What is agentic commerce?

What is agentic commerce? A plain definition

Agentic commerce is buying and selling in which an AI agent acts on a shopper’s behalf — finding products, reading prices and policies, comparing options, and, where the rails exist, paying. The human still decides; the agent does the legwork. Strip the buzzwords and the agentic commerce meaning comes down to one shift: the buyer is no longer only a person with a browser. It is also a program with delegated intent and a token budget.

That shift sounds small. It is not. Every layer of a storefront — theme, mega-menu, image gallery, review widget, newsletter modal — was built to persuade a human eye. An agent has no eye. It has a parser and a context window, and it judges your store on exactly one axis: whether the buying facts are legible.

Two species of buyer

Commerce is splitting into two species of buyer. Humans need a rendered visual experience. Agents need structured data. Both arrive at the same URL, and a store that serves only one of them is now half a store.

Seen this way, the storefront stops being the product. It becomes one output format among several, compiled from a single source of truth — catalog, pricing, policies — and rendered differently per requester. The constraint that keeps this honest: same substance, different format. Prices, availability, and policies never vary by who is asking; only the packaging does. Substance that varies by requester is cloaking, and search engines penalize it.

The protocol timeline: how this became real

Agentic commerce stopped being a thought experiment inside roughly one year. The dates matter, because they explain why merchant readiness moved from optional to urgent:

One shift inside this timeline is easy to miss. OpenAI moved away from in-chat checkout toward discover in chat, transact on the merchant’s site: the agent finds and shortlists, then the human lands on your store to finish. That makes the merchant-side experience — what your site does when an agent-referred buyer arrives — a live surface, not a hypothetical one.

Agentic commerce examples you can measure

The clearest agentic commerce examples are not concept demos — they are traffic patterns already visible in merchant data:

• A shopping agent reads a product page. A CLI or chat assistant fetches your PDP with Accept: text/markdown in the request headers — a standard HTTP way of asking for a machine-readable format. A typical store ignores the header and returns full theme HTML — our reference capture shows what that costs.
• An assistant recommends a store; the human buys. AI-referred visitors convert ~38% higher than search visitors — they arrive pre-qualified, having already compared options through the agent.
• Agent-integrated retailers pull away in peaks. Cyber Week 2025: retailers with agent integration saw ~7x sales growth versus those without.
• Protocol checkout. With ACP live since September 2025 and all three major card networks supporting agent-initiated payments, an agent can complete a purchase end to end where the merchant supports it.

Conversion lift, AI-referred visitors vs search

~38%

Cyber Week 2025 retail data

Cyber Week 2025 sales growth with agent integration

~7x

Cyber Week 2025 retail data

Content negotiation vs llms.txt, accurate retrieval

~4.2x

300k-domain study, mid-2026

That last number carries a warning worth reading twice. The same mid-2026 study found llms.txt alone produces no measurable citation lift. Publish one — it is a useful index — but the mechanism that actually moves retrieval accuracy is content negotiation: answering the Accept header with clean text.

The before and after, in tokens

Numbers make the problem concrete. Our reference capture asks one question — what does a shopping agent receive from a product page? — against the same demo store twice: before and after installing a gateway that answers Accept: text/markdown. The capture is deterministic, generated from committed fixtures with no live network, and re-verified on every run.

Roughly 69x less context for the same buying facts — and the last row is the part merchants underestimate. Size aside, the before-HTML leaves availability in a <span> with no label on it and tucks the shipping terms behind collapsed accordions, eighteen reviews of widget markup deep. The markdown opens like this:

# [Alpine Trail Pack 28L](https://basecamp-supply.example/products/alpine-trail-pack-28l)

- **Brand:** Basecamp Supply Co
- **Price:** $148.00
- **Availability:** In stock
- **Shipping:** Free US shipping on orders over $75; standard shipping $6.95.
- **Returns:** 60-day returns. Items must be unused with tags attached.

The facts get better, not just smaller: price, stock, shipping, and returns are front-loaded instead of recovered by markup archaeology. An agent working with ~328 tokens of signal answers its user faster, cites more accurately, and has no reason to guess.

What changes for merchants

The strategic change is upstream of any single feature: your store’s legibility to agents now sits in front of revenue the way crawlability sat in front of search traffic twenty years ago. When an assistant cannot extract your price and return policy cleanly, it recommends the store where it can. Concretely:

• Your analytics undercount a new audience. An agent visit is a fast fetch that runs no JavaScript — in most analytics it registers as a bounce, or not at all. Seeing this traffic requires request-level observation, not page-view scripts.
• Format becomes a serving decision, not a redesign. Nothing about your human storefront changes. Agent readiness means answering one HTTP header with clean text — middleware, not a re-platform.
• The cloaking line must be respected. Same substance in a different format is the defensible position. Different prices or availability per requester is not — and known search crawlers should always receive your canonical HTML.
• Handoff quality matters now. Under discover-in-chat, transact-on-site, agent-referred humans land on your normal pages — and AI-referred visitors convert ~38% higher than search visitors. Treat them as your best traffic, not an anomaly.

How to get ready this week

Readiness is not a re-platform, and the first steps are free. In order of payoff:

1. Look at what agents currently get. Fetch a product page twice — once with curl -H 'Accept: text/markdown' and once plain — or use the side-by-side preview in the Rebilder Console, which fetches your public pages both ways and works before you install anything.
2. Install a gateway. Rebilder’s agent gateway is free middleware: it detects agents, answers Accept: text/markdown with markdown rendered from your product data, and passes every human and search crawler straight through to your normal site. Adapters cover Next.js, Shopify, Express, Fastify, and Cloudflare Workers.
3. Ship llms.txt — with honest expectations. The gateway generates it from the same source data, or you can build one by hand with the free generator. It is a useful index; it is not, on its own, a citation strategy.
4. Watch the log. The Console records every classified request — agent, human, crawler, or protocol probe; which platform; what was served — so agent traffic stops being invisible.
5. Plan for protocols; don’t wait for them. UCP/ACP/MCP endpoint support is on the Rebilder roadmap — announced, not yet live. The same source-of-truth wiring that powers markdown today is the input those endpoints will read, so nothing you set up now is throwaway.