Agentic commerce: preparing product data for AI shopping agents

Agentic commerce is a model of online shopping in which AI agents search, compare and, where permitted, complete purchases on behalf of a buyer. Agents read product data through feeds and APIs, so the completeness, structure and freshness of that data decide whether a product is found, recommended and bought. In 2026 several open protocols define how agents, merchants and payment providers interact, and all of them assume machine-readable product data.

Which agentic commerce protocols exist in 2026

The protocols address different layers. Product data teams mainly deal with the catalogue layer, while checkout and payment layers sit with commerce and payment teams.

UCP lists support for AP2, the Agent2Agent protocol (A2A) and MCP. Card networks run related programmes for agent payments, such as Visa Intelligent Commerce and Mastercard Agent Pay.

How agentic checkout works in practice

Checkout models are still settling. Google offers UCP-based checkout on eligible listings in AI Mode in Google Search and in Gemini, starting in the United States, with Australia and Canada announced next. Merchants must complete a technical UCP integration and apply through Merchant Center.

OpenAI launched Instant Checkout in ChatGPT on ACP in September 2025. In March 2026, it was widely reported that OpenAI would wind down Instant Checkout and shift towards merchant-built apps inside ChatGPT, with many purchases completed on merchant checkouts. Product discovery in ChatGPT continues to rely on merchant product feeds.

For product data teams, the practical conclusion is stable even while checkout models change: agents need accurate catalogue data, whichever protocol completes the order.

What product data AI agents need

Agents compare offers on facts. The OpenAI product feed specification illustrates the baseline: nine required fields (item ID, title, description, URL, brand, seller name, image URL, availability and price), conditional fields for variants and checkout eligibility, and recommended fields such as GTIN, MPN, material, colour, size, dimensions, weight, shipping and return policy.

Beyond any single specification, agents perform better with:

• Granular, typed attributes that answer comparison questions, such as capacity, compatibility, materials and certifications.
• Global identifiers (GTIN, MPN, brand) that let agents match one product across retailers.
• Variant structures with clear parent and child relationships.
• Real-time price and availability, since stale offers lead to failed transactions or exclusion.
• Explicit policies on shipping costs, delivery times and returns, stated as data.
• Use-case content that matches conversational requests, covered in generative engine optimisation.

Where PIM fits in the agentic commerce stack

Agentic commerce adds channels to an existing syndication model. The PIM holds master attributes and content, feed management tools such as Productsup, Channable or Feedonomics (see vendor profiles) transform them per channel, and commerce platforms expose price, stock and checkout through APIs. A headless architecture with well-documented product APIs makes it easier to serve agents that query data directly.

Governance requirements rise with automation. When agents act without a person reading the page, an incorrect dimension or compatibility claim can lead directly to a wrong purchase and a return, so validation rules and data quality monitoring need to cover agent channels like any other channel.

Content rights also need attention. Image licences, embargoed launch content and retailer-specific prices should be released to agent channels under the same rules applied to other syndication targets, using the rights metadata described in digital asset rights management. Each agent channel should appear as a distinct target in the PIM, with its own completeness rules and publication status, so teams can see which products are ready for which agent surface.

Practical next steps

1. Map which AI shopping surfaces matter in target markets and which feeds or protocols they use.
2. Compare the PIM data model with the OpenAI feed specification and Google Merchant Center requirements, and close attribute gaps.
3. Ensure GTINs, brand names and variant structures are consistent across all channels.
4. Move price and availability updates to near real time for agent channels.
5. Publish shipping and return policies as structured data.
6. Involve commerce and payment teams in UCP or ACP checkout decisions, since those integrations sit outside the PIM.