Draft cart

How a cart gets drafted

The buyer states an intent — "I need a base layer, a beanie, and liner gloves for spring skiing, around $150 total" — and the agent decomposes it, evaluates candidates per slot against the stated constraints, and assembles a cart for approval. The buyer edits or accepts; checkout completes either back on the merchant's storefront or, in pilots, through agent checkout protocols. ChatGPT Shopping and Perplexity's Buy with Pro run this pattern in production as of mid-2026; see OpenAI's shopping announcements for the first-party description.

The selection step is the part that matters for brands: each cart slot is an elimination tournament over structured data. Candidates missing a constrained attribute — price, size, a stated feature — drop out before subjective ranking even begins.

Why it matters for ecommerce

Draft carts move the moment of selection off the storefront. By the time a session arrives — if one arrives — the multi-brand comparison already happened inside the agent, against whatever data the agent could parse. A SKU whose attributes are machine-readable competes in every relevant draft; a SKU whose attributes live in description prose competes only when the agent guesses correctly.

This stage also rewards multi-item coherence: agents drafting a basket favor catalogs where sizing, compatibility, and bundle relationships are explicit. Variant structure and typed attributes — Shopify metafields, normalized options — stop being hygiene and start being distribution.

A draft in practice: an example

A buyer asks ChatGPT to put together the spring-skiing basics above. For the base-layer slot the agent considers four brands (illustrative example): one is eliminated because no men's medium shows as in stock in structured data, one because weight class is unstated, leaving two finalists ranked on review signal and price fit. The winning brand's PDP carried a 200gsm weight attribute, live availability, and aggregated review markup. The losing brands may never know the draft happened — no impression, no session, no signal in analytics.

How it relates to neighboring terms

Draft carts are stage two of agentic commerce — after research answers, before agent checkout. The readiness practice is ACO; the per-SKU readiness measure is the agent-readability score; and the structured layer agents evaluate is Product schema plus typed metafields.

How eCommerce Insights helps

The agent-readability score checks exactly the signals draft-cart selection runs on — structured attributes, machine-readable price and availability, review markup — and the recommended diffs fill the gaps that get SKUs eliminated silently. The free Agentic Readiness Grader shows where any single PDP stands.

Related terms

• ChatGPT Shopping — the highest-volume draft-cart surface.
• Perplexity Buy with Pro — Perplexity's transaction-capable equivalent.
• Agent checkout — the stage after approval, in pilot.
• Agentic commerce — the three-stage frame this sits inside.
• Agent-readability score — the measure of whether your SKUs survive drafts.