A shopper asks an assistant for a waterproof walking jacket under one hundred and fifty pounds that will handle Scottish weather and pack down small. What comes back is not a category page. It is three or four named products with reasoning attached, assembled from retailer listings, review sites, forum discussions and specialist buying guides.
Whether your products appear in that answer depends less on your category page optimisation than on whether the specific attributes in that question exist, in plain text, somewhere a system can retrieve them. This article covers how to make that true across a range.
The general principles sit in our guide to generative engine optimization. What follows is specific to retail.
What Changes in AI Shopping Behaviour
Three shifts matter commercially.
Queries carry constraints. Shoppers describe a situation rather than a product. Budget, use case, body type, room size, dietary requirement, compatibility. A system answering that question needs product data granular enough to filter against those constraints.
The shortlist forms before the visit. By the time a shopper reaches your product page, the comparison may already have happened elsewhere, using information you did not write.
Third party sources carry the evaluative weight. Your listing establishes what a product is. Reviews, buying guides and forum discussion establish whether it is any good. No system treats your own description as evidence of quality.
Google also notes that generative AI responses can include product listings and product information, and points retailers towards Merchant Center feeds as a way of helping products be visible in AI responses alongside other Search results. That is a straightforward, documented action rather than an inference.
Product Information Quality Is the Foundation
Most retail AI visibility problems are data problems. A product cannot be matched against a constraint that is not recorded anywhere.
Here is a typical listing and a rewritten version.
Weak:
“Summit Trail Jacket. Our bestselling waterproof jacket combines technical performance with everyday style. Perfect for adventures in all conditions. Available in three colours.”
Stronger:
“Summit Trail Jacket. A 3-layer waterproof shell rated to 20,000mm hydrostatic head with 20,000g breathability, weighing 340g in a men’s medium and packing into its own chest pocket to roughly the size of a one litre bottle. Fully taped seams, YKK AquaGuard front zip, helmet compatible hood with wired peak. Cut for layering over a mid-weight fleece. Suits hillwalking and multi-day hiking in sustained rain. Not insulated, so it is a shell rather than a winter jacket. Sizes XS to XXL, men’s and women’s fits, three colours. £139.”
The second version can be matched against every constraint in the original question. The first cannot be matched against any of them. It is also more useful to a shopper, which is the recurring pattern in this entire subject.
The Attributes Worth Capturing
Work through your catalogue and check that these exist as text, not only as filter values buried in a faceted navigation system:
- Precise measurements and technical specifications with units
- Materials and composition
- Compatibility, where relevant
- Care, maintenance and lifespan
- Sizing guidance including fit notes, not just a size chart
- Country of manufacture and any certifications
- What the product is designed for, stated explicitly
- What it is not suitable for
That final item is the one nearly everyone omits, and it does disproportionate work. A statement such as “not suitable for dishwashers” or “does not fit the 2023 model” is precisely the kind of specific, checkable claim a system needs when it is trying to avoid recommending the wrong thing.
Pricing and Availability
Both change frequently, which makes them the information most likely to be wrong in an AI answer about your products.
The risk is stale retrieval. If a system holds a cached version of your page from three weeks ago, it may tell a shopper a product costs £99 when it now costs £129, and cite you as the source of the error.
Practical mitigations: keep price and stock status in the page’s server rendered HTML rather than loading it entirely through client side scripts, maintain accurate product structured data and Merchant Center feeds, and avoid burying the price in an image or a widget. Where you run frequent promotions, be clear about what is a permanent price and what is temporary.
Product Comparisons Within Your Own Range
Shoppers regularly ask which of two products in the same range they should buy. Most retailers publish nothing that answers this, so the comparison happens on a forum instead.
A short comparison covering the genuine differences between your models, including who should buy the cheaper one, is unusually valuable content. It is also commercially sensible, because a shopper steered to the correct product returns fewer items.
Weak: “The Pro model offers enhanced features for the discerning user.”
Stronger: “The Pro adds a second thermostat zone and a larger 2.4 litre tank. If you brew for one or two people, the standard model does the same job for £80 less. The Pro is worth it if you regularly make more than four drinks in a session or want to steam milk and pull a shot without waiting for temperature recovery.”
Reviews and User Generated Content
Reviews serve two functions here. They provide evaluative evidence that your own copy cannot, and they surface vocabulary and use cases you never thought to write about.
Read your reviews for retrieval value, not just sentiment. If customers repeatedly mention that a jacket runs small, or that a mattress suits side sleepers, that is a constraint appearing in real shopper questions. It belongs in your product content as well as in the review section.
Ensure reviews are rendered in a way that can be crawled rather than loaded exclusively through third party scripts. And do not manipulate them. Fake reviews carry regulatory exposure in the UK under consumer protection law, quite apart from the credibility problem.
Category and Collection Pages
Category pages that contain nothing but a product grid and a paragraph of keyword filler contribute little.
The version that earns retrieval is one that genuinely helps someone choose: what distinguishes the options in this category, which characteristics matter for which uses, what the sensible price bands are and what changes as you move between them. Write it as a buying guide that happens to sit on a category page.
Original Buying Advice Is the Real Differentiator
Product data gets you matched. Buying advice gets you cited.
Retailers hold knowledge that manufacturers do not: which products get returned and why, which sizing runs true, what customers ask before purchase, which combinations work, what fails after two years. Almost none of this is published.
Content built from that knowledge cannot be replicated by anyone summarising manufacturer specifications, which is exactly the position worth occupying. A guide on choosing a walking jacket for British conditions, written by someone who has processed ten thousand returns, is genuinely better source material than the fiftieth restatement of what hydrostatic head means.
This is the same argument made in our article on how interior design businesses get cited by AI search engines, applied to retail: publish the operational knowledge, because it is the part nobody else can produce.
Brand and Merchant Signals
Systems also form a view of the retailer, not just the product. Delivery terms, returns policy, guarantees, customer service reputation and trading history all feed into whether a shop is worth recommending.
Make the practical details explicit and easy to find: delivery costs and timescales, returns window and who pays for return postage, warranty terms, and a real UK address and contact route. Keep your Google Business Profile and Merchant Center details accurate and consistent with your site.
Structured Data, With the Correct Expectation
Use product structured data covering price, availability, reviews and variants. Do it because it supports rich results and merchant listings in ordinary search, which remains a substantial traffic and revenue channel.
Do not implement it expecting an AI citation effect. Google states that structured data is not required for its generative AI features and that no special schema is needed. It is good practice with clear benefits elsewhere, not an AI lever.
A Practical Sequence
- Audit whether product pages are indexed, crawlable by AI search crawlers, and rendering price and stock without requiring script execution.
- Identify your top hundred products by revenue and rewrite their descriptions to include specific attributes and explicit unsuitability notes.
- Add comparison content for products customers regularly confuse.
- Turn your highest traffic category pages into genuine buying guides.
- Publish two or three pieces of original buying advice drawn from returns and support data.
- Verify Merchant Center feeds and structured data are accurate and current.
- Track a prompt set covering constrained product questions, category questions and brand questions, using the approach in our guide to measuring AI search visibility.
Frequently Asked Questions
Should I rewrite manufacturer supplied product descriptions?
Where you can, yes. Identical manufacturer copy appears across dozens of retailers, giving a system no reason to prefer your listing. Adding your own fit notes, use case guidance and unsuitability warnings creates differentiation that duplicated copy cannot.
Do I need to be in Google Merchant Center?
Google’s guidance points retailers towards Merchant Center feeds as a way of helping products appear in AI responses and other Search results. For any retailer of scale it is worth maintaining regardless.
How do I stop AI answers quoting outdated prices?
Keep pricing in server rendered HTML, maintain accurate feeds and structured data, and ensure product pages are recrawled regularly. You cannot control cached versions directly, so the aim is to reduce the window during which stale data exists.
Are long product descriptions better for AI search?
Length is not the variable. Google states there is no ideal page length. What matters is whether the specific attributes a shopper might constrain on are present and clearly stated.
What about marketplaces like Amazon and eBay?
Marketplace listings are frequently retrieved and often outrank retailer sites for product queries. Keep them accurate and detailed, while accepting that your own site is where you can publish the buying advice a marketplace listing cannot accommodate.
Published by BrandingX UK.
