Generative engine optimization · Measurement
Your AI traffic is hiding in GA4's "Direct" bucket — here's how to find it
You added your first Quick Check link, you started showing up in ChatGPT answers, and then you went hunting through Google Analytics for proof it's actually sending anyone your way. What you find instead is a heap of sessions labeled "Direct" that ballooned right around the same time. No source, no campaign, no way to point at any single one and say that came from an AI answer.
That's not a tracking screwup on your end. It's how the traffic actually arrives.
Why AI traffic ducks Google Analytics in the first place
When someone clicks a link inside a Google search result, the browser hands over a referrer, a little tag that tells your analytics "this person came from Google." Most AI apps don't bother. ChatGPT's iOS app opens links inside its own browser view, which strips the referrer entirely. Someone reading an answer on their phone and copying a link into a fresh tab loses it too. No referrer, no source. Google Analytics has one bucket for sessions it can't account for, and that bucket is Direct.
Google did try to patch part of this. In May 2026, GA4 rolled out a native AI Assistant channel that automatically recognizes traffic from ChatGPT, Gemini, Deepseek, Copilot, and Grok, no setup needed on your end. It just starts showing up in your channel reports. Claude was in the announcement but got quietly yanked from the official list soon after, so if you're pulling Claude traffic, don't expect it to sort itself.
Perplexity is arguably the AI tool most likely to send someone straight to your Quick Check page with real buying intent, and it isn't in the native channel at all. It still lands in Referral, not AI Assistant, not Direct. You have to go hunting for it specifically.
So Direct isn't the whole picture, but it's most of it
Even with the new channel live, industry estimates put somewhere around 60 to 70 percent of real AI-driven sessions still landing in Direct, mostly from mobile app browsers and copy-pasted links that never carried a referrer to begin with. The native channel catches what it can see. It was never built to catch everything else.
Which means the honest read on your GA4 right now is: whatever the AI Assistant channel shows you is a floor, not the real number. Your actual AI traffic is higher than that report says. You just can't point at exactly which Direct sessions belong to it.
Why this is worth untangling, not just accepting
The traffic that does get correctly labeled behaves differently from your average visitor. Industry benchmarks put AI-referred visitors converting at roughly 4.4 times the rate of standard organic search. Conductor's 2026 numbers are the most quoted version of that figure, though the underlying methodology isn't published. Nobody's put out a rigorous study behind that number yet, so treat it as directional, not gospel. But it lines up with something that checks out: someone who read a full answer about your business before clicking is already most of the way to a decision. They're not comparison-shopping ten tabs. They came in warm.
If a meaningful slice of your best-converting visitors is sitting unlabeled in Direct, you're making decisions about what's working without seeing where your actual wins are coming from.
What to actually do about it this week
You can't fully unmask Direct traffic, and anyone promising a clean fix is overselling it. But you can close most of the visible gap.
First, check whether your GA4 property has the native AI Assistant channel showing data yet. Go to Reports, Acquisition, User acquisition, and look for "AI Assistant" as a Session default channel group. If you don't see it, it can take a property a while to pick it up after rollout. It's not something you toggle on.
Second, build a custom channel group for Perplexity specifically, since Google isn't going to. It's a source/medium rule matching referrer contains perplexity.ai. It takes about five minutes in GA4 admin, and it moves that traffic out of generic Referral into something you can actually track over time.
Third, for anywhere you control the link, your Quick Check page shared in a guide, a mention you've placed somewhere citable, add a simple UTM tag. It won't help traffic that arrives through an AI app's stripped browser, but it recovers everything that comes through a normal link click, which is more than you'd think.
Fourth, treat Direct sessions landing on a high-intent page, your pricing, your Quick Check, a specific service page, as a signal worth watching even without a clean label. It's not proof, but a spike in unlabeled Direct traffic to your Quick Check the same week you got mentioned somewhere is not a coincidence you should ignore.
None of this gets you a perfect number. What it gets you is a real floor instead of a guess, and a way to watch that floor move as you get named more places. That's enough to know whether the work is doing anything. That is the actual question you were trying to answer when you went looking for that traffic in the first place.