How to Track AI Referral Traffic in GA4
Configure GA4 to identify AI-assistant referrals, preserve source detail and report the traffic alongside visibility and citation metrics.
To track AI referral traffic in GA4, create a custom channel that groups known assistant referrers, validate it against source and referral-path data, and report landing pages and key events separately from prompt-based visibility. GA4 can measure clicks that arrive at your site; it cannot measure answer impressions, unclicked citations or recommendations that influence later branded visits.
Know what GA4 can see
When an assistant sends a normal web click, GA4 may record the referring domain. Some products add UTMs, while browsers, apps and redirects can remove referral information. A user may also see an AI recommendation and return later through search or direct navigation.
Therefore label the report “observable AI referrals,” not total AI influence.
Inventory referrers from your own data
In Reports, open Acquisition and inspect session source/medium. In Explorations, add:
- session source;
- session medium;
- page referrer;
- landing page plus query string;
- session campaign;
- key events and revenue where configured.
Search for known domains such as ChatGPT, Perplexity, Claude, Copilot and Gemini, but confirm the exact values found in your property. Products and referral patterns change.
OpenAI states that ChatGPT search referrals can include utm_source=chatgpt.com; use that as a validation clue, not the only rule.
Create a custom AI channel
GA4 supports rule-based custom channel groups and explicitly provides an AI-assistant example in its documentation. Create the channel under Admin → Data display → Channel groups.
Use source conditions based on observed domains. Keep AI referrals above a generic Referral rule because traffic is assigned to the first matching channel. Document the regex or domain list, owner and review date.
Do not overwrite the primary channel group immediately. Test the custom group alongside the default until classification is stable.
Build a useful exploration
Create a free-form exploration with:
- rows: session source, landing page;
- columns: month or week;
- values: sessions, engaged sessions, key events and revenue;
- filters: custom channel equals AI Assistants.
Add content group or page-type dimensions if configured. This shows whether assistants send traffic to guides, product pages, tools or brand pages.
Separate acquisition from attribution
Use session-scoped dimensions for the visit that arrived from an assistant. For conversion analysis, compare GA4’s attribution views and document the selected attribution model and lookback window.
Do not claim a key event was “caused by AI” merely because AI was one touchpoint. Report direct AI-referred conversions and AI-assisted paths separately where the available reports support it.
Preserve raw identifiers
Export source, medium, referral path, landing page and UTM values before applying a presentation label. A broad regex can hide platform changes or misclassify domains.
GA4 notes that custom channel groups are not available in the BigQuery export schema. Recreate the logic in SQL if you use exported event data.
Add annotation and change control
Maintain a change log for:
- new or retired referring domains;
- channel-rule edits;
- consent or tag changes;
- site migrations;
- major assistant launches.
Because custom channel groups can apply retroactively in reports, an edit may restate historical classification. Store a monthly extract when auditability matters.
Connect traffic to visibility data
Join GA4 trends to the prompt tracking library and citation-rate analysis. A cited page may receive no click, and a page may receive traffic after being linked without the brand being recommended.
Use a simple monthly table: observable referrals, landing-page mix, engaged-session rate, key events, prompt visibility and cited-page count.
Common mistakes
- treating Direct traffic as provably AI-driven;
- relying on a fixed third-party domain list without validating property data;
- mixing user acquisition and traffic acquisition scopes;
- using UTMs that the publisher does not control as if they were complete;
- comparing tiny conversion counts without uncertainty.
Frequently asked questions
Can GA4 show impressions inside ChatGPT or Perplexity?
No. GA4 measures activity on your tagged properties, not exposure inside external assistants.
Should AI assistants be Organic Search or Referral?
Keep a separate custom channel for analysis. GA4’s default classification may place many visits in Referral.
Can custom channels classify past data?
Google says custom channel groups can be applied retroactively in reports, but changing definitions can create reporting inconsistencies. Maintain change notes.
