Agentic commerce

The three stages, and where each stands

Stage one — research — is current behavior: a shopper's first touch with a category is increasingly an AI answer rather than a results page. Stage two — draft carts — is live and expanding: ChatGPT Shopping and Perplexity's Buy with Pro assemble options and carts, then hand the buyer back for approval. Stage three — agent-completed checkout — is protocol-mediated and in pilot as of mid-2026: ACP from OpenAI/Stripe (backing ChatGPT Instant Checkout), Google's UCP, with AP2 on the payments layer and Shopify Agentic Storefronts wrapping it for merchants.

Agentic commerce names the phenomenon. ACO — Agentic Commerce Optimization — names the practice of preparing a catalog to perform inside it. Keeping the two distinct keeps vendor conversations honest.

Why it matters for ecommerce

For a VP of Ecommerce this is a planning-horizon question. Agent-driven volume is modest today across D2C; its trajectory is not. The catalogs that lose when volume arrives are the ones that are messy now — and the fixes (clean schema, machine-readable price and availability, admitted crawlers, discoverable policies) are the same fixes that lift AI-research citations this quarter. The work compounds; nothing is stranded if protocol timelines slip.

For a VP of SEO it changes the yardstick: a SKU that wins an agent's evaluation does not need a click to convert. The optimization target shifts from visit to selection, which is why structured attributes start outweighing persuasion copy at the margin.

An agent evaluation: an example

A ceramic-mug brand's 12oz hand-thrown mug is one of seven candidates when a ChatGPT user asks for "a gift mug under $50 that is dishwasher safe." The agent evaluates against explicit constraints: three candidates eliminated on price, two on missing dishwasher information, leaving two finalists. The final pick turns on schema clarity and review data (illustrative example). The brand whose dishwasher-safe status lives in a typed metafield surfaced via Product JSON-LD survives the elimination round; the brand whose status lives in a lifestyle paragraph may not — even if both products are identical.

The strategic claim worth internalizing

The PDP work that wins citations in stage one is the same work that makes a SKU draftable in stage two and purchasable in stage three. Brands do not have to bet on which checkout protocol wins; the groundwork is shared. That is the reasoning behind the two-score model: the citation score measures today's outcome, the agent-readability score measures readiness for the next two stages. OpenAI's commerce announcements and Google's UCP materials are the primary sources to track as the pilots expand.

How eCommerce Insights fits in

The platform scores readiness rather than implementing checkout: per-SKU agent-readability checks, protocol-aware recommendations, and the PDP groundwork shipped as reviewable diffs. The free Agentic Readiness Grader runs the same checks on any PDP.

Related terms

• ACO (Agentic Commerce Optimization) — the practice of preparing for the phenomenon.
• ACP (Agentic Commerce Protocol) — the OpenAI/Stripe checkout protocol.
• UCP (Universal Commerce Protocol) — Google's counterpart protocol.
• Draft cart — the live middle stage of agent involvement.
• Agent checkout — the piloting final stage.