Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
The Protocol Question Is Not Your Question
There are two live standards for agent-driven purchasing and no resolution in sight. That sounds like a decision you need to make. It is not, because both protocols consume the same underlying thing: a product record a machine can read without guessing. Fix the record and the protocol question becomes your platform's problem rather than yours.
The Agentic Commerce Protocol arrived first, launched by OpenAI and Stripe on 29 September 2025 alongside ChatGPT Instant Checkout, with US Etsy sellers available at launch. It is open source, concentrates on the checkout step, and settles through Stripe. Google's Universal Commerce Protocol was announced at NRF in January 2026, reaches further up the funnel into discovery, and settles through Google Pay.
The politics moved quickly after that. On 24 April 2026 Amazon, Meta, Microsoft, Salesforce, and Stripe joined the UCP Tech Council alongside founding members Google, Shopify, Etsy, Target, and Wayfair, producing a ten-member governance body that includes companies backing the competing standard. Read that as a signal that nobody knows who wins either, which is a further argument for building on the part that is common to both.
What an Agent Cannot See
A person reading a product page infers an enormous amount from design: that the crossed-out number is the old price, that the green text means in stock, that the badge image means free delivery, that the greyed-out swatch means unavailable. An agent sees none of that. It sees whatever is in the markup, and treats everything else as absent.
| Fact | How humans get it | What an agent needs |
|---|---|---|
| Price | Visual hierarchy and strikethrough | A single unambiguous current price with currency, in structured data |
| Availability | Colour, badge, or a greyed-out option | An explicit availability value, kept current |
| Delivery time | A banner, or a separate shipping page | Shipping terms attached to the product record |
| Returns | A footer link nobody reads | Return window and conditions as machine-readable policy |
| Variants | Swatches and dropdowns | Each variant addressable, with its own price and stock |
| Identity | The photo and the title | GTIN, MPN, or brand identifiers that survive matching |
Work down the right-hand column against your own product template. Most stores fail on three or four rows, usually availability, shipping, returns, and variant-level pricing, and usually because those facts live in a theme setting or a separate policy page rather than in the product record. The markup foundations are in ecommerce product page SEO, and the general rule that structured data must describe what is actually on the page still applies here.
The Readiness Checklist That Is Worth Doing Now
Everything on this list improves conventional shopping surfaces, search appearance, and conversion whether or not agentic commerce ever reaches meaningful volume for you. That is the test for whether a readiness task is worth doing before the market arrives.
- Make availability true in real time.
Stale stock status is the single most damaging fact in an agent-mediated purchase, because the failure surfaces after a commitment rather than before one. If your availability updates nightly, that is a data pipeline problem worth fixing before any protocol conversation.
- Put shipping and returns in the product record.
Not only on a policy page. Both protocols expect these as fields, and shoppers comparing options weigh them heavily. A policy an agent cannot read is a policy that does not exist for the purposes of a comparison.
- Give every variant its own addressable record.
Price, stock, and identifiers per size and colour. A parent product with a "from" price and a dropdown is ambiguous to a machine, and ambiguity resolves as exclusion far more often than it resolves as a guess in your favour.
- Fix product identity.
GTINs, MPNs, and consistent brand naming are how an agent confirms that your listing and a competitor's are the same object. Without them you are not in the comparison, which is a quieter failure than losing it.
- Check what your crawler policy permits.
Some agent traffic arrives as a user-triggered fetch and will be refused by a blanket AI block, which means declining sales to make a point about training data. Separate those decisions using the crawler decision guide.
- Keep the human path excellent.
Agent-mediated purchasing is a minority of transactions today. A readiness programme that degrades the experience for the people actually buying from you has optimised for a forecast over a customer.
Trust Is the Constraint, Not Technology
The bottleneck on agentic commerce is not whether agents can transact. It is whether people will let them. A Checkout.com study in June 2026 found consumers wanted concrete controls before delegating purchases, spending caps and instant revocation among them, with three quarters of merchants surveyed calling real-time permission revocation critical to adoption.
That has a direct implication for merchants that gets lost in protocol coverage. The commercial risk in agent-mediated selling is concentrated after the sale: disputed authority, cancellations, returns initiated by someone who did not personally choose the item. A store whose returns process is painful will find that pain amplified, because the buyer feels less ownership of a decision they delegated.
So the readiness question is not only "can an agent buy from us". It is "can we handle an order placed by software on someone's behalf, and cancel or reverse it cleanly when asked". That is an operations question with no protocol attached, and it is the one most likely to determine whether early agentic orders are profitable or expensive.
What to Defer, Honestly
Direct protocol integration, custom agent endpoints, and anything sold as agentic commerce consulting. If you run on a major hosted platform, support will ship as a feature. Building it yourself in 2026 means maintaining an integration against a standard that may lose.
Be equally sceptical of the market-size numbers being used to sell urgency. Forecasts of multi-trillion-dollar agentic commerce markets by 2030 are projections, not measurements, and the gap between a protocol existing and meaningful revenue flowing through it for a small store is currently large. Nothing in this guide depends on those forecasts being right, which is deliberate: every task in the checklist above pays for itself through ordinary shopping surfaces.
The adjacent development actually worth watching is on the browser side, where a proposed standard lets a site declare what an agent may do rather than being scraped for it. That is covered in our guide to WebMCP. Until either standard settles, the durable work is the same as it has always been for ecommerce: accurate data, complete records, and a catalogue that can be crawled and understood, which is the ground covered by the ecommerce SEO checklist.
Questions People Ask About Agentic Commerce
- What is agentic commerce?
Agentic commerce is a purchase where an AI agent does some or all of the work on a person's behalf: finding products, comparing them, and in some implementations completing checkout without the buyer visiting the store. The distinguishing feature is that the thing reading your product data and making a decision is software rather than a person looking at your design.
- What is the difference between UCP and ACP?
The Agentic Commerce Protocol was launched by OpenAI and Stripe in September 2025 alongside ChatGPT Instant Checkout; it is open source, focused on checkout, and settles through Stripe. Google's Universal Commerce Protocol was announced at NRF in January 2026, covers discovery as well as checkout, and settles through Google Pay. They are competing standards with overlapping backers rather than complementary layers.
- Which protocol should a small store adopt?
Most small stores should adopt neither directly and instead wait for their platform to support whichever wins, while doing the underlying data work that both require. If your store runs on a major hosted platform, protocol support will arrive as a feature rather than an integration project. What will not arrive automatically is accurate, complete, machine-readable product data, and that is the part worth your time now.
- How do I make my product pages readable by AI agents?
Expose the facts an agent needs as structured data rather than as design: price with currency, real-time availability, shipping timelines, return terms, and variant-level detail such as size and colour. Anything that a human infers from layout, from a photo, or from a badge image is invisible to a machine, and an agent that has to guess will usually pick the competitor it did not have to guess about.
- Is agentic commerce actually happening or is it hype?
Both, and the split matters. The infrastructure is real: two protocols are live, major platforms and retailers have joined governance bodies, and checkout integrations exist in production. The volume flowing through it is still small relative to conventional ecommerce, and forecasts of its eventual size are forecasts. The sensible posture is to do the data work, which pays off regardless, and to defer the integration work until your platform makes it cheap.
