Some of your traffic is no longer human

For most of the web's history, a session meant a person. That assumption is quietly breaking. A growing share of product research, comparison, and increasingly checkout itself is being delegated to AI agents acting on a real buyer's behalf — inside ChatGPT, inside Google, inside a browser assistant, inside a retailer's own app.

This is agentic commerce, and the important part for a business owner is not the technology. It is that a machine is now sometimes the thing evaluating your product page, and a machine reads a page very differently from a person. It does not respond to a hero video. It responds to whether your price, availability, variants, and return policy are stated in a form it can parse without guessing.

The three acronyms, in plain terms

Three standards get named constantly in this conversation, and they are usually presented as competitors when they mostly are not.

  • ACP (Agentic Commerce Protocol) — maintained by OpenAI and Stripe, launched alongside ChatGPT's Instant Checkout in late 2025. It is narrowly scoped to the checkout session: the handoff where an agent completes a purchase against a merchant.
  • UCP (Universal Commerce Protocol) — built by Google with Shopify, Etsy, Target, Walmart, Wayfair and more than twenty other companies. It is broader, covering discovery through post-purchase, is open source and free to implement, and went live with real US retailer checkout in May 2026.
  • AP2 (Agent Payments Protocol) — the payments rail underneath. It answers the liability question: who authorised this, for what, and for how much.

AP2 is worth understanding because it is the piece that makes the rest legally workable. Every agent purchase is expressed as three signed mandates — an intent mandate for what the user asked for, a cart mandate for what the agent assembled, and a payment mandate for what will actually be charged. That chain of signatures is what lets a merchant or card network prove a real person authorised a specific basket. Google published AP2 under an open licence in September 2025 and handed governance to the FIDO Alliance in April 2026, with roughly sixty contributing organisations including Mastercard, American Express, PayPal, Adyen, and Coinbase.

You probably do not implement any of this yourself

This is the part most coverage gets wrong. If you sell on Shopify, or through a major commerce platform, protocol support is arriving at the platform layer. You inherit it. Hand-rolling ACP or UCP endpoints is a decision for platforms and very large direct merchants, not for a D2C brand with a Shopify store.

What you are responsible for is the layer underneath, and that layer is unglamorous: whether your product data is accurate, complete, and machine-readable. An agent cannot buy what it cannot understand, and no protocol rescues a product page where the price is injected by a script and the stock status is a coloured dot with no text label.

Six things to fix before any of this matters

  1. Put the commercial facts in the HTML. Price, currency, availability, variant options, and delivery estimate should be in the server-rendered markup, not assembled client-side after load. This is the single highest-impact change, and it helps human-facing search at the same time.
  2. Ship real Product and Offer structured data. Not a partial block that validates but omits availability, priceCurrency, and sku. We cover the details in our guide to schema markup in 2026.
  3. Write attribute-complete product copy. Agents match on attributes. Material, dimensions, compatibility, care instructions, and what is in the box are now ranking and matching signals, not filler.
  4. State returns and shipping in text. A returns policy locked inside a PDF or an image is invisible. Agents weight return terms heavily when comparing near-identical options.
  5. Keep URLs stable. Agents cache and revisit. A product whose URL changes on every catalogue update looks discontinued.
  6. Do not block the crawlers you want. Aggressive WAF rules and bot filters routinely block the exact agents you are hoping will recommend you. Check what your firewall is actually rejecting.

What this does to brand

An agent flattens a product to a comparison table. Photography, motion, and art direction do not survive that step — which sounds like bad news for brand investment and mostly is not, because it moves the value of brand earlier and later in the journey rather than removing it.

Earlier: whether the buyer types your name into the agent at all, or asks a generic category question and takes whatever comes back. Later: whether they trust the recommendation enough to complete. Both are brand problems. If your store competes purely on being the cheapest row in a table, an agent will find someone cheaper. This is the same argument we make about generic-looking brands, with higher stakes.

The realistic timeline

Treat trillion-dollar projections as directional. What is verifiable today is narrower: the protocols exist, the largest retailers and platforms have committed to them, checkout has gone live in production, and payment authorisation has real governance behind it. The direction is not in doubt. The pace is.

Which is why the sensible response is not a rebuild. It is a data and markup pass that pays for itself immediately in ordinary search and conversion, and positions you for agent traffic whenever it arrives at volume. If your checkout already loses human buyers, fix that first — our piece on checkout abandonment is the better starting point, because an agent will abandon for the same structural reasons a person does, only faster.

If you want an honest read on whether your store is legible to machines, send us the URL and we will tell you what an agent can and cannot see.