The short answer
A GA4 AI referral dashboard filters session source or source/medium for known AI referral domains, then reports sessions, landing pages, engaged sessions, conversions, and revenue beside an explicit source list.
Business outcome
You can see which AI referrals reach the site and what those visitors do without pretending untracked searches or citations are traffic.
The process
Build it in five passes
Inspect raw acquisition data
Open Traffic acquisition and export session source, session medium, landing page, sessions, engaged sessions, key events, and revenue. Look for exact source values already reaching the property before writing a filter. Preserve raw values because product labels and referral patterns change.
Create an explicit source rule
Build a regex or channel rule from observed and documented referral hosts. Start narrow with domains you can explain, such as chatgpt.com, perplexity.ai, claude.ai, and copilot.microsoft.com. Review unknown referrals rather than adding every domain containing 'ai.'
Report landing pages and outcomes
Show sessions and users, but lead with engaged sessions, key events, qualified leads, purchases, or another business outcome. Break results down by landing page to learn which evidence and content AI users choose to visit.
Keep visibility metrics separate
Do not combine referral sessions with answer mentions, citations, or recommendation share into a single traffic number. OpenAI notes that ChatGPT referral URLs include a source tag, but not every AI interaction creates a click and not every click will retain attribution.
Audit monthly
Review source values, self-referrals, redirects, consent behavior, and landing-page tags. Add newly observed sources with a dated note. Reconcile important conversions against the underlying event setup before presenting them as revenue impact.
Before it ships
Quality checklist
- The filter is based on session source or source/medium, not page text.
- Every included domain has a documented reason and review date.
- Raw source values remain available for audit and future reclassification.
- Landing pages, engaged sessions, key events, and revenue are visible.
- AI referrals are not labeled as all AI-search influence.
- Consent, redirects, cross-domain tracking, and event quality are checked.
Copyable artifact
Starter exploration specification
Use this as a build note for a GA4 Exploration or BI dashboard, then replace the starter sources with values observed in your property.
SEGMENT NAME: AI referral sessions SCOPE: Session INCLUDE: Session source matches regex STARTER REGEX: ^(chatgpt.com|perplexity.ai|claude.ai|copilot.microsoft.com)$ DIMENSIONS - Session source / medium - Landing page + query string - Date - Device category - Country METRICS - Sessions - Engaged sessions - Engagement rate - Key events - Session key event rate - Total revenue GOVERNANCE - Owner: [name] - Last source review: [date] - Newly observed sources awaiting review: [list] - Known attribution limitations: [notes]
Validation
How you know it is ready
- 01A sampled session source matches the dashboard classification and raw report.
- 02Key events fire once at the intended business milestone.
- 03Dashboard totals reconcile with the same filtered date range in GA4.
Do not overclaim
Referral analytics measure attributable visits, not all recommendations, citations, zero-click answers, or assisted decisions. Source values can change, and privacy controls or redirects can remove attribution.
Questions
What teams usually ask
Can GA4 show how often ChatGPT recommends us?
No. GA4 observes site activity after a visit. Recommendation and citation measurement require answer-level research or platform reporting such as Bing's AI Performance view.
Should AI referral traffic be a custom channel?
It can be useful, but retain the raw source and medium. A channel rule should make analysis easier without destroying the ability to inspect individual sources.
Why not use one giant public regex?
Referral hosts and product behavior change, and some lists mix assistants, search engines, developer tools, and unrelated domains. Start with observed sources and maintain a dated rule.
Sources reviewed