AI Referral Traffic in GA4 and PostHog: A Two-System QA Guide

Direct answer

Check whether your GA4 property already shows an AI Assistant channel, then inspect session source and landing page. If your property needs a separate definition, create a custom channel group that matches known assistant hosts and place it before channels those sources could otherwise match. In PostHog, inspect $referring_domain on web pageviews and group the same hosts. Report GA4 sessions and PostHog pageviews separately: they answer different questions and their totals should not be added together.

This workflow measures visits with a recognizable assistant source. It does not measure every mention or citation in an AI answer. It also cannot recover the origin of a visit whose referrer was stripped before either analytics tool saw it. GA4 channel mechanics and regex are covered by Google’s documentation and established guides; the value here is the two-system QA, the citation-versus-click distinction and a dated one-property example.

Question	Use	Metric to report
How many visits did GA4 attribute to AI assistants?	GA4 Traffic acquisition	Sessions by session source and landing page
Which assistant hosts appeared on actual pageviews?	PostHog event properties	Web $pageview events by $referring_domain
Did an AI answer cite our URL without a click?	A dated citation sample	Citation present/absent, platform and prompt
Did these visitors act?	A defined conversion event	Leads or other key events, with a stated attribution rule

1. Check the channel you already have

Open Reports → Acquisition → Traffic acquisition in GA4 and look for AI Assistant in the channel dimension. Our property shows this channel. In equal complete windows August 1–28 and August 29–September 25, 2026, it recorded 123 → 78 AI Assistant sessions. That is a property-specific channel comparison, not a count of all AI citations or an estimate of AI influence. Check your own property’s group definition before relying on the label.

Switch the report dimension to Session source / medium and add landing page where available. Record the property time zone, exact start and end dates, session count, engaged sessions and key events. Export the table before changing any channel definition. This gives you a baseline that remains interpretable if you later refine the grouping.

If AI Assistant is absent or you need a specific host list, use the official GA4 custom channel group instructions. Google explicitly includes an AI-assistant example and says custom groups can be used in acquisition reports. It also says channel order matters: traffic goes to the first matching channel.

2. Create a controlled GA4 AI group

In Admin → Data display → Channel groups, create a new group from the default group. Add a channel named “AI assistants”, set Source → matches regex, and start with a narrow list of known assistant hosts:

^(chatgpt\.com|chat\.openai\.com|(www\.)?perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com)$

Move the AI channel above Organic Search and Referral, save the group, then select that custom group in Traffic acquisition. This priority prevents a source such as gemini.google.com from being taken by a broader Google rule first. Keep the original default/primary group available for comparison. The host list is intentionally narrow and can undercount alternate hosts or stripped referrers; review unmatched sources monthly before adding another domain. Test candidate hosts such as chat.bing.com or grok.com against observed rows before adding them. Avoid matching a bare word such as ai or google, which could classify unrelated sources as assistant referrals.

Google states that custom reporting groups can apply retroactively to reports, while changing a property’s primary channel group affects its future primary-channel definition. Verify that behavior in your property after saving. A reporting group is enough for this analysis; there is no need to change the property’s primary group just to count referrals.

3. Cross-check real referrers in PostHog

Filter PostHog to web $pageview events on your own host. Break them down by $referring_domain and check the landing pathname alongside each assistant host. Compare the observed host strings with the GA4 source values before claiming the two definitions align. Keep an “unknown/missing referrer” row rather than silently assigning it to AI.

The validation table should be small enough to inspect:

Check	Expected result	If it fails
Known assistant visit	Assistant host in source/referrer	Check redirect chain and source naming
Direct typed visit	Direct or empty referrer	Do not label it an AI visit
Google Search visit	Google organic source	Ensure a broad regex did not capture google.com
Landing page	Same canonical URL family in both tools	Check trailing slashes, query strings and redirects
Conversion	Defined event tied to the reporting window	Check event firing before claiming a channel lift

PostHog event counts are not GA4 sessions. Different blockers, collection rules, time zones, identity rules and event definitions can all move the totals. Compare the direction of a trend only after aligning the windows and metrics; do not report a percentage discrepancy as “missing AI traffic” without investigating it.

4. Separate clicks, citations and business outcomes

Use a second table for citations: prompt, platform, market, date, cited URL, mention type and screenshot or answer URL. The AI citation tracking template defines its fields. A citation can produce zero measured visits; a measured visit may come from a link that no longer appears in a later answer. This is why a fall in a visibility tool’s response count cannot be explained by referral sessions alone. Our AI referral benchmark gives a dated historical example, while the 2026 AI search evidence review documents broader click-behavior research.

For a monthly stakeholder report, keep sessions, citation coverage and qualified actions on separate rows with stated windows; the five-metric report shows the full format. If visits and citations move in opposite directions, investigate the prompts and pages before rewriting content. For a Perplexity-specific example, see referrals versus citations; for Google-organic reconciliation, use the GSC and GA4 worksheet.

What this method cannot prove

• A direct visit is not proof that an AI assistant sent it.
• An Ahrefs “AI response” is not a GA4 session or a lead.
• A platform citation count is sensitive to prompt set, market and collection date.
• A custom channel group changes classification; it does not create visits that were never measured.

The practical goal is a repeatable decision trail. If the channel declines, identify which assistant, which landing page and which outcome changed. Then decide whether the problem is content, discovery, conversion or measurement. Our first-party AI visibility case study shows why those distinctions matter for a small site.

If your team needs help validating attribution and the action events before a content decision, see our consulting scope. A clean source definition is more valuable than a larger but unverified AI-traffic number.

FAQ: AI referral measurement

Can GA4 count every visit influenced by an AI answer?

No. A GA4 source rule can classify sessions only from the acquisition information it receives. A person may see an answer and later search for the brand, type the domain, or open a copied link without its original referrer. Keep those sessions in their observed channels. An AI Assistant channel is a useful attributed-referral series, not a count of everyone influenced by an assistant. The GA4 traffic acquisition documentation explains the session acquisition dimensions used in reports.

Should we add GA4 and PostHog assistant counts together?

No. The two tools can record the same visit with different session boundaries, client-side event timing, filters or source labels. Compare directions after aligning dates and landing URLs; do not sum the two totals as though they were separate audiences. If they diverge, inspect the definition and a small set of landing pages before attributing a business change to the difference.

What if an assistant host is missing from our rules?

Keep the raw source or referrer dimension available and review new hosts periodically. Add a specific rule only when the host belongs to an assistant and the classification is useful. Google’s custom channel group guidance says rule order determines which channel gets the first matching session. Document rule changes beside your trend chart so a classification edit is not mistaken for traffic growth.

Last verified: September 2026. Source definitions and product interfaces can change; recheck the linked documentation before updating a production GA4 group.