Why AI traffic hides inside Direct
Ask most SaaS teams how much traffic they get from ChatGPT and Perplexity, and the honest answer is "we don't really know." That's not a tooling failure on their part. It's a structural gap. An analysis of over 446,000 site visits found that roughly 70.6% of AI-driven traffic arrives with no referrer header at all, which means Google Analytics has no way to attribute it and simply files it under Direct (Loamly, via Seresa). Your actual AI referral number is very likely several times larger than whatever GA4 shows you today.
Four reasons the referrer gets stripped
- Referrer policy settings. Some AI platforms configure outbound links with a strict or no-referrer policy by default.
- Explicit noreferrer attributes. Google's own AI Mode links carry a noreferrer attribute, a deliberate choice that makes AI Mode traffic effectively untraceable to any client-side analytics tool.
- Mobile app in-app browsers. Traffic from the ChatGPT or Perplexity mobile apps often routes through an in-app browser that doesn't pass a normal referrer.
- Copy-paste. A user who reads an AI answer and manually types or pastes your URL generates no referrer at all, indistinguishable from someone who just remembered your domain.
None of this is unique to your setup. It's architectural, and it means every AI visibility conversation needs the caveat that whatever number you can see is a floor, not the real total.
What changed in GA4 in May 2026
There's real, recent progress worth knowing about. On May 13, 2026, Google added a native "AI Assistant" channel to the GA4 Default Channel Group, which automatically classifies sessions from recognized AI referrers, ChatGPT, Gemini, and Claude among them, based on the referrer header when one is present (Search Engine Journal). It requires no configuration. If GA4 receives a recognized referrer, the session is tagged and grouped automatically.
GA4 processes data forward only. The new channel does not retroactively reclassify historical sessions, and it still can't recover the 70%-plus of AI traffic arriving with no referrer at all. It's a real improvement, not a complete fix.
Recovering signal in older properties and dark traffic
Since the automatic channel only catches what has a referrer, a few manual steps close some of the remaining gap:
- Custom channel grouping. Build a channel definition that matches known AI referrer domains and session source strings so historical data going forward is captured consistently, not just whatever the default catches.
- UTM parameters where you control the link. Anywhere you can influence how a link is shared, llms.txt aside, structured data, brand mentions you control, tag it. It won't cover organic AI citations, but it removes ambiguity where you have the choice.
- Watch Direct traffic behavior, not just volume. A spike in Direct sessions with unusually high engagement or a landing page pattern that matches your comparison and pricing content is a reasonable proxy signal for AI-referred dark traffic, even without a clean attribution.
The Search Console side
Google Analytics only sees traffic that reaches your site. Search Console sees something upstream of that: how your pages perform inside Google's own AI features before a click even happens, through the Generative AI performance report. I go through what that report actually shows and how AI Overviews eligibility works in the companion guide on appearing in AI Overviews and AI Mode.
Why the effort is worth it
The dataset behind the 70.6% figure also measured what identifiable AI-referred traffic actually does once it arrives: a meaningfully higher transactional conversion rate than non-AI traffic, roughly a four-times premium in that data (Seresa). That tracks with what you'd expect: someone arriving after an AI system already answered their comparison questions is further along than someone arriving cold from a generic search. Undercounting this channel doesn't just hide a vanity number. It hides your highest-intent traffic.
If AI-referred visitors convert at multiples of your average traffic and most of them are currently invisible in your reporting, the real ROI case for AI visibility work is probably larger than whatever number you're currently using to justify it.
The setup checklist
- Confirm your GA4 property has the AI Assistant default channel active (rolled out through June 2026, check your Default Channel Group report).
- Build a custom channel grouping to catch AI referrer domains the default group might miss.
- Cross-check Direct traffic for engagement and landing page patterns consistent with AI-referred dark traffic.
- Set up the Generative AI performance report in Search Console to see the upstream picture.
- Treat any AI referral number you get as a floor, and revisit it as GA4's own coverage improves.
A proper audit sets this measurement up correctly from the start, so you're not guessing. That is what I do.
Why doesn't AI referral traffic show up properly in GA4?
Does GA4 track ChatGPT traffic automatically now?
Is AI referral traffic worth tracking closely?
Primary sources: Loamly, 446,405-visit AI traffic analysis, via Seresa · Search Engine Journal, GA4 AI Assistant channel launch coverage. All sources accessed July 25, 2026.