For twenty years the job of an online store was to convince a human. Good
photos, a clear price, a trust badge near the checkout button. In 2026 a second
kind of shopper is arriving, and it does not care about your hero image. It is
an AI agent acting on someone's behalf: "find me a waterproof jacket under 120
euros that ships to Lisbon by Friday, and buy it."
This is agentic commerce, and it stopped being a slide in a keynote this year.
OpenAI and Stripe shipped the Agentic Commerce Protocol in ChatGPT in late 2025.
Microsoft's Copilot Checkout went live in the US in January 2026 with Shopify,
PayPal and Stripe support. Google announced its own Universal Commerce Protocol
and a "Buy for me" flow inside Search and Gemini. Morgan Stanley thinks nearly
half of online shoppers will use a shopping agent by 2030. You do not need to
believe the most bullish forecast to see the shift: a growing slice of your
traffic will soon be machines deciding whether your product makes the shortlist.
The stores that win that shortlist are the ones that made themselves legible to
an agent on purpose.
What "agentic commerce" actually means
Strip the hype and an agentic purchase is a chain of steps that used to need a
person: understand the request, search across stores, compare options on price
and delivery, pick one, and complete checkout. The new part is that a model runs
the whole chain, and increasingly completes the payment without the buyer
visiting your site at all.
That last point is the one merchants underestimate. In a classic funnel the
agent might send a human to your product page. In the protocol-based flows now
live in ChatGPT and Copilot, the agent reads a structured feed of your catalog,
builds the cart, and pays through a payment token, all server to server. Your
beautifully optimized landing page never renders. What gets evaluated is your
data: title, price, availability, shipping, returns, reviews.
The protocols you will keep hearing about
Two competing standards are forming, and you will likely support both rather
than bet on one.
ACP (Agentic Commerce Protocol), from OpenAI and Stripe, has been live in
ChatGPT since September 2025. It lets an agent fetch a product feed and submit
an order, with payment handled through a delegated token so the merchant still
gets a normal Stripe charge.UCP (Universal Commerce Protocol), Google's coalition-backed answer
announced in January 2026, powers the "Buy for me" experiences in Search AI
Mode and Gemini.
The practical takeaway is not to memorize the spec. It is that both depend on a
clean, machine-readable representation of your catalog and your policies. If you
sell on Shopify, much of this is being wired in for you. If you run a custom or
headless store, supporting these flows is an integration project, the same shape
as any other agent build where the model is easy and the plumbing is the work.
Why this is really an SEO and data problem
Here is the uncomfortable part for a lot of brands: agentic commerce rewards the
same things AI search already does, and punishes the same neglect. An agent
choosing between five jackets is doing structured comparison. It wants accurate
structured data (Product and Offer schema), a real-time feed with correct stock
and price, unambiguous shipping and return terms, and reviews it can parse.
Vague copy and missing fields do not just hurt conversion now, they get you
filtered out of the consideration set entirely.
This is the commerce edge of the shift we wrote about in
generative engine optimization
and llms.txt: being citable, and
now being buyable, by machines. A store with sloppy product data is invisible to
an agent the way a slow site is invisible to a human who bounced.
The new ranking factors are boring on purpose
Accurate price. Live stock. Clear delivery dates. Honest return policy. Parseable
reviews. None of it is glamorous, and all of it is exactly what an agent reads
before it decides your product is or is not on the list.
What to actually do in 2026
You do not need to rebuild your store this quarter. You need to stop assuming
every shopper is a human with eyes.
Audit your product data first. Make sure every SKU has correct Product and
Offer structured data, accurate availability, and shipping and return terms
that a machine can read without guessing. This is the highest-leverage work
and it helps human SEO too.Get your feed real-time. An agent that quotes a stale price or sells a
sold-out item creates a support mess and a trust problem. Stock and price
accuracy stop being nice-to-haves.Decide your protocol posture. On Shopify, turn on the agentic checkout
features as they roll out and test them. On a custom stack, scope ACP and UCP
support like any integration: payment tokens, order APIs, webhooks.Watch agent traffic separately. Start distinguishing bot and agent hits in
your analytics so you can measure this channel instead of guessing. You cannot
optimize a funnel you are not even counting.Keep the human path excellent. Most revenue is still humans clicking. This
is a second front, not a replacement, so do not let the shiny new channel
starve the one paying the bills today.
The honest read for 2026
Agentic commerce is early, US-first, and noisier than the headlines. Most
European stores will see modest agent-driven sales this year. But the work it
demands, clean structured data, a trustworthy real-time feed, clear policies, is
work you should do regardless, because it pays off in human search and
conversion immediately and positions you for the agent channel as it grows.
The mistake is treating it as a 2030 problem. The stores that are legible to
agents in 2027 are the ones that fixed their product data in 2026. If you want a
read on where your catalog stands, or help wiring up an agentic checkout flow
on a custom store, talk to us. We will tell you honestly whether
this is worth your time yet, and what to fix first if it is.
Top comments (0)