Tracking & attribution engineeringCapability
Server-side capture for orders taken by AI checkouts: each order tagged agentic in GA4, and sent to ad platforms only when it carries a usable identifier.
Sounds familiar?
How I solve it
I build server-side order capture for AI-channel orders. A webhook takes each order as Shopify writes it, the order is classified and tagged agentic, and it is sent to GA4 through the Measurement Protocol. It goes to Meta CAPI or Google Ads only when the order carries an identifier those platforms can actually use — a click id, a matching browser id, or a hashed identifier you have consent to send. Anything else is recorded in GA4 and left out of the ad platforms, on purpose.
The classification reads what Shopify itself records about where an order came from. In the Admin API that is Order.attribution (from API version 2026-07), Order.app, Order.publication and sourceName. Shopify has deprecated channelInformation, so I don’t build on it. ChatGPT buyers pay on your own checkout, so for them the signal is the visit. The API keeps a visit’s referrer, landing page and UTM parameters in customerJourneySummary, though nobody documents how much of that survives ChatGPT’s in-app browser. What Shopify doesn’t document, as of 2 October 2026, is which values its AI channels write into these fields. On your store I read them on real AI-channel orders first, and only then fix the rules.
The process
6 steps. Scope and a fixed price are agreed before the first one.
orders/create and orders/paid — the order is the source of truth, not the browser.
Rules read Shopify's own source fields on the order (attribution, app, publication, sourceName), with the values confirmed on your real AI-channel orders first. agentic orders then stay separate in reporting instead of blending into "direct".
A Measurement Protocol event for each eligible order, so the order reaches GA4 whether or not the browser ever loaded.
Present → the event also goes to Meta CAPI / Google Ads. Absent → GA4 only, and the order is flagged as identifier-free rather than padded.
Every order is matched against what GA4 and the ad platforms recorded: what arrived, what did not, and why.
Which AI channel produced which orders, and what share carried a usable identifier.
Handover
agentic, separated from direct and organic.Built with
Proof
Capability Work I do; the proof below is from related projects.
Before → after
The closest real result is adjacent work, not an AI-channel one: the headless store rebuild behind the same pattern. 573 of 612 sales were invisible to every report before the rebuild, and the share of sales traceable to a source went from 6% to 76% — read from the live GA4 and Google Ads APIs in the August 2026 handover.
Headless Shopify automotive-repair retailer, US market
573/612
573 of 612 sales were invisible to every report before the rebuild; sales traceable to source 6% → 76%.
Read the case: 16 Days Before Shopify's Deadline: Rebuilding Measurement for a Headless Store[He] is very responsive and is an expert at fixing pixel and tracking issues in Meta and Shopify.
What you can look at
I built a demo with synthetic orders: channel labels, GA4 purchase payloads and weekly reconciliation. You can read the demo repository on GitHub; it builds ad-platform payloads but never sends them.
I have no AI-channel client results yet. What you can read is the adjacent work: the case file for the headless store rebuild, built on the same server-side pattern.
To check from outside whether your storefront tags can see AI click-outs at all, run the free AI Shopping Readiness Check. Its last row, AI orders in your analytics, also explains which AI orders browser tags never reach.
I have no AI-channel client results yet. The figures on this page come from adjacent work, a server-side measurement rebuild for a headless store. Client details are anonymised, and the figures were read from live platform APIs in August 2026.
Price and timeline
A store build starts by reading how Shopify labels your AI-channel orders. Then it is priced as a Focused Fix from USD 450, with a fixed quote before it starts. If you only need one question answered — whether your AI-channel orders are being captured at all — a Working Session is USD 195.
The rest of the AI shopping work, from channel settings to product data, is on Shopify in AI shopping channels.
QuoteFixed before work starts
A note from Daniilbefore you decide
If AI-channel orders are a rounding error in your revenue and stay that way, the honest answer is to wait. If you sell neither to the US nor, through Meta, to Canada or Mexico, Shopify’s AI channels don’t offer your buyers in-chat checkout as of October 2026, so this capture has nothing to catch.
If you already run a server-side container that captures these orders, you may only need the classification and the eligibility rule, not a new pipeline. And if your ad platforms are already receiving valid, identifier-carrying AI-channel conversions that match Shopify, there is nothing here to fix.
— Daniil
Questions
The capture works from the order, so it covers any checkout that writes an order into Shopify. Shopify names four AI channels: ChatGPT, Google AI Mode and Gemini, Microsoft Copilot, and Meta. It says their orders show in your admin with channel or referrer attribution, and that Meta orders are attributed to Meta. ChatGPT sends buyers to your own checkout, so its orders arrive as referrals. Google, Copilot and Meta can take payment inside their own surface, and that is where your tags stop. As of 2 October 2026, Shopify does not document the exact Admin API values each channel writes, so I read them on your real orders before the rules go live.
Yes — through the Measurement Protocol. That is the point: the order reaches GA4 even when no browser session ran. With no session behind the order there is no real client ID to join it to, and Google says Measurement Protocol data on its own may give only partial reporting.
A click id, a matching browser id, or a hashed identifier you have consent to send. If none is present, the order stays in GA4 and is deliberately kept out of the ad platforms.
Because an identifier-free conversion teaches the platforms to optimise on noise. It also inflates reported conversions, which is the opposite of what you asked for.
Not for a client yet. AI-channel orders are new, and I have no client results for them. The proof on this page is adjacent work, the same server-side pattern in a measurement rebuild for a headless store, and I label it that way.
The first answer is free, within one working day. Or write directly: next@taskfordaniel.com