AI visibility for ecommerce: product data first, content second

AI visibility for an online shop is won on product data accuracy first and content second. When somebody asks an assistant for the best product for a job, the answer is assembled from retailer pages, marketplace listings, review sites and editorial round ups, and your product appears only if its current price, stock and specification are readable in plain text on a crawlable page. This page covers the five query shapes shoppers use, the feed and markup work that decides whether you are eligible, and how to stop an assistant quoting a price you stopped charging last year.

The five shapes a shopping question takes

Shape	Example phrasing	What decides the answer
Best for a use case	Best running shoe for flat feet under a set budget	Editorial round ups, review sites, and category pages with real specification text
Product comparison	Model A compared with Model B for durability	Specification tables and independent reviews that mention both
Where to buy	Who stocks this in my country right now	Current stock and price data, and your presence on marketplaces
Fit and compatibility	Will this part fit this machine	Structured compatibility data in readable text, not a lookup widget
Brand or shop quality	Is this retailer reliable, what is the returns policy	Review platforms, your own policy pages, and consistent trading details

Only the first two are about marketing. The other three are about data, and that is where most shops lose. Our audit of where shopping recommendations come from is in where ChatGPT shopping recommendations get their sources.

Product data is the work

An assistant answering a product question needs facts it can commit to. Five things make your catalogue readable.

• Price and availability in the HTML. If price loads from a script after the page renders, treat it as a price the fetcher may never see. View source on a product page and search for the number.
• Product and Offer structured data that matches the visible page. Google asks that markup match what a user sees, so markup showing a sale price the page does not show is a defect rather than an advantage. Validate with the Rich Results Test.
• One canonical URL per product. Variant URLs for colour and size multiply the same product into a dozen weak pages. Pick the canonical and let the variants point at it.
• Specification in text, not an image. Dimensions, materials, compatibility and care instructions as a list. A specification sheet in a PDF or a JPEG is a specification nobody can quote.
• A feed that matches the site. Merchant feeds, marketplace listings and your own pages should carry the same price on the same day. When they disagree, the stale one is the one that gets quoted back at you.

That last point is the most common complaint we hear from retailers, and the diagnosis is in why AI quotes an old price.

Category pages carry more weight than product pages

Buyers ask for a category with a constraint far more often than they ask for a product by name. That makes the category page the highest value page in the shop for this work, and most of them are a grid of tiles with 40 words of text.

A category page that gets quoted has four things: a title that matches how buyers describe the category, two or three opening sentences that say what is in the category and who each option suits, a short comparison table of the real decision criteria, and a buying guide section that answers the questions a shop assistant would be asked. Keep the grid, add the words above and below it.

Reviews and third party presence

A claim that exists only on your own domain has one source behind it. For a shop that means four places are worth getting right.

• Review platforms. A complete, verified business profile with current trading details. The platform page is structured and easy for a model to read.
• Marketplaces. Your listings on large marketplaces are often better indexed than your own product pages, so keep the title, specification and images current there too.
• Editorial round ups. The "best X for Y" articles that assistants lean on for use case questions. Being in the round up is a pitch and a sample, not an SEO task.
• Community threads. Reddit and forums come up repeatedly in citation lists. A transparent, named presence that answers questions is worth more than a thread nobody from your company has ever read.

Why these move answers is set out in why listings and reviews move AI answers.

Access, in fifteen minutes

Shops are more likely than most sites to be blocking AI fetchers, because bot management is turned up hard to protect checkout and stock. Check robots.txt on every host, read the access log for OAI-SearchBot, GPTBot, PerplexityBot and ClaudeBot by name, and confirm your bot rules are not refusing them at the edge. The robots file is only half the answer, since a CDN rule can override it. The copyable blocks are in how to write a robots.txt.

Measuring it

Build a prompt set from the five shapes above, with your brand name in none of them, and run each prompt several times in two or three assistants every month. Record whether you were named, which retailer was named instead, and every source cited. The cited sources are your pitch list for the next quarter. AI Knows Us runs this across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews and then writes the pages, and its free scan needs no account.

This page covers AI visibility for an online shop. For organic search mechanics, see ecommerce SEO. For a walkthrough by business type, see how an ecommerce brand gets recommended by AI.

The AI visibility for ecommerce questions worth an answer

Do assistants recommend specific products or only categories?

Both, and the trigger is how the question is framed. A question with a constraint in it, such as a budget, a size or a condition, usually produces named products. A broad question about a category often returns guidance with no names, which is why your prompt set should contain constrained questions.

Does Merchant Center or a product feed get me into AI answers?

A feed keeps your price and stock correct in the surfaces that read feeds, which is valuable in its own right. For Google's AI surfaces, Google states that a page must be indexed and eligible to be shown in Search with a snippet, so the crawlable product page is still the thing that makes you eligible.

Why does an assistant show an old price?

Because it is reading a source that still carries the old price: a cached page, a marketplace listing, a stale feed or an affiliate page. Fix it at the source rather than arguing with the output, then publish the current price in plain text with the date you checked it.

What about product reviews on my own site?

They give a model text to summarise about a product, so they are worth having and worth marking up accurately. Independent reviews carry further, because they count as a separate source saying the same thing.

How many pages do I need to write?

Start with your top ten category pages and the two or three comparison questions where a competitor is named and you are not. Ten strong pages beat a hundred thin ones, because retrieval rewards a page that answers one question properly.

What to do first

• View source on three product pages and confirm price, stock and specification are in the HTML.
• Run the Rich Results Test on those three pages and fix any mismatch between markup and visible text.
• Compare the price on your site, your feed and your largest marketplace listing for ten products today.
• Add two opening sentences, a comparison table and a buying guide to your top three category pages.
• Ask five constrained buying questions in two assistants without naming your shop, and save every cited source.