Skip to content

Agentic commerce: what shopping agents mean for measurement

Agent purchases reach a store either as a normal session on your own checkout or as an order with no session. GA4 sees the first; for the second you rely on your commerce platform and whatever attribution the platform chooses to pass.

€99 once, excl. VAT. Full refund within 30 days, no questions asked. You keep the read.

By , Founder & CEOPublished 6 min read

Shopify's Help Center documents two shapes for a purchase that starts in an AI channel, and only one is a session on your store. ChatGPT users finish on the store's own checkout, while on Google AI Mode and Gemini, Microsoft Copilot and Meta a shopper can buy through direct checkout, where it is switched on, without leaving the AI channel. In the second shape none of your pages loads, so you can measure the order and whatever attribution the platform chooses to pass. Count agent orders where the platform reports them, then reconcile to total orders.

What is agentic commerce, and what do the builders say?

Stripe's documentation defines it as AI agents supporting "transactions between buyers and sellers", including agents that "transact on behalf of the people they serve". The dated primary announcements:

  • 29 September 2025: OpenAI launched Instant Checkout with the Agentic Commerce Protocol (ACP), built with Stripe, and said orders, payments and fulfillment stay with the merchant.
  • 11 January 2026: Google's developers blog introduced the Universal Commerce Protocol (UCP), developed with Shopify, Etsy, Wayfair, Target and Walmart, and says the business stays the merchant of record.
  • 19 May 2026: Google introduced Universal Cart: a shopper checks out with Google Pay or transfers items to the merchant's site, and the brand stays the merchant of record.
  • Undated page, read today: OpenAI's merchants page says it is "moving away from a standalone Instant Checkout experience" and that purchases are completed on merchant-owned websites or apps.

What does the store receive when an agent buys?

The documents differ:

  • OpenAI's ACP. OpenAI still publishes the checkout spec, which has merchants run checkout flows inside ChatGPT while keeping orders, payments and compliance on their own commerce stack. Its Buyer object holds a name, an email and an optional phone number, and the order holds an ID, a checkout session ID and a permalink. It lists no campaign or referrer field. OpenAI's best-practice page tells merchants to add a UTM parameter such as utm_medium=feed to feed URLs for click tracking.
  • UCP (Google, Shopify and others). An optional attribution field carries campaign and click parameters. The spec calls it "informational and optionally provided by the platform", says UCP "does NOT prescribe attribution models, windows, or assignment logic", and describes it on an order as "a business-emitted snapshot of the originating checkout's attribution".
  • Shopify. Orders from AI channels show in the admin "with channel or referrer attribution". Its Agentic section reports sales, orders, online store sessions and conversion per AI channel, and sales are "an aggregate of referral-based sales and sales that happen through direct checkout". On its Agentic plan, for businesses without a Shopify online store, "You can't review your ChatGPT order history in Shopify."

In all three the merchant keeps the order: OpenAI and Google say the merchant stays the merchant of record, and Shopify says you keep full ownership of the buyer relationship and the post-purchase experience. What you lose is the path to the order. The gap follows from the flow: a page tag fires only on a page that loads, and in direct checkout the shopper never leaves the AI channel. That step is inference from the documents. What reaches you is whatever the platform passes.

What does the evidence say?

Ethercycle, a Shopify agency, reports 44,737 ChatGPT-referred sessions across ten US stores that are its retainer clients, July 2025 to June 2026: 3.1% completed checkout against 1.6% for all traffic, and it calls direct handoff to checkout "effectively zero" (agency-published, not independently audited; it says ten stores is not a sample for significance tests).

Two arXiv preprints, not peer-reviewed, look at how agents choose. Allouah and co-authors (revised December 2025) audited how AI agents pick products and report that preferences are unstable: "model updates can drastically reshuffle market shares". Wadi and Ma (revised September 2026) randomised the order of one hundred hotel listings across 5,000 AI agent sessions with four language models and report that AI agents "never decline to buy". That is hotels, not retail, but it is a reason to keep agent conversion apart from human conversion. None of the sources cited here says how many orders agents place for a typical store.

How do you measure agent orders without guessing?

Four steps, in order:

  1. Count where the platform counts. On Shopify, Sales channels, Agentic shows orders and sessions per AI channel. Elsewhere, ask your commerce or payment provider which order events and fields it sends.
  2. Put GA4 beside it. For the same dates and time zone, list the purchases GA4 credits to its AI Assistant channel and subtract them from the first step's orders. The remainder is agent orders GA4 gave no channel credit (how to read it). Keep it as its own line, not inside Direct.
  3. Ask for attribution in writing. Does each order carry any, and what is stored for how long? UCP says businesses "SHOULD NOT persist signal data beyond the operational needs of the transaction", so fraud signals are not analytics.
  4. Test the channel, not the dashboard. Shopify lets you turn product access off for ChatGPT, Microsoft Copilot or Meta, and its page says it can take up to 7 days before product data stops being shared. First measure your usual week-to-week swing in total orders. Then switch one channel off for full weeks in an ordinary period, back on, and compare total store orders, not that channel's count. Pass, the channel adds orders: total orders fall by more than the swing while it is off. Fail, it re-routes orders or is too small to see: no fall beyond the swing. Log the date of every agent-side change.

Sources, 30 September 2026: Agentic commerce (Stripe, 2026); Buy it in ChatGPT (OpenAI, 2025); Power product discovery in ChatGPT (OpenAI, 2026); Agentic Checkout Spec (OpenAI, 2026); Best practices (OpenAI, 2026); Under the Hood: UCP (Google Developers Blog, 2026); UCP specification (ucp.dev, 2026); Introducing the Universal Cart (Google, 2026); Shopify agentic storefronts (Shopify, 2026); Managing agentic storefronts (Shopify, 2026); Agentic plan setup (Shopify, 2026); Default channel group (Google Analytics Help, 2026); State of Ecommerce 2026 (ethercycle, 2026); What Is Your AI Agent Buying? (Allouah et al., arXiv, 2025); Does Rank Still Matter? (Wadi and Ma, arXiv, 2026).

Frequently asked questions

  • What is agentic checkout?
    A purchase completed inside an AI interface instead of on the store's own pages. Shopify calls its version direct checkout: the shopper buys in the AI channel without leaving it, and the merchant stays the merchant of record. OpenAI's merchants page says it is moving away from a standalone Instant Checkout.
  • Which agentic commerce solution is best?
    None of the sources cited here ranks them. OpenAI's protocol, Google and Shopify's UCP, Stripe's tooling and Shopify's managed channels do different jobs. Choose by where your checkout lives, then measure orders per channel against total orders.
  • What does an agent's order carry that I can use for attribution?
    It depends on the protocol. OpenAI's checkout spec lists a buyer name, an email and an optional phone number, and no campaign field. UCP has an optional attribution field that the platform may or may not send. Shopify shows AI-channel orders with channel or referrer attribution.

Go deeper: Causal attribution, explained.

Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.

Keep reading

Terms in this article

Browse the full glossary

Your platforms guess.
We run the math.

Upload a GA4 export and see what each channel caused, next to last-click, in 1–2 minutes. The read is yours to keep.

Free, in your browser: your file is not uploaded. The full read is €99, refundable within 30 days. Prices exclude VAT.
Or book a 30-min call.