The AI attribution data pipeline:
GA4, server logs, CRM and a dashboard
AI attribution needs five data sources that answer five different questions: GA4 sessions, server logs, CRM records, payments and a prompt panel. Join them on email, URL and utm values in a warehouse, then chart the result in a dashboard so the whole story is one report, not five tabs.
Five data sources join in one place, so AI attribution is one report and not five tabs
Diagram of an AI attribution data pipeline. Not measured data.
Diagram, not measured data.
How to read it. Each source answers one question. The join keys, email, URL and utm, are what let a crawler hit, a visit, a signup and a payment line up.
In ThriveStack citedby The sources are connected once, and the joins are built, so there is no pipeline for your team to maintain. See it for your site →
AI attribution needs five data sources, and each answers one question.
No single system holds the whole path from citation to payment. Each source holds one link in the chain.
Five data sources for AI attribution, and the question each answers
| Source | Question | Join key |
|---|---|---|
| GA4 sessions | Where did the visit come from? | utm, landing page, client id |
| Server logs | Which bots read which pages? | URL |
| CRM | Who signed up? | |
| Billing | Who paid? | Email or customer id |
| Prompt panel | Which engines cite us? | URL |
GA4's BigQuery export keeps session level data you can join.
GA4 can export raw events to BigQuery. That gives you the source, medium and landing page for each session, which you can join to CRM contacts.
- 1
Link GA4 to BigQuery
Turn on the export in GA4 Admin.
- 2
Load server logs into BigQuery
Send log lines to the same warehouse.
- 3
Load CRM and billing data
Export contacts, deals and payments.
- 4
Join the sources and label AI visits
Join on email, URL and utm. Label sourced and influenced separately.
A dashboard can chart the joined data without a new tool.
Looker Studio connects to BigQuery and to spreadsheets. Build four panels and you have a working AI attribution view.
- AI sessions by engine
- Crawler hits by bot type
- Signups by first source
- Revenue by first source, sourced and influenced
Write down the rules for joining the five sources.
The joins are where numbers go wrong. Written rules make them repeatable and auditable.
Which field is the first source. What counts as sourced and what counts as influenced. How a missing referrer is labelled.
ThriveStack citedby runs your prompt panel across 10+ AI engines, reads server-log crawler hits, matches human visits to signups and reports AI-sourced and AI-influenced pipeline.
AI traffic data pipeline, answered.
What data do I need for AI attribution?
GA4 sessions, server logs, CRM records, payments and a prompt panel.
Do I need a warehouse?
Not at first. A spreadsheet works for one site. A warehouse helps when the joins grow.
What should I join on?
Email for people, URL for pages and utm values for sources.
All 10 questions this lesson answers
AI traffic in Looker Studio
Looker Studio reads GA4, so start there. GA4 now has an AI Assistant default channel, added in May 2026, that groups visits from chatbots like ChatGPT, Gemini and Claude. Chart that channel by source and landing page. It only catches visits that carry a recognizable referrer.
AI traffic in BigQuery
In the GA4 export, use session_traffic_source_last_click.manual_campaign.source and medium for session level attribution. Filter the source with a regex such as chatgpt, perplexity, claude, gemini and copilot. Visits with no referrer still show as direct, so BigQuery cannot fix missing data.
Track AI traffic in Adobe Analytics
Adobe Marketing Channel processing rules include a Referrer Is Conversational AI condition. Place your AI rule above the generic referral and direct rules, because rules run in order. Changes are not applied to past data, so set the rule up early.
Track AI traffic in Mixpanel
Mixpanel's JavaScript library tracks UTM tags and the $referrer and $referring_domain properties by default, plus initial values for the first visit. Group chatgpt.com, perplexity.ai and similar domains into one AI property. If a link uses noreferrer, the property is not sent.
AI traffic in Amplitude
The Amplitude browser SDK captures UTM parameters, referrer and referring_domain, and saves first touch values as initial_ properties. Build a channel classifier that maps AI domains to an AI channel. Direct visits store an empty value, so do not read EMPTY as AI.
Track AI traffic in HubSpot
HubSpot labels recognized visits as AI Referrals in its traffic source properties. Its rules run in a set order, so a ChatGPT click tagged utm_medium=cpc lands in paid search instead. Review the order and test one tagged click.
AI traffic in Salesforce campaigns
Salesforce needs help to see AI traffic. Capture UTM values in hidden form fields, store them on the Lead, and copy them to the Campaign Member when a response is logged. Without those fields, Salesforce only knows the Lead Source you typed in.
AI referral tracking in Segment
Analytics.js reads UTM parameters into context.campaign and the referrer into context.page.referrer on its own. UTM values do not carry to later calls unless you pass them. Add a rule in your warehouse that tags AI domains, then send the result to every destination.
AI traffic in Plausible
Plausible groups visits from ChatGPT, Claude, Gemini, Perplexity and Copilot into an AI Assistants channel. It filters known AI crawlers and bots, so the channel shows people who clicked. It runs without cookies, which makes setup light.
AI traffic in Matomo
Matomo 5.5.0 added an AI Assistant referrer channel for ChatGPT, Copilot, Gemini, Claude, Le Chat, Meta AI, iAsk and Perplexity. Find it under Acquisition, then AI Assistants. It needs referrer data, so visits without one are not attributed.
Sources, checked November 2026: Google: BigQuery traffic attribution data · Search Engine Journal: GA4 AI Assistant channel · Adobe: Marketing Channel processing rules · Mixpanel: Traffic attribution · Amplitude: Marketing Analytics SDK · Segment: Common fields spec · Plausible: AI traffic · Matomo: AI Assistants.