Generative commerce is online shopping mediated by generative AI: instead of typing keywords and clicking through filters, shoppers describe what they need in plain language, and an AI assistant discovers, compares, and recommends products — increasingly completing the purchase inside the conversation. The buying decision moves off the retailer's website and into the model's answer, so brands now compete for a place in an AI recommendation rather than a rank on a results page.
Key takeaways
- Generative commerce replaces keyword search with conversational discovery: the AI reads specs, reviews, and prices across sources and returns one synthesized recommendation, often as product cards.
- The transaction layer is live. ChatGPT's Instant Checkout, built on the open Agentic Commerce Protocol with Stripe, and Google's agentic checkout both complete purchases on the shopper's behalf.
- Placement is a machine-readability contest: Product schema, accurate real-time feeds, and a deep review corpus decide whether an agent can parse — and trust — your catalog.
- The metrics change with the medium. Share of Model, product-card activation, and share of card matter more than raw sessions once decisions conclude inside the answer.
- Generative commerce names the whole AI-mediated shopping funnel; agentic commerce is its execution layer, where agents actually transact.
How generative commerce works
Follow one purchase end to end. A shopper types "I need a waterproof running jacket under $150 that packs down small" into ChatGPT. The engine fans that single sentence out into a set of grounding queries — best packable rain jackets, waterproof rating comparisons, running jacket reviews under $150 — retrieves data from merchant feeds, review sites, and editorial roundups, then compresses everything into a short answer: two or three products with images, prices, and a sentence of rationale each.
Three properties separate this from a search results page. Synthesis: the model has already read the reviews and spec sheets, so the comparison work the shopper used to do across ten tabs happens before the answer renders. Personalization: the same prompt from a trail runner in Seattle and a treadmill user in Phoenix can produce different shortlists. Action: with checkout protocols in place, "buy the second one in medium" is now a valid next message.
The interface is no longer text-only, either. Google's AI Mode draws on the Shopping Graph — more than 50 billion product listings, roughly 2 billion of them refreshed every hour — to answer photo-based queries like "find a table that matches this chair."
