product page is never visited. Only your data is.

If your business is set up for SEO but not for AI, you risk losing ground in a channel that is growing fast, and the cost of that gap looks likely to rise.

SEO and AI optimise for different things

The simplest way to understand the shift: SEO optimises for retrieval. Did the right URL surface for a query? Did the user click through and read your page?

AI shopping increasingly optimises for extraction and action. Can a model lift a specific factual claim from your data, attribute it correctly, and use it in an answer? Can an agent filter, compare, and transact on it without ever showing the user a page at all?

Different goals lead to different tactics. SEO rewards keywords, headings, internal links, and authoritative backlinks. AI shopping rewards typed attributes, controlled vocabularies, machine-readable feeds, and complete variant-level data. There is overlap. Clean HTML and good structured data help both. But the centre of gravity is moving.

What "legible to AI" actually means

When a customer asks an AI agent something like:

"Find me a women's waterproof running jacket in size medium, in a bright colour for visibility, machine washable, under £150, in stock for next-day UK delivery."

…every constraint in that sentence ideally maps to a structured field in your catalogue. The agent is not necessarily reading your prose. It is filtering on data, then ranking what is left.

That single query touches nine structured fields. Women's maps to an audience field, ideally typed rather than only the word "women's" in the title. Running jacket maps to a category from a controlled taxonomy. Waterproof maps to a typed attribute or rating, ideally not just the word "waterproof" buried in the description. Size medium maps to a sized variant with an explicit size system (UK, EU, US) and group (Regular, Petite, Tall). Bright colour maps to a colour family from a controlled list, ideally alongside the marketing name. Machine washable maps to a care instruction from a defined enum. Under £150 maps to a typed price with currency. In stock maps to variant-level availability, not only parent-product. Next-day UK delivery maps to structured shipping data with destination and delivery time.

A typical ecommerce store will have some of these as structured data and others only in the product description. Constraints that live only in free text are harder for an agent to filter on reliably. That can mean the product gets skipped in favour of one with cleaner data, or surfaces with the wrong attributes.

This is not speculative. It is how Google Shopping's filters already work today. Google Merchant Center has dedicated typed fields for most of the constraints above (gender, size, size system, colour, material, and many more), and for apparel categories several are required. Google has also publicly announced new Merchant Center attributes specifically aimed at conversational commerce, going beyond traditional keywords to include things like answers to common product questions and compatible accessories.

Why this matters commercially

If you sell products online, the gap between catalogues that are AI-ready and catalogues that are not looks set to widen. Early movers can compound, being cited, ingested, recommended, and transacted with by AI agents while less-prepared competitors are harder to find on these channels. The cost of catching up later is likely to be higher than the cost of preparing now. Data debt compounds, and the work is engineering, not a content tweak.

The honest version of the argument also acknowledges the upside if AI shopping evolves more slowly than expected. The same structured catalogue that wins on AI channels is what powers serious analytics, sharper merchandising, and better operations. One investment, two payoffs. We will come back to this in Part III of the series.

For now, the takeaway is simple. AI shopping is no longer a forecast. It is a channel, with its own rules about how it reads your business. Those rules favour structured data over polished copy, typed fields over free text, and machine-readable feeds over keyword optimisation.

The next post in this series gets specific about what "AI-ready data" actually looks like in practice, why most catalogues fall short, and what the gaps tend to be.