What do AI agents actually see when they recommend your products?

What do AI agents actually see when they recommend your products?

Can AI agents like ChatGPT and Gemini actually “see” your products?

It’s a question more ecommerce leaders are asking as Unified Commerce Protocol (UCP) gains traction. The goal: a universal language so AI can instantly understand your catalog and boost sales. But how do these systems actually process your catalog?

The three layers of product data AI agents use

AI agents process product data from three main sources, each with specific limitations:

• 1. Storefront/Web Scraping: Most large language model (LLM) crawlers use US IPs and extract information from your public storefront. This means they typically access your US Market version and pricing, descriptions, and product availability may not match what international shoppers see.

• 2. Web Search Layer: Platforms like ChatGPT and Gemini frequently use web search to enrich their responses with real-time product information. These search results are influenced by the user's browser location, which means pricing and availability can vary per user. While this improves local relevance, it creates inconsistencies, especially for stores using advanced localization (such as Shopify Markets).

• 3. API/Product Feed Layer: The most structured source, Shopify Catalog or OpenAI Product Feed, offers detailed product data. However, localization support is limited, and most feeds default to US data. Global pricing, duties, and inventory are often missing or incorrect.

Risks of data mismatch

Without fully localized, accurate data, AI agents can present incorrect prices, availability, or details to customers. For global brands, this creates risks of misinformation at scale.

How Shopify Catalog handles product data (and how to influence it)

Shopify published a technical post (link) explaining how product attributes are processed using machine learning models. Merchant data is often messy, so Shopify standardizes attributes and descriptions to a unified format and length.

How can you influence what AI agents learn about your products?

• Write concise, well-structured product descriptions that fit the AI-preferred format. Shorter, clearer descriptions are more likely to be interpreted correctly.

• Fix product groupings using Shopify’s catalog mapping tool (link). Custom variant grouping is a new feature and worth testing, even if it’s still early days. Note: replace YOUR_BRAND_SHOP in the url with your own Shopify store ID to access directly.

• With Glara, these optimizations can be automated, ensuring your catalog is structured for AI discovery without manual effort.

How to audit what an AI Agent "knows" about your products

• Shopify Catalog: Use the Shopify dev dashboard (link) to query your product catalog. This provides a “JSON snapshot” of what Shopify’s AI interprets for your products.

  Open the link → “Catalogs” → “Create a catalog” → write any Query

• Probing LLMs: Ask ChatGPT or Gemini about your products from different locations to reveal gaps and inconsistencies.

What’s next? (and how Glara can help)

UCP and AI-driven product discovery are evolving rapidly. Better support for global markets is on the roadmap (UCP priorities). Until then, product data quality and structure are critical.

Glara helps ecommerce brands optimize product data for AI discovery, ensuring your catalog is accurately represented across all channels.

Want to benchmark your catalog? Reach out for an audit or learn more at glara.ai.

References: Shopify technical post on ML product attributes (link), OpenAI Product Feed spec (link).