78.9% is ChatGPT’s share of AI referral page views.

Use the same source domains, but measure the entrance page and the conversion event in one funnel. This keeps referral volume separate from what visitors did after the click.

Referral and landing-page report

1. Open Web analytics, set the last 30 days, and review Acquisition or Referrers for the AI source domains in the table below.
2. Compare visitors, sessions, views, and conversion goals by referring domain. Keep broad aliases separate until you have checked the raw values.
3. Open Product analytics and create a funnel. Make the first step $pageview, filtered where $referring_domain matches the AI source regex.
4. Use your real signup, lead, checkout, or purchase event as step two. Break down the entrance $pageview by $pathname to compare landing pages.

PostHog fields to use

Referrer

$referring_domain on the entrance $pageview

Source filter

Matches the same anchored AI domain regex used for GA4

Landing page

$pathname on the filtered entrance $pageview

Conversion

Your real lead, signup, checkout, or purchase event

First touch

$initial_referring_domain only when the question is first acquisition, not the current visit

Filter the entrance event, not every funnel step. A conversion event may not carry the original referrer property.

Values recognized by the current panel

chatgpt.com, www.chatgpt.com, sora.chatgpt.com, chat.openai.com, chatgpt, openai, openai.com

Operational rule

Keep the raw source dimension in exports so a broad alias can be removed later without rebuilding the report.

• Treat chatgpt.com and chat.openai.com hostnames as high-confidence sources. Review openai and openai.com because those values can include non-ChatGPT referrals or tagged campaigns.
• ChatGPT app-to-browser handoffs, redirects, and privacy settings can strip the referrer and move a real click into Direct.
• GA4 does not expose the ChatGPT prompt, answer, or citation behind a click. Use landing pages and your own key events as separate evidence.

Keep blind spots and bots in separate reports

• In-app browsers, app-to-browser handoffs, redirects, privacy settings, and referrer policies can strip the source. Those visits can appear as Direct or have an empty referrer.
• Gemini referrals can show as gemini.google.com, but Google AI Overviews and some AI Mode clicks can remain google / organic. Standard web analytics cannot reliably isolate them from conventional Search clicks.
• Treat dark AI traffic as an unmeasured remainder, not a number to relabel. Report identifiable AI referrals as the observed minimum and keep Direct and unassigned movement as a separate watchlist.
• A source domain identifies the sender when the referrer survives. It does not reveal the prompt, answer, citation, or citation position that caused the click.
• AI crawlers are bots, not human referrals. Keep crawler activity in server or CDN logs, and exclude known bots and internal traffic from visit and conversion reporting.