Agentic commerce: how to prepare your catalog for buyers who shop through AI

Agentic commerce is buying through AI agents that search, compare and, increasingly, complete purchases on behalf of a person or a company. To be chosen by those agents, a catalog needs complete product attributes, structured data, machine-readable policies, and prices and stock that are always current. Agents recommend what they can read and verify, and skip what they cannot.

What changes when the buyer is an AI agent?

A human shopper tolerates gaps. They zoom into photos, read reviews, open the returns page and message the store with questions. An agent works from the data it can retrieve in seconds and from the sources it considers reliable.

That shifts the competitive ground. A product with a vague title, missing dimensions and a return policy published as an image may be invisible to an agent even if it is the best offer. Completeness and consistency become as important as price.

The infrastructure is moving quickly. In 2025, OpenAI and Stripe introduced the Agentic Commerce Protocol and Google introduced the Agent Payments Protocol, both designed to let agents complete purchases on a buyer's behalf, and search engines already show AI-generated answers above traditional results. Adoption will vary by market and category, but the preparation work is the same everywhere, and it also improves regular search.

Which product data do AI agents need?

Start with the fields a careful buyer would check before purchasing, and make every one of them explicit in text.

• Clear titles that state product type, brand, model and the key variant, without promotional filler.
• Complete attributes: dimensions, weight, materials, compatibility, capacity, certifications and anything a buyer filters by.
• Variants modeled as variants, each with its own identifier, price and availability.
• Standard identifiers such as GTIN, MPN or SKU, consistent across the website, marketplaces and feeds.
• Descriptions written as answers to real questions: what the product is for, who it suits and what it does not do.

B2B catalogs need the same discipline, plus the fields that procurement teams and their agents check: minimum order quantities, lead times, technical documentation and the commercial terms that apply to each customer segment.

How should structured data be implemented?

Structured data turns a product page into facts an agent can parse. The schema.org vocabulary used by search engines covers what matters: Product for the item, Offer for price and availability, OfferShippingDetails for delivery and MerchantReturnPolicy for returns. Put it in the HTML the server delivers, because many crawlers do not execute JavaScript.

Keep structured data and visible content identical. If the page shows one price and the markup says another, both people and machines lose trust. Validate the markup on every catalog release, not once a year.

Why do price and stock freshness matter more now?

An agent that recommends an out-of-stock item or a stale price creates a bad experience for its user, and the platforms behind these agents have every reason to prefer reliable sources. Inventory and pricing need to flow from the ERP to the storefront, feeds and marketplaces in near real time.

This is integration work. Catalog data usually lives across the ERP, a commerce platform, marketplace connectors, logistics systems and spreadsheets maintained by the commercial team. Agents that monitor these sources, flag divergences and correct feeds are among the most practical first AI use cases for commerce teams.

How should policies be written for machines?

Shipping times, return windows, warranty terms and payment conditions often live in PDFs, banners or images. Rewrite them as plain text with explicit values, and mirror the key facts in structured data. Ambiguity that a human resolves by asking support becomes a reason for an agent to choose another store.

• State delivery times by region and shipping method, including order cut-off times.
• Publish return windows, conditions and costs as text, not as an image.
• Describe warranty coverage and how to make a claim.
• Keep a public, crawlable FAQ for pre-purchase questions.

How do you measure discovery when there is no click?

As AI answers resolve more questions directly on the results page or inside an assistant, traffic stops being a complete proxy for visibility. Marketing and commerce teams need indicators that capture presence inside AI answers, not just visits.

1. Track how often your brand and products appear in AI answers for a fixed set of buying questions, tested on a regular schedule.
2. Compare impressions and clicks in search consoles to find queries where answers are absorbing clicks.
3. Follow branded search and direct traffic as signals of influence that happened elsewhere.
4. Report referral traffic and conversions from AI assistants as a separate channel.
5. Audit the accuracy of what assistants say about your products, and correct the source data when it is wrong.

No single metric replaces the click. A small dashboard that combines these signals gives a more honest picture of whether the catalog is being found and chosen.

What does a catalog readiness audit check?

Before investing in new tools, run a short audit on the products that generate most of the revenue. It shows where the gaps are and which fixes will matter first.

1. Attribute coverage: the share of required attributes filled for each product category.
2. Identifier consistency: whether GTIN, MPN and SKU match across the ERP, the storefront and marketplace feeds.
3. Price and stock latency: how long a change in the ERP takes to reach every channel.
4. Structured data validity: whether markup is present, valid and identical to visible content.
5. Policy clarity: whether shipping, returns and warranty terms exist as text with explicit values.
6. Answer accuracy: what the main AI assistants say about your top products today.

Score each item for the top products, fix the worst gaps first and repeat the audit every quarter. Catalogs drift just like models do, as new products, suppliers and promotions arrive.

Who should own catalog readiness?

Catalog quality sits between commerce, marketing, operations and IT, which is why it often belongs to nobody. Treat it as an operation with an owner, service levels for data freshness and a backlog of attribute and policy improvements reviewed every month.

AI can carry much of the repetitive work: enriching attributes from supplier documents, detecting inconsistencies between ERP and storefront, generating structured data and monitoring how assistants describe your products. At AI Infinity, this work runs inside the client's existing commerce and ERP stack, with people approving any change that affects prices or customer promises.

Whatever the operating model, the principle holds. The companies that agents trust are the ones whose product data is complete, current and consistent everywhere it appears.