An AI shopping assistant recommends the product it can describe with confidence, and it gets that confidence from your product page: a plain statement of what the product is, specs stored as attributes, facts that settle a comparison, real questions answered in text, honest reviews, and variants that each resolve to the right price and stock. This guide covers how to write and structure each part.
It assumes the groundwork from our post on getting your store ready for ChatGPT and AI shopping agents: catalog access, Shopify's AI channel settings and crawler rules. This one is about the page itself.
Why do AI shopping assistants send shoppers straight to product pages?
Because the product page is the unit every AI shopping system is built around. Feeds, catalogs and structured data all describe one purchasable item and point to one URL, and that URL is the product detail page. When an assistant recommends something, the click lands on the PDP, not your home page or a collection.
You can see it in the specs. OpenAI's product feed reference asks for one row per purchasable item or variant, and its required url field is the "product detail page for the item, with the variant selected when possible." Google splits product structured data into product snippets and merchant listings, and merchant listings are for "pages where customers can purchase products from you." On Shopify, ChatGPT sends shoppers to your own online store checkout, as covered in the earlier post.
What should the top of an AI-ready product page say?
The product's name, what it is, who it is for, and the one or two facts that set it apart, in plain sentences above the fold. Write the first paragraph so it still makes sense if an assistant quotes it alone, with no image, tab or carousel around it.
OpenAI's feed spec asks for a "factual product description for this item" in plain text, and notes that ChatGPT may generate simplified product titles and descriptions because merchants describe the same product in different ways. Copy that states facts gives it something accurate to work from.
Compare two openings for the same bag:
- "Designed for the modern explorer, our iconic carry-all redefines everyday adventure."
- "A 28-liter waxed-canvas backpack with a padded sleeve that fits a 16-inch laptop, built for daily commuting and weekend trips."
The second answers three questions an assistant is likely to be asked: how big, what material, will my laptop fit. Keep the brand voice for the paragraph after.
The title should say what the item is ("Waxed canvas backpack, 28L"), not only a collection name, and this copy has to be in the page HTML. Google's guidance on AI features in Search lists "making sure that important content is available in textual form" as a best practice, and the safe assumption is that other AI crawlers read no more than Google does. Text hidden in images, or loaded only after a click on a tab, is the usual miss.
Which product attributes should be fields, not prose?
Anything a shopper filters or compares on: dimensions, weight, material, color, size, capacity, compatibility, care instructions, age group, condition and identifiers like GTIN and MPN. Store each as its own field, then show it on the page as a readable spec list. Fields feed catalogs and schema; the visible list is what crawlers and shoppers read.
The attribute list in OpenAI's feed reference is a useful checklist because it shows what an assistant can use when stored as data: material, color, size, gender, age group, product dimensions (not shipping box dimensions), weight, condition and product category. Units are not inferred, so "12" with no unit is noise.
On Shopify, category metafields are the native place for this. They map to Shopify's Standard Product Taxonomy, so choosing the category Shirts gives you fields like neckline and sleeve length, and they can be connected to your variant options. If your attributes already live in custom metafields or are baked into product titles, Shopify's Catalog Mapping can point the catalog at them without a rebuild. On BigCommerce, custom fields do the same job as name-value pairs on the product page.
How should you write comparison facts?
State the facts a shopper uses to choose between your product and the closest alternatives, in numbers and plain terms: capacity, runtime, fit, weight, warranty length, what is in the box, what it works with. Assistants are often asked to compare. They can only compare what is written down.
Good comparison facts are specific and checkable:
- "Fits laptops up to 16 inches (36 x 25 cm sleeve)."
- "Works with iPhone 12 and later. Does not support Android."
- "Machine washable on cold. Do not tumble dry."
- "Includes the charging cable. Wall adapter sold separately."
Comparisons with your own range belong here too. If you sell three blenders, a short "which one is right for you" block on each PDP that names the differences (motor wattage, jar size, price) helps a shopper and gives an assistant a clean answer to "what is the difference between model A and model B." A confident, wrong fact on your page gets repeated by every assistant that reads it.
Do FAQs on product pages help AI recommendations?
They help when the questions are real and the answers are visible on the page. Pull them from support tickets, chat logs, returns reasons and on-site search, and answer each in one or two direct sentences. Invented FAQs written to hold keywords help nobody, and FAQ markup alone is not a ranking lever.
There is a reason to put this text on the page rather than rely on a feed. OpenAI's feed reference says "raw review entries and question-and-answer lists are not part of this discovery contract"; the feed carries review counts and ratings only. If you want sizing advice, compatibility answers and care questions to inform an answer, they need to be on a page an assistant can read.
Good PDP questions are narrow: "Does it run small?", "Is the lid dishwasher safe?", "Will it fit a queen bed with a 14-inch mattress?" Keep shipping and returns answers consistent with your policy pages.
How do reviews affect AI product recommendations?
Reviews are one of the inputs. OpenAI says ChatGPT's shopping results draw on structured metadata from third parties, such as pricing, product descriptions and reviews, and that it may show review summaries. The practical moves are to collect more genuine product reviews, show them on the PDP as text, and mark up only what is really there.
Three rules keep this honest and useful:
- Keep review text on the page. Google's review snippet guidelines say marked-up review content must be "readily available to users from the marked-up page." Reviews that load only inside a widget after the page renders may not be read at all.
- Keep product reviews and store reviews apart. OpenAI's feed tells merchants to "exclude seller and store reviews" from product review counts. Google ignores star markup for reviews an organization controls about itself on Organization pages.
- Never fabricate. Google requires that "ratings must be sourced directly from users." No seeded reviews, no rating schema on products with zero reviews. Our post on Shopify product structured data covers the markup side in detail.
How should variants be set up so assistants pick the right one?
Give every variant its own price, availability, image and, where possible, its own URL or URL parameter, and use the same option names across the product. When a shopper asks for "the black one in size 10," the assistant needs a variant that resolves to that exact item and its real stock status.
OpenAI's feed asks for a separate row per variant, sharing a group ID, with each row carrying "its own price, availability, URL, and images." Google says product variant structured data "can help Google better understand which products are variations of the same parent product." The common problems:
- One image for every color, so the variant shown does not match the one described
- Option names that drift ("Colour" on one product, "Color" on another, "Shade" on a third)
- Colors split into separate products with no link between them. Shopify's Catalog Mapping can group separate products as variants of one listing
- Out-of-stock variants still marked in stock in the page markup
Where should you start?
Pick your ten best-selling products and fix those pages first, in this order: a factual first paragraph, a spec list built from fields, three to five real questions answered, reviews visible as text, and variants that resolve correctly. Then roll the template across the catalog.
Before you edit anything, run the free DYODE store scan. It reads your home, collection and product pages as raw HTML, the way a non-JavaScript crawler does, and flags whether the product title and price are in that HTML, whether Product schema with a price is present, whether robots.txt lets GPTBot, PerplexityBot, ClaudeBot and Google-Extended in, and whether a return policy is linked from the product page. For the crawler side of that, see our guide to robots.txt and llms.txt for AI crawlers. Once the pages are fixed, track ChatGPT, Perplexity and Gemini traffic in GA4 to see which PDPs assistants actually send shoppers to.
DYODE is a Shopify Plus and BigCommerce partner, and PDP templates and product data are a core part of our AEO and GEO optimization work. If you want a second set of eyes on your product page template, talk to us.
About DYODE. Founded in 2016 by former heads of E-Commerce at PacSun and Tilly’s. Shopify Premier Partner.
Add DYODE as a preferred source on Google
Google shows you more from the sources you pick.
