Ask ChatGPT, Gemini or Perplexity "what's the best robot vacuum under $400" and something structurally new happens: an AI assistant reads structured data from across the web, forms an opinion, and hands the shopper a shortlist — often with a buy button attached. The shopper never sees page two of Google, never opens ten tabs, and often never visits the brands that lost.
That is agentic commerce: shopping in which AI agents do the discovering, comparing and increasingly the transacting on a human's behalf. This guide explains what actually changed, how the agentic purchase funnel works, and what a brand has to publish for AI agents to recommend — and sell — its products.
Key Takeaways
- Agentic commerce shifts the buyer from a person to an AI agent. Agents don't browse product detail pages; they consume structured catalogs, schema and brand context — so the assets that win are machine-readable, not visual.
- The traffic is real and compounding. Visa reports AI-driven traffic to US retail sites grew over 4,700% in a year, and platforms responded by building standard rails — ACP, UCP, AP2, MCP — instead of ad-hoc scraping.
- Discovery is the layer brands actually control. After OpenAI pulled back from Instant Checkout toward discovery that hands shoppers to merchant sites, whether AI recommends you at all became the battleground — checkout increasingly returns to your own store.
- The playbook is context, not ads. Publish one authoritative, agent-readable version of your catalog, claims and brand story; serve it consistently across engines; measure where AI recommends you and fix the gaps.
From search-and-browse to ask-and-delegate
The web's commerce funnel assumed a human at every step: search, click, compare, add to cart, check out. Agentic commerce collapses those steps into a conversation. The shopper states an intent ("hypoallergenic pillow for side sleepers, under $80"), and the agent expands it into dozens of sub-queries, reads reviews, specs and availability, then returns a decision-ready answer.
Three consequences follow. First, the consideration set shrinks: an agent surfaces three to five options where a search results page surfaced twenty. Being the cited answer matters far more than ranking eighth. Second, the inputs change: agents weight structured data, consistent claims and third-party evidence over hero images and landing-page copy. Third, : with protocols like ACP and Google's UCP, an agent can carry the shopper all the way to a completed order — or hand them to your checkout to finish.
