Is Your GA4 Hiding 70% of Your AI Traffic?

Seventy percent of the visitors your AI tools are sending you? Your analytics platform has no idea they came from AI. Not 5%, not 15% — 70.6%. That's the proportion of sessions arriving at sites from AI assistants that carry no referrer header at all, according to an aggregated study of over 446,000 sessions flagged as AI-originated. GA4 sees them, classifies them as Direct, and everyone assumes the uptick in direct traffic is brand awareness improving. It usually isn't.

The headline figure comes from a study that cross-referenced server log UA-string matching against GA4 session data across a large multi-site sample. Crawler bot-mix data in this post draws from CDN edge telemetry published by major network providers for May 2026. AI referral platform breakdowns come from independent studies of browser-origin sessions totalling millions of visits.

The Referrer Gap Isn't Accidental

When a user browses to ChatGPT or Perplexity in a dedicated mobile app, asks a question, clicks your link, and arrives on your site — the browser often sends no referrer header. The app didn't load your page inside a webview that passes referrers; it handed off to the OS, which opened your page fresh. The result is a session that looks, to GA4, exactly like a direct visit. You'd need behavioural signals (deep landing pages, session length anomalies, zero-scroll bounce patterns) to separate these out — and even then it's probabilistic.

There's also the question of platform design. Some AI platforms have historically stripped referrers more aggressively than others. The gap isn't random noise; it's structurally baked into how these products work. The mobile/desktop split matters enormously: the same AI platform will produce very different attribution rates depending on whether users are on a web browser or a native app.

Crawlers and Referrals Are Two Different Animals

Here's something worth internalizing: the bots crawling your site and the traffic your site receives from AI platforms are almost entirely separate populations. ClaudeBot recently surged from 9% to 20% of global AI crawler traffic by volume. GPTBot holds around 10%. But neither of these crawl volumes tells you anything about whether the AI assistants are actually referring visitors to you.

The crawl is reconnaissance — the AI gathers and indexes your content. The referral happens much later, if ever, when a real user asks a question and your content surfaces in an answer. For every one referral visit, crawlers may hit the same URL thousands of times. Which means optimizing for crawler access and optimizing for referral traffic require completely different thinking, and conflating them leads to very wrong conclusions about how AI is affecting your site.

Did GA4's May 2026 Update Actually Fix This?

On 13 May 2026, Google added a native AI Assistant channel to GA4's default channel groups. Sessions from recognizable AI platforms now get a distinct medium instead of folding into Organic or Referral. It's a genuine improvement — for the first time, AI-attributed sessions are a first-class segment in GA4 without custom regex work.

But the referrer-stripping problem it doesn't fix is substantial. Estimates suggest the new channel captures somewhere between 60 to 80% of actual AI-sourced sessions — the rest still land as Direct. One frequently-cited analysis puts the hidden share at roughly 1 in 3 AI visits remaining uncounted even after the channel is in place. Mobile app contexts and certain embedded browser configurations are the main culprits. Unless those platforms change their referrer behaviour, no analytics configuration will fully close the gap. You're tracking the sessions that leave a trace, not all sessions.

Why Does It Matter Which Platform Sends Your Traffic?

Not all AI platforms leak referrers at the same rate. Perplexity, being primarily a web-first product used on desktop browsers, passes referrers more consistently than platforms with heavy mobile app usage. That means your Perplexity traffic is likely closer to correctly attributed in GA4 than traffic from other major AI assistants.

This matters for prioritisation. If you're trying to understand which AI platform is actually driving business value, raw referral counts in GA4 systematically undercount platforms with higher mobile share. An AI assistant sending you 200 GA4 sessions might realistically be sending 300 to 400. Acting on the GA4 numbers alone risks deprioritizing the platform that's actually most valuable for your audience.

What should site owners actually do?

The first practical step is to stop treating Direct traffic as opaque noise and start treating it as a signal. A spike in Direct sessions on deep, specific pages — product detail pages, pricing comparisons, technical docs — that doesn't correlate with brand search volume or email campaigns is worth investigating as possible AI-assisted traffic. Building a GA4 segment for Direct sessions landing on non-homepage, non-category pages is a reasonable starting point.

Second, if AI crawlers are visiting your site, instrument that layer separately. The gap between crawler visit volume and eventual referral traffic is enormous, but crawl activity isn't meaningless — it tells you which pages are being indexed and how often. If crawlers are hitting your homepage repeatedly but skipping your product documentation, that's an indexing problem that will eventually show up as a referral gap. Tracking crawl patterns is a leading indicator for referral trends, not a lagging one.

Third: the referrer stripping problem is structural and isn't getting fixed by analytics configuration changes alone. The only reliable way to understand what AI systems are doing with your content is to look upstream — at the crawl layer. Crawler access logs tell you when your content is being harvested; referral data tells you (partially, imperfectly) when it's being surfaced. You need both to understand the full pipeline, and right now most site owners only have one of them.