How product schema tells AI shopping agents which products to recommend

A product can be perfect for a customer, but if it does not have complete schema markup, AI shopping agents will never recommend it. The agent reads your HTML like a human would, but it understands your product through structured data. No schema equals no understanding.

When a customer asks an AI shopping assistant to find a laptop bag that fits under an airplane seat and costs less than $100, the agent does not browse the web like a person. It reads structured data from product pages. It searches for Offer schema with specific prices, Product schema with dimension attributes, and AggregateRating schema with customer reviews. If your product page has a beautiful description but no schema, the agent skips it entirely.

This chapter covers how product schema works, what markup AI shopping systems actually read, and how to structure your product data so AI agents can find and recommend your products.

What is product schema and why AI agents depend on it

Product schema is machine-readable code that tells AI systems what your product is, what it costs, how it is rated, and where customers can buy it. Instead of asking an AI agent to guess whether "laptop bag" means a briefcase or a backpack, schema tells the agent explicitly. This product has wheels. It has a 15-inch laptop compartment. It costs $89.99.

Schema markup uses a format called JSON-LD. JSON-LD is a way of writing data that any computer system can read without having to understand HTML or design. An AI crawler visits your product page, reads the HTML, and also extracts the JSON-LD script tag. That script tag contains structured data. The agent parses that data and understands your product instantly.

Stores that use complete product schema get 58.3% more clicks from AI shopping interfaces than stores with incomplete or missing schema. Products with clear schema markup appear 3.1 times more often in AI Overviews and AI shopping results.

The difference between Product schema and Offer schema

Product schema describes what the product is. Offer schema covers the price, availability, and seller information. Together they form the complete picture AI shopping systems need.

Product schema includes product name, description, image, brand, product category, color, size, weight, dimensions, material, and GTIN (the barcode number). This is the identity layer. It tells the AI what you are selling.

Offer schema includes price, currency, availability (in stock, out of stock, preorder), item condition (new, refurbished, used), seller name, and return policy. This is the transaction layer. It tells the AI where and how customers can buy it.

A product page needs both. A product without an offer is just information. An offer without a product leaves the AI confused about what is being sold. Schema markup for AI shopping always includes Product schema plus at least one Offer schema block.

How AI shopping agents use schema to match products to customer needs

When a customer asks an AI agent to find something, the agent breaks the request into searchable attributes. A request for "a water bottle that keeps drinks cold for 24 hours" gets translated into search attributes like product type equals water bottle, material includes stainless steel or double-wall, capacity is at least 16 ounces, and color can be anything.

The agent then searches product feeds for products that match all of those attributes. If your product schema does not include the material attribute or the capacity attribute, the agent cannot evaluate whether your product matches. Even if your product is perfect, the agent skips it.

This is why complete schema matters more than marketing copy. Your beautiful product description that says "keeps drinks cold all day" does not give the agent a temperature or duration. Your schema that specifies the material as stainless steel and capacity as 32 ounces tells the agent exactly what to search for.

What AI shopping systems actually read in your schema

Not every field in schema markup gets equal weight. AI shopping systems prioritize certain fields because those fields help them match products to customer constraints.

Critical fields that are required include product name, price, currency, and availability. These are non-negotiable. Without them, the agent cannot include your product in results.

High-priority fields that get used heavily include brand, image, GTIN or MPN, color, size, material, weight, dimensions, and AggregateRating. These fields directly influence whether an agent recommends your product when a customer is comparing options.

Secondary fields that appear when relevant include warranty information, shipping details, return policy, product category, and manufacturer. These appear in the agent's response but do not drive the initial recommendation.

Most ecommerce brands get the critical and high-priority fields right but skip secondary fields entirely. A complete schema that includes warranty, return policy, and shipping details increases the likelihood an agent recommends you when customers are evaluating trustworthiness.

JSON-LD is the format AI systems prefer

There are multiple ways to write schema markup. Microdata uses HTML attributes. RDFa uses HTML tags. JSON-LD uses a script tag that lives separate from your HTML.

AI systems prefer JSON-LD because it is the easiest format to parse. JSON is valid, structured data. A computer program can extract it without understanding the layout or design of your page. When an AI crawler visits your page, it looks for a JSON-LD script tag, reads the data, and uses it immediately.

89% of schema implementations use JSON-LD. It is the format Google recommends. It is what AI shopping platforms expect to read. If your ecommerce platform generates schema automatically, it is almost certainly generating JSON-LD.

Schema markup should sit in the page <head> or immediately after the <body> tag. It should be static HTML in the page source, not loaded dynamically with JavaScript. If your product schema loads after JavaScript executes, some AI crawlers will not wait for it to load and will crawl a blank version of your page.

Building complete Product schema for your catalog

A minimal product schema block needs to include name, description, image, brand, price, currency, and availability. But minimal does not compete well with complete.

A complete product schema includes everything above plus GTIN (barcode number), color options, size options, material, weight, dimensions, product category, manufacturer, aggregate rating (star count and review count), and any attributes specific to your product category.

If your product comes in variants (different sizes or colors), each variant needs its own Offer block with its own price and availability. A shirt that comes in red, blue, and green should have three separate Offer schema blocks. A bag in small, medium, and large should have three price and availability entries.

Major ecommerce platforms generate basic schema automatically. The schema generation is usually good enough for critical fields, but it often misses material, dimensions, and variant handling. Audit your schema. Check that color and size variants have their own Offer blocks. Verify that special product types (bundles, gift sets) are structured correctly.

How to verify your schema is correct

Do not trust that your platform is generating correct schema without verifying it. Use Google's Rich Results Test or the Schema.org validation tool. Enter your product page URL, and the tool shows you exactly what schema your page contains.

Look for these specific issues. Missing GTIN numbers, incorrect availability values, missing currency declarations, missing image URLs, review count set to zero, and missing variant information are the most common problems to fix.

Missing GTIN numbers are the most common problem. If your product has a barcode (EAN, UPC, ISBN), include it in your schema. AI systems use GTIN to verify product identity across retailers. If your GTIN is missing or incorrect, the agent cannot match your product to inventory systems and shopping comparisons.

Review counts set to zero or blank are a red flag. If your schema says a product has zero reviews, AI systems assume it is new and unverified. Even if your product has 500 reviews, if your schema does not include that review count, agents see an unproven product.

Product feeds as your core schema asset

If you sell on marketplaces or use shopping feed services, your product feed is where schema comes to life at scale.

A product feed is a structured list of all your products with standardized fields. Each field maps to schema markup. Your feed generator creates JSON-LD schema from that data. If your feed has incomplete or incorrect data, the schema will be incomplete or incorrect at scale.

Invest in feed quality. This means auditing your product data for accuracy, completeness, and consistency. A high-quality feed that includes GTIN, brand, price, availability, and product attributes will generate clean, complete schema automatically.

Most ecommerce companies leave 10 to 20% of their product data incomplete. An incomplete product does not appear in AI shopping results. A feed audit that fixes just the top 20% of your products by sales volume can increase AI visibility by 40% or more.

AggregateRating and Review schema boost recommendations

A product with reviews gets recommended more often than an identical product without reviews. AI shopping systems treat reviews as third-party verification of your claims.

AggregateRating schema tells AI systems how many people reviewed your product and what their average rating was. Include the ratingValue (the average star rating, like 4.7), reviewCount (how many reviews you have), bestRating (always 5), and worstRating (always 1).

A product with 4.8 stars and 120 reviews signals trustworthiness to an AI agent. A product with no review data signals that it is unverified. If your product has reviews on your site, include the aggregate rating in schema. If your product is new and has no reviews, at least include ratingCount as 0 instead of omitting the field entirely.

How WEMASY helps with product schema

WEMASY's website builder includes schema generation for product pages. You add product information through the WEMASY interface, and the system automatically generates clean, valid JSON-LD schema. You can preview your schema before publishing using WEMASY's built-in validator.

For merchants selling on WEMASY, you can add detailed product attributes like material, dimensions, color variants, and weight directly in your product editor. WEMASY translates those fields into complete Product and Offer schema automatically. Learn more about WEMASY's ecommerce features at our pricing page.

Frequently asked questions