Chapter 12 9 min read

Optimizing Product Data for Agents

If an agent can't read your product, it doesn't exist.

In the previous chapter, the agent selected the "Bellroy Classic" because it knew it had a laptop sleeve. How did it know that? Because the data was structured in a way the agent could understand.

Optimizing for agents is different from optimizing for humans (UX) or search engines (SEO). We call this new discipline AIO (Artificial Intelligence Optimization) or GEO (Generative Engine Optimization).

The Three Layers of Data Maturity

Level 1: Unstructured (The Swamp)

Product descriptions are blobs of marketing copy.
"This amazing bag is great for work and play! It fits lots of stuff."

Agent Reaction: Confusion. "Does it fit a 16-inch laptop? Is it waterproof? I don't know." -> Skip.

Level 2: Structured (The Spreadsheet)

Standard attributes are filled out.
Color: Blue, Material: Nylon, Dimensions: 12x18x6.

Agent Reaction: Comprehension. "I can filter by size and material." -> Consider.

Level 3: Semantic (The Knowledge Graph)

Data is enriched with Context and Use Cases.
Use_Case: ["Commuting", "Light Travel"]
Compatibility: ["MacBook Pro 16", "iPad Air"]
Climate_Suitability: ["Rain", "Moderate Cold"]

Agent Reaction: Preference. "This perfectly matches the user's intent for a 'rainy commute'." -> Recommend.

How to Optimize for Level 3

Merchants need to stop writing for humans and start writing for the machine.

1. Flatten the Hierarchy

Humans love clicking categories: Men > Accessories > Bags > Backpacks.
Agents hate traversing trees. They want flat, searchable tags.

Instead of burying "Waterproof" in a description tab, make it a top-level boolean attribute: is_waterproof: true.

2. Answer the "Negative" Questions

Humans rarely ask what a product can't do. Agents optimize for constraints.

If your data explicitly says is_carry_on_compliant: false, you save the agent a calculation. It will trust you more than a brand that leaves it ambiguous.

3. Schema.org on Steroids

You are likely already using Schema.org for Google. For agents, you need to go deeper. Use the GS1 web vocabulary or industry-specific ontologies.

{ "@context": "https://schema.org/", "@type": "Product", "name": "Classic Backpack", "material": { "@type": "Material", "name": "Cordura Nylon", "sustainabilityLevel": "Recycled" } }

The "Context Window" Economy

Agents have a limited "Context Window" (how much text they can read at once).

If your product page sends 5MB of HTML, JavaScript, and tracking scripts, the agent might truncate it.

The Rule: The highest value information (Specs, Price, Stock) must be the most accessible.

Many brands are creating a dedicated /agent.json endpoint for every product page.

  • Human URL: store.com/products/backpack (Heavy, Visual)
  • Agent URL: store.com/products/backpack/agent.json (Light, Data-only)

Key Takeaways

  • Merchants must move from Unstructured text to Semantic data to be visible to agents.
  • Agents prioritize products with explicit constraints (e.g., 'is_waterproof: true').
  • Providing a dedicated lightweight data endpoint (like /agent.json) ensures your product fits in the context window.
  • This is the new SEO: Optimizing for the machine's understanding, not just keyword matching.
Next: Security & Governance