Your 'Direct' traffic grew and you don't know why. A good chunk is generative AI traffic GA4 can't label — it exists, it converts, and it's hiding in plain sight.
In the Searchability framework we work to make AI discover and cite you. That was Pillar 1: presence. Now Pillar 2 begins — measuring the traffic that presence generates. If you already track your presence in model answers and checked how ready your site is with the AI Visibility Score, the next step is seeing the visits that arrive — and that's where Google Analytics 4 (GA4) fails you.
This post defines the problem. It isn't a tutorial — configuring the AI channel and the regex in GA4 comes in a later post of this pillar.
What 'Direct' Is in GA4 — and What It Isn't
'Direct' sounds like people typing your URL. That's the intuitive reading, and it's the one that misleads you. According to Google, "Direct is the channel by which users arrive at your site/app via a saved link or by entering your URL."
But in practice the channel is broader than that. GA4 marks a visit as Direct when the source is exactly (direct) and the medium is (not set) or (none) — that is, when no origin signal reached it. It isn't only user intent: it's the absence of data.
Google says it plainly. The (direct) / (none) combination represents, according to Google, "website traffic that doesn't have a clear referral source." Among the causes it lists are parameters lost in redirects, link shorteners, and missing referral information. Direct is the "I don't know where it came from" bucket.
Subscribe to the Madbotz newsletter to get the next analysis straight to your inbox. No spam, no noise — just new posts.
Why Generative AI Lands There — Referrer Stripping
When someone clicks a link, the browser usually sends the referrer: the signal that says which page the visit came from. According to MDN, the HTTP Referer header "contains the absolute or partial address from which a resource has been requested," and each site's referrer policy defines how much is sent.
GA4 reads that header to decide the channel. If chatgpt.com arrives, it groups it; if nothing arrives, it sends the visit to Direct. The whole attribution model depends on a signal AI often doesn't send.
This is called referrer stripping. Chat apps, in-app browsers, and copy-and-paste open your link without attaching where it came from. According to Search Engine Journal, AI assistant traffic "that arrives without a referrer header still lands in Direct" — and that happens with in-app browsers, mobile apps, and when the user copies and pastes the link. The model doesn't fail: it just doesn't get the signal.
Not Everything Disappears — How Each AI Source Appears
Referrer stripping isn't uniform. Some AI sources do leave a trail and others erase it, so it's worth knowing which is which before drawing conclusions from the report.
A 2026 development helps here. According to Search Engine Journal, as of May 2026 GA4 groups traffic from recognized assistants — it names ChatGPT, Gemini, and Claude — into a native "AI Assistant" channel, with no manual setup, but only when a valid referrer arrives. Whatever arrives without a referrer stays in Direct.
AI Overviews are the other special case. According to Google, Organic Search includes "Google's AI Overviews and AI Mode" — because the click comes from google.com and GA4 treats it as one more organic result. It doesn't hide in Direct, but it doesn't isolate itself either.
Table 1 — how GA4 records each AI traffic source. The last column, "What to do about it," is the recommended action for each case.
| AI source | Passes referrer? | How it appears in GA4 | What to do about it |
|---|---|---|---|
| Chat assistants in app and mobile (ChatGPT, Claude, Gemini) | Usually no | Direct / (not set) | Review Direct spikes and prepare server-side measurement |
| Chat assistants on desktop web | Usually yes | "AI Assistant" channel (native since May 2026) | Confirm they land in the native AI channel |
| AI browsers (ChatGPT Atlas, Perplexity Comet) | Varies — Atlas tends to strip it, Comet tends to pass it | Direct (Atlas) or Referral perplexity.ai (Comet) | Server-side measurement for the referrer-less cases |
| Perplexity (web) | Usually yes | Referral (perplexity.ai) | Group it with your AI channel |
| Google AI Overviews and AI Mode | Yes — it's google.com | Organic Search | Segment inside Organic, knowing it won't fully isolate |
The pattern is clear: the more a source resembles a closed app, the more likely you end up blind. Perplexity on the web and AI Overviews are the friendly exception; assistants in their apps and some AI browsers are the problem.
The 2026 Aggravator — AI Browsers
There's a new layer that makes the picture worse. AI browsers like ChatGPT Atlas and Perplexity Comet open links in ways that strip the trail even further.
According to MarTech, ChatGPT Atlas opens shared links in an internal environment that strips the referrer header, so GA4 records the session as Direct or (not set); Comet, by contrast, tends to pass perplexity.ai. The behavior varies from one browser to another and isn't stable.
We won't go into the detail here — the AI-browser deep dive is a later post in this pillar. For now it's enough to know the trend worsens the problem, it doesn't ease it.
What This Costs the Business
The point isn't technical, it's financial. If your AI channel hides in Direct, you're undercounting exactly the source most worth measuring.
AI traffic doesn't behave like the rest. According to Similarweb, generative AI referrals to transactional sites grew 357% year over year, and those visitors "convert at around 7%, higher than many traditional channels." They arrive further along in their decision, because the model already did the upfront research.
Now put the two pieces together. A large share of that high-value traffic arrives without a referrer and lands in Direct, where it blends with bookmarks and typed visits. The result is a blind spot: your highest-converting channel, reported as noise.
Where to Start Recovering Attribution
You don't need to solve all of it today. This pillar will hand you each piece; for now, three high-level moves to stop flying blind.
First, audit your Direct traffic for signals. If you see Direct spikes that line up with your long-form pages, your FAQs, or your most citable content, a good part is likely AI — the same content a model prefers to cite.
Second, get ready to build an AI channel in GA4. Google already documents custom channel groups with referrer-based rules, and the new native "AI Assistant" channel covers part of the path; the full regex and the rule order are the subject of the next post in this pillar.
Third, consider server-side measurement for the referrer-less cases. When the signal never reaches the browser, capturing it on the server is the way to recover what referrer stripping erases — another later post in this pillar.
How Madbotz Measures Its Own Blog
Honesty before hype. Madbotz faces exactly this problem on its own blog: part of our AI traffic lands in Direct, just like yours.
That's why this pillar exists. AI attribution isn't guaranteed or recovered 100% — you work it layer by layer, starting by understanding where it's lost. Tracking helps recover attribution; it doesn't promise to return every session.
And the cause of the traffic is visibility. Before measuring what arrives, it's worth making sure AI sees and cites you — you measure that with the AI Visibility Score, the same 130+ check-item engine we run against our own site. Start there, then come back to the measurement layer.
Frequently Asked Questions
Why does my AI traffic show up as 'Direct' in GA4?
Because many AI apps and assistants don't send a referrer when they open your link. Without that signal, GA4 can't tell where the visit came from and files it under Direct, the bucket for any traffic with no identifiable source. It isn't people typing your URL: it's that the origin data got lost on the way.
Do all AI sources hide in Direct?
No. Perplexity on the web usually passes perplexity.ai as a referral, and Google's AI Overviews arrive as organic search because the referrer is google.com. Chat assistants inside their apps and AI browsers like ChatGPT Atlas are the ones that strip the trail most often and end up in Direct.
Does GA4's new 'AI Assistant' channel fix the problem?
Only partly. As of May 2026, GA4 groups traffic from recognized assistants into a native "AI Assistant" channel, but only when a valid referrer arrives. Referrer-less traffic — apps, mobile, copy and paste — still falls into Direct. The native channel captures what GA4 can identify, not what was lost.
Why does recovering this attribution matter?
Because AI traffic tends to convert differently and grows fast. According to Similarweb, generative AI referrals to transactional sites grew 357% year over year and convert at around 7%, above many traditional channels. If that traffic hides in Direct, you can't measure your highest-value channel or report it to the business.
Wrap-Up
Three takeaways:
- Your 'Direct' in GA4 isn't only people typing your URL — it's the bucket for source-less visits, and a good chunk today is AI traffic.
- Not everything disappears: Perplexity web and AI Overviews leave a trail; chat apps and AI browsers are the ones that erase it.
- That traffic converts and grows — measuring it is the work of Pillar 2, layer by layer, starting with the visibility that creates it.