Being recommended by an AI assistant is the visible half of the shift. The quieter, larger half is transaction: assistants and autonomous agents that don't just name a product but discover it, compare it, and complete the purchase on the buyer's behalf. When the agent — not the human — is the one navigating your store, a different set of things has to be true.
Most sites are built for a person with eyes, a cursor, and patience. An agent has none of those. It reads structured data, follows documented endpoints, and gives up fast when a flow assumes a human.
What an agent needs to buy
To transact, an agent has to answer a sequence of machine questions about your catalog — and find unambiguous answers:
- What is this product? A
Productschema with a stable identifier, not just a pretty page. - What does it cost, and is it available? Explicit
price,priceCurrency, andavailability— in markup, not baked into an image. - Can it ship to me, and what are the terms? Shipping, returns, and policies it can read, not infer.
- How do I actually complete this? A path to checkout (or a handoff) that doesn't depend on a logged-in human clicking through a cart.
If any of those answers is missing or buried, the agent does the rational thing: it picks a competitor that answered cleanly.
The protocols to watch
A small stack of emerging standards is forming the rails for this:
- ACP (Agentic Commerce Protocol) — conventions for agents to browse catalogs and initiate purchases.
- UCP (Universal Commerce Protocol) — interoperable product, price, and availability data across stores.
- MCP (Model Context Protocol) — the increasingly common way assistants connect to external tools and data, including commerce surfaces.
You don't need to bet on a single winner. What they share is the underlying demand: structured, authoritative, machine-readable commerce data plus a documented way to act on it. Get that right and you're ready for whichever protocol your buyers' agents speak.
How to get ready — without rebuilding your store
- Make your catalog legible. Complete
Product/Offerschema with price, currency, availability, and SKUs on every product. - Publish your terms as data. Shipping and returns an agent can parse, not just a PDF.
- Don't block the agents. Many stores quietly disallow AI fetchers in
robots.txtand never realize they've opted out of the channel. - Expose an authoritative layer. A machine-readable manifest of what you sell, where, and how to transact — so an agent gets the truth from you, not a guess.
- Keep it current. Agents act on what they read; stale price or stock data converts into wrong orders or abandoned ones.
Why now
This won't arrive all at once, and it won't replace human shoppers. But the brands that are transactable by agents early get a structural advantage: they're the default an assistant can actually complete, while competitors are still a link a human has to chase. The cost of getting ready is mostly markup and discipline — cheap insurance against being the store the agent couldn't use.
Want to know whether an agent could discover, price, and check out from your site today? Run a free Legible report — commerce readiness is one of the eight dimensions it scores, with the fixes to close the gaps.