How to write a product page an AI can cite

A growing share of buyers no longer types a query: they ask a question. And what answers is not a list of ten links, it is an engine that builds an answer and cites a few sources. If your product is not among those sources, for that person it does not exist, even if you rank first on Google.

The practical question is: what does a product page need in order to be one of those sources.

What an answer engine takes

A generative engine does not summarise your page out of politeness: it looks for fragments it can repeat without getting them wrong. In practice it takes four kinds.

Data stated plainly. Measurements, weight, material, compatibility, consumption, what is in the box. Written as data, not drowned in a promotional paragraph. "Brushed steel frame, 42 cm diameter, fits half-inch fittings" is quotable. "Designed with attention to every detail" is not, because it says nothing anyone can repeat.

The question with its answer next to it. If the page contains the question a person would ask and the answer immediately below, that block is already in the shape required. It is the same reason it works on Google, but sharper here: the engine does not have to interpret, only to quote.

The comparison. "What is the difference between this and that" is a typical question for an answer engine, and not a question for a list of links. If the page does not contain the comparison, the engine builds it from somebody else's sources.

The limit. This is the counterintuitive part: writing when a product is not the right choice raises your chances of being cited. An engine that has to recommend also looks for constraints, and a source that states them is more useful than one promising everything to everybody.

Why a beautiful description does not get quoted

Pages written to persuade are built on adjectives, and adjectives cannot be repeated. An engine quoting "elegant and versatile" has told its user nothing, so it does not quote it. It prefers the boring sentence that contains a number.

That is an uncomfortable habit change for anyone who writes copy, and it should be said clearly: copy has not stopped mattering. Copy is for the person reading the page. Data is for whoever cites it. Both live on the same page, but if only one is there, on the new surface you are invisible.

What to put on a page, in practice

No theory, just the list of what is most often missing.

Complete attributes, including the obvious ones. An empty field never gets quoted by anyone, and "obvious" is a judgement you make knowing the product, not one the person searching makes.

The questions your support inbox receives. They are already written down: they are the real questions, asked in the real words. A page that answers those also answers the ones put to an engine.

Service terms on the page, not three clicks away. Shipping, returns, warranty, installation. The doubt appears on the product page, and that is where it has to be resolved.

The use context. What it suits, what it does not, what it works alongside. That is the information that lets an engine recommend it for the right case instead of skipping it.

What cannot be measured yet, and is worth saying

Honesty is needed here, because the industry is avoiding it. On traditional search you have positions, impressions and clicks: you know whether you are improving. On generated answers, today, there is no reliable and widely available tool that tells you how many times a page was cited, by which engine, and for which question.

Anyone selling you a precise count of "AI citations" is estimating, and should say so. What can genuinely be done is more modest and still useful: run the questions that matter in your category and see who gets cited, watch the traffic arriving from AI chat referrers, and measure the things that remain measurable, namely catalogue completeness and performance on Shopping and search.

That is not a reason to skip the work. It is a reason not to let anyone promise you results nobody can verify yet.

Frequently asked questions

What is the difference between SEO and GEO?

SEO gets you into a list of results. GEO gets you quoted inside an answer, when a generative engine answers instead of listing. The first rewards titles and descriptions, the second rewards complete data and explicit answers. It is one piece of work on the same page.

Do I need to add structured data to my pages?

It helps, and on Shopify most themes already generate some. But structured data describes what is on the page: if the information is missing from the page, marking it up does not create it. Content first, markup second.

Do I risk duplicate content by answering the same questions across many products?

You do if you paste the same paragraph onto a thousand pages. You do not if the answer contains that product's data, because then every answer is different, and it is also the only useful version: a generic answer serves neither the reader nor the engine.

Should I write for ChatGPT or for Google?

The question is badly framed, and it is the trap of the moment. The work is the same: complete data and explicit answers about the product. What changes is what the recipient does with it, not what you have to write. Optimising for one engine means chasing behaviour that changes every few months.

Where do I start with thousands of products?

Not with the top hundred, which somebody already fixed by hand and which already rank. Start with the categories where buyers' questions are most technical, because those are where a thin page costs the most and where complete data shows the most.