Tracking AI Referral Traffic: GA4 Setup for LLMs

Why can't GA4 show your AI referral traffic out of the box?

GA4's channel taxonomy was designed before generative AI became a meaningful traffic source. Its default buckets — Organic Search, Paid Search, Direct, Referral and Social — have no native slot for an answer engine. So when someone clicks a citation inside ChatGPT or Perplexity, that session has nowhere obvious to land.

In practice it splits two ways. If the AI platform passes a referrer header, GA4 files the visit under Referral and buries it among dozens of other domains. If the header is stripped — common with desktop apps, in-app browsers and certain privacy settings — the session shows up as Direct, the same bucket as someone typing your URL by hand.

The result is a blind spot. Most teams genuinely cannot see how much traffic AI engines send them, so they under-invest in the work that earns those citations. You can't justify a GEO budget against a number you've never measured. Isolating the channel is the first step to proving it's worth the effort.

What is AI referral traffic and why does it matter now?

AI referral traffic is the stream of visitors who reach your site by clicking a link an AI assistant surfaced in its answer. Instead of scanning a page of blue links, the user reads a synthesized response, sees your brand cited as a source, and clicks through. It's a distinct intent profile from classic search.

The volume is still modest as a share of total traffic, but the trajectory is steep — AI-assisted sessions are growing well over 80% year over year on many sites. More importantly, the quality is unusually high. The user has often already been pre-qualified by the AI's answer, so they arrive warmer, more informed and closer to a decision.

That matters because the people you most want — researchers, buyers, founders comparing tools — increasingly start in an AI chat, not a search box. If you're publishing content but only measuring Google rankings, you're scoring half the game. Measuring AI referrals lets you connect GEO effort to actual sessions, leads and revenue.

Which AI domains should your GA4 setup track?

Accurate tracking starts with a clean, current list of source domains. Different engines expose themselves differently, and new entrants appear almost monthly, so treat this list as something you review quarterly rather than set once and forget.

At a minimum, your pattern should cover the heavy hitters: ChatGPT, Perplexity, Gemini, Claude and Copilot. Beyond those, a handful of fast-growing players — DeepSeek, Grok, Meta AI and You.com — are now producing measurable referrals where a year ago they sent almost none. Including them future-proofs your reporting.

The engines also behave differently in the wild. Perplexity is the most reliable, consistently passing its domain as a referrer on both desktop and mobile. ChatGPT only began appending its source tag to desktop citation links in mid-2025, so its measured numbers undercount reality. Claude and Gemini web interfaces pass headers inconsistently, which means some genuine AI visits will still slip into Direct no matter how clean your setup is.

How do you build a custom AI channel group in GA4?

The most reliable method is a custom channel group with a regex rule. Open Admin, then under Data Display choose Channel Groups. Copy the default group so you keep every existing channel, and give the copy a clear name like "Default and AI Engines". This preserves your normal reporting while adding the new view.

Next, click to add a new channel — call it "AI Engines" — and set a condition where Session source matches a regex. A workable pattern is: chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|bard\.google\.com|claude\.ai|copilot\.microsoft\.com|deepseek\.com|grok\.com|meta\.ai. Remember that GA4 regex is case-sensitive, so include any variants you actually see in your data.

Now the step people miss: ordering. GA4 evaluates channel rules top to bottom and stops at the first match. If Referral sits above your AI channel, those AI sessions fire on Referral first and never reach your new bucket. Drag the AI Engines channel above Referral. If your channel shows zero sessions while AI domains clearly appear in your referral report, bad ordering is almost always the cause.

Should you use GA4's native AI Assistant channel too?

In 2026 Google added a native AI Assistant channel to GA4's default channel group. When an incoming session carries a referrer matching a recognized AI domain, GA4 tags it with the medium "ai-assistant" and files it automatically. To find it, go to Reports, then Acquisition, then Traffic acquisition, and set the dimension to Session default channel group.

It's a useful baseline, but it has real gaps. Google's recognized list is narrow — it captures the biggest names but misses several engines, so Perplexity, Copilot and newer tools can still end up in Referral. It also processes data forward only, with no backfill, so it counts sessions from the day it switched on and won't reconstruct the months before.

That's why the best practice is to run both. Let the native channel give you a quick, low-maintenance signal, and let your custom regex group catch the engines Google misses and keep pace with new entrants. Together they give you the fullest picture GA4 can offer, with the custom group doing the heavy lifting.

What can't you track, and how do you handle the gaps?

Even a flawless setup leaves blind spots, and it's healthier to name them than pretend they don't exist. Mobile AI app traffic frequently strips the referrer, so those visits land in Direct. Copy-and-paste link sharing does the same. And zero-click answers — where the AI cites you but the user never clicks — leave no session at all, yet still build awareness and brand recall.

There's also a hard limit on context. GA4 can tell you a session came from perplexity.ai, but it can't tell you what the user asked, which of your pages the AI quoted, or what it said before the click. The referrer carries the domain, not the prompt. For that intelligence you'll lean on AI-visibility tools that test prompts directly rather than on analytics alone.

Treat measured AI referrals as a confident floor, not a ceiling. A practical move is to watch your Direct channel for unusual lifts that track alongside content you know is getting cited. It won't be precise, but combined with your custom channel it gives you a more honest sense of true AI influence.

How do you prove AI traffic is actually worth it?

Once the channel is isolated, judge it on outcomes, not raw volume. Build a comparison in GA4's Explorations using your AI channel as a segment, and line it up against Organic Search on the metrics that matter: engagement rate, conversions and conversion rate. This is where the case for GEO usually makes itself.

The pattern that keeps appearing is striking. AI referral visitors tend to convert at several times the rate of traditional organic search traffic, because the engine has effectively pre-sold them before they arrive. On some sites ChatGPT and Perplexity sessions convert in the double digits while plain Google organic sits below two percent. Even a small number of these visits can outperform a much larger generic crowd.

That's the argument that unlocks investment. When you can show that a single-digit slice of traffic drives an outsized share of qualified leads, "is AI traffic worth it" stops being a debate. The next logical step is to publish more of the structured, citation-friendly content that earns those mentions — which is exactly where an organic-marketing autopilot like Artiql fits. If you'd like to see how that works end to end, book a demo.

Frequently asked questions