The Attribute Gap: Why AI Search Skips Incomplete Listings

AI shopping reads fields, not keywords

When a buyer asks an AI shopping engine to compare products, the engine does something a human shopper never does: it reads the listing as a set of discrete fields, then filters and compares them. Brand, identifier, price, availability, size, color, shipping speed, and return policy each become a lever the engine pulls. If a field is empty, that lever does nothing — and the listing quietly drops out of the comparison.

Google’s Merchant Center product data specification is unusually blunt about this. It states that Google “uses this data to match your products to the right queries” and that “incorrect, inaccurate, or missing product information can cause disapprovals, limited eligibility, incorrect displays for your products.” The spec names the same offenders repeatedly: a wrong product category, a missing GTIN, and “missing or incorrect variant attributes (such as item group id, color, or size).”

This is the attribute gap — the distance between the standard fields AI engines match on and the fields your listings actually carry.

The evidence: entity authority is your weakest signal

FirstShelf scores every listing across four dimensions. An anonymized audit of 21 marketplace listings over a 90-day period found that entity authority — the dimension measuring whether a listing carries the factual attributes AI engines match on — averaged just 30.8 out of 100. That was the lowest of the four dimensions, trailing platform compliance (49.5), semantic density (41.9), and structure quality (31.7).

The most commonly missing attributes were exactly the ones comparison filters depend on:

• Authenticity guarantee (missing on the most listings in the sample)
• Size and dimensions
• Software compatibility
• Editability — what can and cannot be changed
• Delivery method (download versus access)
• License or usage terms (personal versus commercial)

In plain terms, the average listing lost more than half of its possible entity-authority score before the AI ever read a sentence of its description. No amount of persuasive copy can recover a listing that the engine cannot match in the first place.

What AI shopping actually does with your attributes

Google confirms that AI Mode is “particularly helpful for queries where further exploration, reasoning, or complex comparisons are needed,” and that both AI Overviews and AI Mode may use a “query fan-out” technique — issuing multiple related searches across subtopics and data sources to build an answer.

Picture a buyer asking for running shoes under $120 in size 9 with free returns. Fan-out fires several sub-searches: one filters by price, one by size, one by availability, one by return policy, then the reasoning layer compares what survived. Every sub-search pulls on a specific attribute. A listing that never declares a size cannot survive the size-9 filter. A listing whose return policy lives only in a paragraph marked “easy returns” cannot survive a structured returns filter. Only a declared field survives a structured filter — prose never does.

This is why attribute completeness, not keyword density, is now the gatekeeper for AI shopping visibility.

The four attribute groups that decide eligibility

Google’s product data specification groups attributes into categories. Four of them decide whether a listing is matchable in AI shopping:

1. Product identifiers

Brand, GTIN, and MPN are the keys an engine uses to recognize your product as a unique entity and group it across sellers. Google’s spec notes that “your product’s category type determines which unique product identifiers (GTIN, MPN, brand) are required,” and that omitting them can disapprove a listing entirely. Without an identifier, the engine cannot confirm two sources are describing the same product, which weakens the corroborated entity profile that drives recommendations.

2. Price and availability

Price and availability are non-negotiable filters. Google’s Product structured data documentation shows that shoppers can see “price, availability, review ratings, shipping information” directly in results. A listing missing price or availability is invisible to every price-bounded and in-stock query — which is most comparison queries.

3. Variant attributes

For products sold in multiple versions, variant attributes — item group id, color, size, material — tell the engine which listings are siblings of the same parent product. Google recommends product variant structured data “to help Google better understand which products are variations of the same parent product.” Missing variant data is one of the spec’s named causes of limited eligibility, and it is the single most common reason an AI comparison shows the wrong color or size.

4. Shipping and returns

Shipping cost, speed, and return policy are the filters buyers add when they narrow a shortlist. Google recommends declaring these as structured policy data — “Share shipping costs, especially free shipping” and “Share return information, such as your return policy, fees involved in returns, and how many days customers have to return” — because shoppers weigh total cost, not sticker price. A listing that omits shipping and returns loses the final, highest-intent filter.

Why new “AI markup” is the wrong fix

The tempting response to AI search is to bolt on something new — a dedicated AI file, a chunking scheme, or a special AI-only schema. Google’s AI features guidance pushes back directly: “There’s also no special schema.org structured data that you need to add” to appear in AI Overviews or AI Mode, and “You don’t need to create new machine readable files, AI text files, or markup to appear in these features.”

The attribute gap is not a missing-markup problem. It is a completeness problem with the markup Google already documents. The Product structured data vocabulary (brand, gtin, color, size, material, offers, aggregateRating) and the Merchant Center feed attributes (gtin, mpn, color, size, shipping, returns) already describe everything an AI comparison needs. The work is filling in the fields you skipped, not inventing new ones.

Crucially, Google says some shopping experiences combine page structured data with Merchant Center feed data — “product snippets may use pricing data from your merchant feed if it’s not present in the structured data on the page.” That means the two sources reinforce each other, and a gap in one can sometimes be covered by the other. Completing both is the highest-coverage move.

Your attribute-completeness checklist

Run every listing through this checklist before it goes live:

• Declare a brand and a product identifier (GTIN where one exists, MPN and brand otherwise). If the product genuinely lacks identifiers, set the identifier-exists flag to no rather than leaving the field blank.
• Add a category using Google’s product taxonomy. A wrong or missing category is one of the spec’s named causes of limited eligibility.
• Set price and availability in the Product offer. Every comparison query filters on at least one of these.
• Declare variant attributes — item group id, color, size, material — for every version of a parent product.
• Add shipping details (cost, speed, free-shipping flag) and a return policy as structured policy data, not just as body copy.
• Fill the attributes that buyers in your category actually filter on. For digital and marketplace goods, that includes delivery method, license or usage terms, software compatibility, and what is included in the purchase — the exact fields most often missing in the FirstShelf sample.
• Keep your structured data and Merchant Center feed in agreement. Google flags conflicting data between feed and website as a top eligibility problem.

How to verify before you publish

Catching the gap before the engine does takes two free checks. Run each listing URL through Google’s Rich Results Test to confirm the Product structured data parses and that required properties are present. Then use Merchant Center diagnostics (the Issue Details Page) to surface missing or incorrect attributes at scale — it lists exactly which fields are missing, conflicting, or malformed.

Pair both with the new Generative AI performance report in Search Console, which shows which of your pages appear in AI Overviews and AI Mode. If a listing you expected to surface is absent, the attribute gap is the first place to look: a listing that fails a filter never reaches the ranking layer at all.

The takeaway

AI search did not invent a new requirement for product listings. It exposed one that was always there. Buyers have always filtered by size, price, and returns — AI simply enforces those filters automatically and at scale. The sellers who win in AI shopping are the ones who treated product attributes as data all along, not as optional decoration. Close the attribute gap, validate it, and your listings become matchable in the comparisons that now drive discovery.