Why your product pages might be invisible to AI agents

When ChatGPT or Google AI Mode gets a query for your category - like "fragrance-free mauve cream blush under $30" or "cordless drill under $200 with brushless motor" - it doesn't read your product page. It reads the feed your store sends to the agent.

For Shopify stores, Shopify Catalog generates this feed automatically and contains title, description, category, and metafields for every product.

That's what the agent has to decide whether your product matches the query. Photography, brand copy, and campaign descriptions are not in that payload.

Whether your store shows up depends entirely on whether those fields are structured, specific, and correctly assigned. I have audited thousands of brands and the pattern is consistent - the data exists, it's just in the wrong places for an agent to use it.

Why your PDP copy isn't working for agents

The copy on your product page does its job for humans. Descriptions have voice, titles carry brand identity, and attributes get mentioned throughout.

→ fragrance-free formula
→ velvet upholstery
→ brushless motor included

But they're embedded in sentences, not set as structured field values. When an agent filters for fragrance-free, it looks for a field value, not a word in a paragraph.

The fields that need to be filled are the category metafields in your Shopify admin, and the fastest way to know what to put in them is to look at how a marketplace in your category has already done it.

The fastest way to know what to fill

The benchmark is already on a marketplace product page in your category. Home Depot, Amazon, Wayfair - depending on what you sell, find a product from a well-established brand and look at the structured attributes section, the table, or the specification list that the marketplace requires sellers to fill.

Those fields exist because marketplaces solved structured product retrieval way before ChatGPT and other AI agents - to match a specific intent to a specific product in a catalog of hundreds of thousands. Agents are solving the same problem with the same data. The attributes that power marketplace search are the same ones AI agents filter by.

Shopify's Standard Product Taxonomy was mapped from this structure. When you assign a product to its correct category in Shopify admin, the metafields for that category appear automatically and mirror what marketplaces require for the same product type.

I have recorded a short video - comparing a product on Home Depot with the Shopify metafields for the same furniture category.

The same principle applies across every category.

Here is what it looks like when every field is correctly structured.

What the agent reads on Sephora's product page

The query from a shopper: "fragrance-free mauve cream blush under $30".

That query never touches the product page. It runs against the feed Sephora pushes through the Agentic Commerce Protocol. For Shopify stores, the same feed is generated automatically by Shopify Catalog, mapped from the fields in your product admin.

Here is how each filter in that query will map to a field in their feed and in Shopify:

1. Category (cream blush) - product_category in the ACP feed. In Shopify, your product category on the Standard Product Taxonomy, not the Makeup > Cheek > Blush breadcrumb your theme renders.

2. Variant and color (mauve) - a variant grouped under item_group_id with color set as an option value. In Shopify, a real Color option on the variant, not a word sitting in the title.

3. Structured attribute (fragrance-free) - attribute in ACP / metafield in Shopify.

4. Price (under $30) - offer.price and priceCurrency. In Shopify, the variant price.

5. Stock - availability on the variant (InStock or OutOfStock) - refresh on OpenAI feed every 15 minutes. One of the shades, Sleepy Girl, was OutOfStock - so per-variant availability routed to an in-stock shade instead of dead-ending.

6. Ranking - aggregateRating, ratingValue and reviewCount - the structured rating value and review count are what the agent reads.

Every field above is either one you set in Shopify and the feed carries, or one you left in prose and the agent never sees. An attribute that isn't in the feed is one that the agent can't use, no matter how clearly it's printed on the page.

Where to start (fix your catalog)

• Category first - go through your products and accept Shopify Magic's taxonomy suggestion or assign it manually. This has the highest downstream impact of anything on this list. It unlocks the metafields, establishes the taxonomy node the agent classifies you under, and makes your products eligible for category-level matching on every UCP-compatible surface at once.

• Color and variant attributes - check whether color and other variant attributes are set as variant options and metafields, or whether they currently only exist in titles and description text.

• Category-specific attributes - for category-specific attributes like fragrance, closure type, or active ingredient, use a marketplace PDP from a strong brand in your category as the template. The fields they've filled are the ones you need.

• Per-variant availability - check that stock is tracked at variant level, not product level. A product-level availability setting gives the agent nothing to route with when one variant is unavailable.

I have built a tool that shows which fields are empty across your catalog. If you want to see the full gap → Take a free AI audit

Filling the fields at scale

Manually reviewing and assigning taxonomy categories and structured attributes across hundreds or thousands of products is where most merchants stop.

Catalog Genius handles this automatically. It reads your catalog, assigns the correct Standard Product Taxonomy category to each product, and enriches the relevant metafields based on what the product actually is. Interested?

→ Schedule a call with me

- Ankit