Can AI Shopping Assistants Read Your Product Pages? A Checklist

Lean ecommerce teams keep asking the same question in two forms. Should we fix product data before we spend on AI visibility? And what, exactly, do we check on a product page to know whether ChatGPT, Google AI Mode or Perplexity can read it? This checklist answers the second question, and in doing so it usually settles the first.

Fix the data first, or spend on visibility first?

Being recommended by an AI assistant is downstream of product information. OpenAI's shopping guidance says ChatGPT selects products using structured metadata from first-party and third-party providers along with other third-party content, and then ranks merchants on availability, price, quality and whether they are the maker or primary seller. Google's guidance for AI Mode and AI Overviews recommends complete, current Merchant Center product data as the way to appear on those surfaces.

So the order is not really a choice. Visibility work that starts before the product data is readable spends money measuring an absence. Visibility work that starts once the pages and feed are in order measures the effect of the fixes. The checklist below is the audit a Paz expert runs first, and a lean team can run most of it on its own.

What AI shopping assistants read on a product page

Structured data. Google documents merchant listing structured data for product pages: Product and Offer markup with name, image, price, currency, availability and identifiers. This is the machine readable layer. Without it an assistant is guessing from raw page text.

The attributes a shopper uses to choose. Materials, dimensions, fit, compatibility, what is in the box, care and warranty. The Merchant Center product data specification lists the fields Google reads, and the right set depends on the category. A running shoe needs support and drop. A sofa needs dimensions and fabric. A missing attribute is a shopper question your product cannot answer.

Conversational context. Google's conversational attributes are optional Merchant Center fields for product specific questions and answers, manuals and related accessories. In Google's own test with lululemon, submitted conversational attributes were incorporated 50% of the time in relevant AI Mode recommendations, per the announcement published in mid September. That is evidence Google uses the information, not a traffic promise.

Content the crawler can see. Google's documentation on JavaScript and search explains that Googlebot queues pages for rendering, which can delay when script loaded content is read. Other assistants document their crawlers but not their rendering. Product facts that only exist after client side scripts run are the facts most likely to be missed.

The most common gaps

Default schema that stops at price and availability. Most platforms emit basic Product markup. Variants as separate offers, identifiers, ratings and shipping are the parts that are usually missing. The Rich Results Test shows what is and is not present on a given URL.

Product details behind a script. Specification tabs, size guides and Q and A widgets that load on click or after render. Fetch the page with scripts disabled and read what remains. If the specifications are gone, so is the assistant's view of them.

Thin descriptions. "Blue cotton t-shirt. Machine washable. Available in S to XL." A shopper asking for a breathable shirt that will not shrink in humid weather needs the page to say breathable, humid and shrink resistant, in words, not in a lifestyle photo.

No questions and answers on the page. When someone asks whether a blender crushes ice, the assistant looks for an explicit answer. A page that answers that question beats one that lists a motor wattage. Put the Q and A on the product page, where every assistant can read it, and in the feed where Google supports it. The two are complementary, not alternatives.

Missing relationships. Accessories, compatible items and replacement parts that are sold but never linked from the main product. Assistants make contextual recommendations, and products without defined relationships are left out of them.

The checklist

Structured data

• Product name, description and brand present in Product markup
• Price and currency specified, matching the visible price
• Availability status present and matching stock
• SKU and GTIN or MPN where they exist
• Review rating and count where you collect reviews
• Variants expressed as separate offers with their own price and availability
• Markup validated with the Rich Results Test on a representative URL per template

Content depth

• Description states use cases, materials and fit in the words shoppers use
• Specifications cover the attributes a shopper in your category compares
• The common pre-purchase questions are answered on the page
• Specifications and answers are present in the server rendered HTML
• Accessories and compatible products linked from the product page

Distribution

• Merchant Center feed complete, approved and current, with Google's guessed categories confirmed or corrected
• Conversational attributes submitted where you have product specific Q and A
• Shopify Catalog connected, or an OpenAI product feed in place where that applies to you

Crawl access

• robots.txt allows OAI-SearchBot, which OpenAI says is the crawler that decides whether a site appears in ChatGPT search results
• robots.txt allows PerplexityBot, which Perplexity says surfaces and links sites in its results
• GPTBot and Google-Extended decided separately, since both are training crawlers and OpenAI states the settings are independent
• No firewall or bot protection rule blocking the published IP ranges of the crawlers you allow

How to prioritize

Do not audit the whole catalog. Start with the products that carry your revenue this quarter and your planned promotions, usually a few hundred items, and run every check above on that list. Fix at the source, in the platform or PIM that generates the page and the feed, so the fixes survive the next sync.

Then expand by template rather than by product. If one product template fails the structured data checks, every product on that template fails them, and one fix covers them all. The long tail follows once the templates are right.

Before you fix anything, record how your shortlist of products appears when you ask ChatGPT and Google AI Mode the questions your customers ask, and note the Merchant Center readings in the same scope. We explain how to read those readings in what 0% and 100% share of voice really mean. Check again after the fixes ship. Appearances, AI referred sessions and orders are three separate numbers.

Frequently asked questions

Should product Q and A live on the product page or only in the feed?

On the page first. Every assistant can read the page, and only Google reads conversational attributes in Merchant Center. Google also says you do not need to duplicate information already in the description or product detail fields. Put the answers on the page, then submit the same facts as conversational attributes where they add information.

Do I need to block or allow AI crawlers?

Allow the search crawlers if you want to appear in answers. OpenAI documents OAI-SearchBot for search and GPTBot for training and says a site can allow one while disallowing the other. Perplexity documents PerplexityBot the same way. Treat the training crawlers as a separate policy decision.

How many attributes does a product page need?

Enough to answer the questions a shopper in your category asks before choosing. There is no published threshold from OpenAI or Google. Read your reviews, your support tickets and your competitors' pages, list the attributes shoppers compare, and make sure each one is stated in text on the page and present in the feed.

This is the audit Paz runs for lean ecommerce teams before any AI visibility work begins. A Paz expert finds the gaps in your product information, across the page, the structured data and the feed, and proposes the fixes from your own sources with the evidence behind each one. You approve, and the platform publishes the changes and tracks how your products appear afterward. If you want to see where one product stands first, run a free AI readiness report. Otherwise, talk to us.