Constructor Blog | Ecommerce Search Industry and Product Information

Agentic Commerce Optimization: AI-Readable Catalogs | Constructor

Written by Noelina Rissman | Nov 20, 2026, 9:06:24 PM

Imagine this: A shopper asks an AI agent for a carry-on that fits under an airline seat, weighs less than 3 kg, and arrives by Friday. To make a useful recommendation, the agent doesn't browse your site the way a person does. It looks for facts it can verify (e.g., dimensions, weight, price, stock, delivery options, etc.) and, more importantly, needs a reliable way to access that information.

If those details are missing or inconsistent, the agent may not be able to determine whether a product meets the request. It might ask for more information, qualify its answer, or recommend another product. Its response will depend on the agent and the information it can access.

This is the work behind agentic commerce optimization: making product information accurate, up to date, and accessible so AI agents can help shoppers discover and evaluate products — and, where supported, take the next step toward purchase.

Below, we show what an agent-ready catalog needs, how different autonomous AI agents and other AI commerce experiences access it, and how to prioritize improvements.

What Makes a Product Catalog Readable for AI Agents?

Short answer: A product catalog is readable to AI agents when there’s enough reliable information to identify a product, understand its attributes, and assess whether the offer meets a shopper’s needs. That information also needs to be available through a path the agent can use, such as a product page, feed, API, or supported integration.

The details vary by product and destination, but the foundation includes:

  • Stable product and variant identifiers
  • Accurate titles, descriptions, images, and product URLs
  • Category-specific attributes, such as dimensions, materials, or compatibility
  • Current price and availability
  • Relevant shipping, return, and seller information

It’s worth noting that even with a solid foundation, good product data doesn’t guarantee a recommendation. An external AI-powered platform decides how it retrieves and presents information. Incomplete or inaccessible data can make it harder for any agent to evaluate a product accurately.

Product facts vs agent interpretation

An agent uses product facts to interpret a shopper’s request.

For example, "waterproof," "30 cm wide," "EU 42," "100% cotton," and "in stock" are catalog facts. "Good for a rainy commute" or "great for a small bedroom" are conclusions an agent can draw from those facts and the context a shopper provides at runtime.

That difference tells you where to invest. Your job is to make the facts complete, accurate, and current. You don't need to pre-write every possible use case into your product copy.

The Three Layers of Agent-Ready Product Data

Organizing product information into three layers can help teams identify gaps and determine who needs to address them.

Layer 1: Product identity and offer information

This layer answers one question: What exactly is this item, and can I buy it right now? Agents may need to verify this information across multiple sources, particularly when checking prices and availability. That includes:

  • A stable product identifier for every product and purchase variant
  • A canonical product URL, ideally with the correct variant preselected
  • Brand and seller information
  • Price, currency, and real-time availability
  • Clear parent-variant relationships, with no identifier reused across different items
  • Global Trade Item Numbers (GTINs) or manufacturer part numbers, where assigned

Variant data deserves extra attention. If a jacket comes in four colors and five sizes, an agent needs to know which of those 20 combinations is in stock, at what price, and at which URL.

For example, OpenAI's product feed specification calls for a separate row per purchasable item or variant. Its basic discovery fields include item ID, title, description, URL, brand, seller name, image URL, availability, and price. These are OpenAI’s requirements for that feed, not a universal formula for every AI platform. ⁠

Layer 2: Attributes that answer shopper questions

This layer answers a different question: Does this particular product meet the shopper's needs?

The attributes depend on the category. A luggage shopper may care about dimensions and weight, while an apparel shopper may need material, size, and fit. And then a home improvement shopper may need information on compatibility or installation.

No matter the attributes, make them as clear and consistent as possible by following these best practices:

  • Use explicit units for measurements
  • Standardize values and field names where practical
  • Record the source of an attribute so your team can review or correct it
  • Avoid filling gaps with unsupported assumptions

Coverage matters, too. If compatibility is populated for only a third of a category, an agent may struggle to answer compatibility questions for the remaining products — even if the field exists in your catalog.

Where Attribute Enrichment helps

Retailers often receive product data from many suppliers, with different levels of detail and inconsistent formats. Constructor’s Attribute Enrichment helps address those gaps by using catalog inputs such as titles, descriptions, categories, and images to predict product attributes and categories. Shopper behavior can also help identify which attributes matter most to customers. ⁠

Retailers can review enriched attributes, use them in Constructor’s search and browse experiences, and export them for use in their product information management system, or PIM. Whether those values reach an external feed or agent depends on how the retailer exports and integrates them. ⁠

Layer 3: Policies and information needed to take action

This layer answers the questions: What should the shopper know about the offer, and what can happen next?

Depending on the experience, useful information may include:

  • Shipping options and costs
  • Return policies and windows
  • Customer ratings, reviews, and other forms of social proof
  • Seller information
  • Product restrictions or safety notices
  • Eligibility for supported cart or checkout features

Not every destination accepts or uses the same fields. Check its documentation before assuming that an agent can access a particular policy, review, or checkout option.

Layer What it helps an agent determine Examples
Product identity and offer Which item or variant is being offered Item ID, URL, price, availability
Product attributes Whether it meets the shopper’s needs Dimensions, material, size, technical specifications, compatibility
Policies and action information What applies after selection Shipping, returns, seller of record, eligibility

How AI Agents Participate in the Buying Journey

Not all agent-led shopping experiences function the same way. Some help shoppers discover products directly on a retailer's website, while others operate independently or connect to retailers through specialized integrations.

Onsite AI shopping agents

These assistants help shoppers find, compare, and evaluate products without leaving the retailer's website. For example, an AI Shopping Agent supports a shopper’s broad discovery goals, and a Product Insights Agent answers last-mile, product-specific questions directly on the product detail page.

When powered by the same discovery system as the retailer's search, browse, and recommendations, these agents can draw on shoppers' past behavior and current session activity to personalize the experience.

Independent AI shopping assistants

Shoppers may ask ChatGPT or other AI assistants for recommendations before visiting a retailer's website. These assistants can help shoppers narrow their options, compare products, and identify retailers, but they don't necessarily have access to retailers' own merchandising or personalization intelligence.

Retailer-connected AI apps

Retailers can also bring product discovery directly into personal AI assistants like ChatGPT or Muse, allowing shoppers to search, browse, and even purchase products without leaving the conversation. This is enabled through retailer-specific apps that shoppers can install or connect to their AI assistants.

When driven by a central reasoning engine, these apps can draw on the retailer's accumulated shopper behavioral data to personalize product recommendations and optimize rankings.

Retailer-enabled agent access

Retailers can also make their websites more accessible to external AI agents that browse the web on shoppers' behalf. Instead of relying solely on information gathered from webpages, these agents can connect directly to the retailer's commerce systems to search and browse products, retrieve up-to-date information, and access supported shopping capabilities.

This enables agents to interact with the retailer's catalog more efficiently, without having to navigate the website as a human shopper would.

Where AI Agents Get Product Information for Decision Making

Each of these AI shopping experiences depends on having access to accurate, reliable product information. But how that information reaches an AI agent varies depending on the platform, the experience, and the integrations available.

Some AI assistants draw on product information already available through search indexes and product feeds, or retrieve additional details from websites in real time. Others can connect directly to retailers' commerce and discovery systems to access product information and capabilities.

Let's look at the different ways retailers can make their product information available to AI agents.

Google: Product pages and Merchant Center

For shoppers discovering products through Google Search, AI Overviews, or AI Mode, Google confirms that existing SEO best practices still apply. Pages need to be indexed and eligible to show with a snippet, and you don't need special AI files or schema markup to appear in these features.

For product information, Google recommends adding structured data to product pages and sharing catalog data through Merchant Center. These can help Google understand product details, though requirements vary by Google surface.

ChatGPT: Product feeds and the Agentic Commerce Protocol (ACP)

For shoppers using ChatGPT to discover and compare products, OpenAI can draw on structured product feeds and other available sources.

OpenAI’s Agentic Commerce Protocol (ACP) provides a framework for sharing structured product data with ChatGPT. Retailers can start an ACP integration by submitting a product feed with details such as product identifiers, titles, descriptions, images, price, and availability. Additional fields can describe variants, shipping, returns, and other product information. ⁠

It’s important to note that a feed helps ChatGPT discover and recommend products, which shoppers can then investigate or purchase through the retailer. It doesn’t, on its own, enable building a cart or checking out directly within ChatGPT. Those actions require separate capabilities and integrations. ⁠

Model Context Protocol (MCP): Connecting agents directly to product discovery

Model Context Protocol (MCP) is an open standard that lets AI applications connect to external tools and information. It's one way to power the retailer-connected AI apps described earlier, allowing shoppers to search and browse a retailer's ecommerce site directly within a supported AI assistant.

With Constructor’s Discovery MCP, a connected AI experience can call Constructor’s product discovery capabilities to search and browse a retailer’s catalog. In other words, an agent can retrieve results informed by Constructor’s behavioral learning intelligence and the retailer’s configured merchandising rules, rather than relying only on the agent’s own interpretation of product pages, creating an experience that mimics Constructor-powered product results within ChatGPT.

Universal Commerce Protocol (UCP): Helping agents connect to commerce abilities

The Universal Commerce Protocol (UCP) is a standard that helps AI agents discover and communicate with retailers' commerce systems. It supports the retailer-enabled agent access experience described earlier, where an external agent can interact with a retailer's systems without the shopper first installing a retailer-specific app.

Through UCP, retailers can make supported capabilities discoverable to compatible AI agents. For example, Constructor's UCP integration can direct an agent to its Discovery MCP, allowing the agent to search and browse the retailer's catalog using Constructor's product discovery intelligence and merchandising rules.

Depending on the implementation, UCP can also support experiences beyond search and browse, such as guided product-finder quizzes, promotional product bundling, cart building, and checkout, through direct integrations with retailers' existing payment systems.

For Google’s UCP implementation, retailers need to review product eligibility, identifiers, policies, and other requirements. Google strongly recommends complete variant information for products with options such as size or color. Again, requirements and eligibility vary by implementation.

For early adopters of MCP and UCP, these integrations offer new opportunities to make product discovery capabilities available to AI agents “offsite.”

If you’re interested in seeing how Constructor can help you control how your products are found, ranked, and surfaced across emerging AI channels via UCP experiences, discover more here.

Improving visibility is different from controlling the answer

Providing accurate information gives an external platform better material to work with, but it doesn’t give a retailer control over how that platform ranks or presents products.

That’s where Constructor’s Commerce AEO fits into the broader discovery strategy for independent AI shopping assistants. It’s designed to help retailers understand their visibility in AI answer engines, identify opportunities to improve content, and connect AI-referred traffic to ecommerce outcomes. It’s a complement to product data work, not a guarantee that an external assistant will recommend a particular item. ⁠

How Product Data Affects an Agent-Assisted Shopping Journey

Consider the carry-on request again: the shopper wants a bag that fits under an airline seat, weighs less than three kilograms, and arrives by Friday.

An agent may need to:

  • Understand the request. Identify the shopper’s constraints and, if necessary, ask which airline or destination they are referring to
  • Find candidate products. Retrieve bags using available catalog information
  • Compare the details. Check dimensions and weight against the shopper’s requirements, and consult an available delivery estimate for the shopper’s location
  • Recommend or take the next step. Present suitable options and, where the agent and retailer support it, help the shopper move to a cart or complete checkout

Now imagine two possible products. Bag A has its dimensions and weight recorded in accessible fields. Bag B has a description that says “perfect for weekend trips,” but no usable measurements.

Bag A gives the agent more information to work with. However, before it can confidently recommend the product, the agent still needs the relevant airline’s size limits and a supported delivery estimate to confirm the full request. For Bag B, it may not be able to assess fit or weight at all.

The fix is more useful than adding more descriptive phrases. The retailer can identify missing luggage measurements, validate them against reliable sources, and make the values available in the systems and feeds that need them. Then the team can test whether the change improves answers to representative shopper questions.

A Practical Sequence for Improving AI Search and Agent Readiness

You don't need to fix your entire catalog at once. The following six-step sequence moves you from audit to measurable progress.

  1. Map where product information lives. List the systems that hold product details, pricing, inventory, fulfillment, policies, and reviews. Assign an owner and expected update cadence to each source
  2. Measure the current state. For the selected category, check the validity of identifiers, attribute coverage, variant consistency, and freshness. Review a sample of populated values for accuracy, not just whether a field is filled in
  3. Compare what each experience receives. Check the same product and variant across the product page, search index, feed, inventory system, and other relevant destinations. Investigate mismatches, particularly around price and availability
  4. Prioritize gaps shoppers encounter. Look at search queries, filters, product questions, and customer support topics. Prioritize information that shoppers use for decision-making, based on its business value and the availability of reliable source material
  5. Test with representative requests. Run representative shopper requests against a defined category or sample. Compare results before and after the change, or use a comparable control where possible. Check both whether relevant products are retrieved and whether answers accurately reflect the available information
  6. Maintain information at the right cadence. Different data changes at different rates. Price and availability may need frequent updates; material or dimensions may change less often. Set refresh and error-handling processes accordingly, and review each destination’s current requirements

Commerce AEO can streamline this work by checking your site’s AI readiness and recommending actions to address gaps, such as missing product attributes or information on product pages. Constructor’s integrations also help keep product data current, while technical foundation checks help identify barriers to AI access.

How to Measure Whether Catalog Improvements Work

Agent readiness can be easy to claim and hard to prove. Use a combination of data-quality, discovery, and business measures, as no single metric tells the whole story.

Start with data quality:

  • Attribute coverage and accuracy
  • Consistency across product pages, feeds, and indexes
  • Price and availability freshness
  • Correct variant identification

Then assess discovery and assistance:

  • Whether relevant products are retrieved for representative queries
  • Whether answers are supported by the available product information
  • Whether agents handle missing information appropriately
  • Whether shoppers can successfully refine their requests

Finally, look at business outcomes where measurement allows:

  • Add-to-cart and checkout completion
  • Conversion and revenue
  • Returns or support contacts related to product uncertainty

Commerce AEO’s Content AI Readiness and Technical Foundation scores can help you assess your starting point, while share of voice and citations help you track changes in AI visibility. You can also measure whether AI-referred traffic leads to add-to-carts, purchases, and revenue.

Where possible, compare an updated set of products with a similar control group, and verify that your evaluation can detect the enriched attributes. ⁠

Connecting Better Catalog Data for Better Agentic Experiences, On and Offsite

Accurate, accessible product information helps AI agents evaluate products and creates a smoother path from discovery to purchase.

Onsite, AI shopping agents can combine product information, shopper behavior, and other signals to help customers explore and narrow down options. As shoppers approach a purchase decision, product insights agents can answer last-mile questions using the information available to them, helping shoppers evaluate whether an item meets their needs.

Offsite, making product information accessible to AI platforms can help retailers improve their ecommerce business’ visibility in answer engines. The next step is to measure not just whether products appear, but whether AI-referred shoppers engage and complete purchases.

Start with a category where shoppers need specific information, identify the gaps, and test whether closing them improves discovery. And if you’d like more concrete help identifying where to start, get in touch.