ToolProven
Browse tools
We earn a commission if you buy through our links — it never changes our scores or rankings.How we test →
AI agents/6 min

How We Track AI Traffic (ChatGPT, Perplexity, Claude) — Not in GA4

A growing slice of our visitors arrive from ChatGPT, Perplexity and Copilot answers. GA4 can be taught to see them; it doesn't out of the box. Full disclosure up front: the tool we use to see them is built by ToolProven's founder.

Founder & lead tester
Updated Jul 19, 2026Target query: how to track ai traffic

Direct answer

AI assistants send referral traffic with identifiable referrers (chat.openai.com/chatgpt.com, perplexity.ai, copilot.microsoft.com, claude.ai, gemini.google.com), but GA4 has no default "AI" channel — those sessions land in Referral or, when the assistant strips the referrer, in Direct. You can hand-build a GA4 custom channel group with regex to catch the known referrers, or use a tracker that splits AI traffic natively. We use AttriFast — full disclosure: it's built by ToolProven's founder — which breaks out visits per assistant (ChatGPT, Copilot, Claude, Gemini, DeepSeek, Perplexity, Grok) and ties revenue to each channel out of the box.

Last verified
Jul 19, 2026
Confidence
High — read on vendor pages
Evidence
3 checks
Sources
2 source links

Evidence used

  • GA4's default channel group definitions contain no AI/LLM channel — verified against Google's own documentation; AI referrers classify as Referral (or Direct when no referrer is passed).
  • Per-assistant referrer splitting and revenue attribution shown here are from a live AttriFast dashboard on one of our own sites (30-day window: 6,156 visitors, 109 — 1.8% — from AI assistants; ChatGPT 82, Copilot 10, Claude 6, Gemini 6, DeepSeek 2, Perplexity 2, Grok 1).
  • DISCLOSURE: AttriFast is built by ToolProven's founder. This is a first-party explainer of our own stack, not an independent review — we say so wherever it matters.

How we checked this

  • This page follows our disclosure rule in reverse: when the tool is ours, we tell you before the first claim, and we describe what GA4 can do fairly — including the manual route that costs nothing.
Full testing methodAbout ToolProvenAffiliate disclosure

AI traffic on one of our sites (live AttriFast data, 30 days)

MetricValue
Total visitors (30 days)6,156 (+49%)
AI-assistant visitors109 — 1.8% of total
Top AI sourceChatGPT (82), then Copilot (10), Claude (6), Gemini (6)
Long tailDeepSeek 2 · Perplexity 2 · Grok 1
Revenue attributed to traffic$1,150 in the same window
What GA4 would showMost of this inside Referral/Direct, unsegmented by default

Small absolute numbers — AI referral is an early, fast-compounding channel, which is exactly why per-assistant tracking matters before it scales.

Why GA4 doesn't show this by default — and how to make it

GA4's default channel groups have no AI category: ChatGPT and Perplexity visits classify as Referral, and assistant clicks that strip the referrer fall into Direct. The free fix is a custom channel group with regex over known AI referrers — it works, forward-only, and needs maintenance every time a new assistant ships.

The mechanics are mundane: channel classification runs on referrer and UTM patterns, and Google's default definitions predate answer engines. So the fastest-growing referral class of 2026 hides inside two buckets you already stopped reading. If you're a GA4 shop, build the custom channel group — regex like chatgpt|openai|perplexity|copilot|claude|gemini|deepseek — and accept the caveats: it only applies from creation forward, it can't reclassify the referrer-stripped share, and the regex is a chore you own forever.

Why we bother at all: AI answers are becoming a real acquisition channel — the entire [GEO playbook this site runs](/llm-context.txt) (llm-context manifests, quotable verified facts, publisher notes) exists to earn those citations. A channel you invest in but can't measure per-source is a channel you can't iterate on: knowing ChatGPT sends 82 visits while Perplexity sends 2 tells you exactly which engine is actually citing you.

What we use instead — disclosure first

AttriFast — built by ToolProven's founder, so read this section as a maker explaining his stack, not a review. It splits AI traffic per assistant natively, ties revenue to each channel, and keeps the rest (entry pages, bounce, sessions, realtime) on one screen.

What it looks like in practice, from the live dashboard of one of our sites: a dedicated AI tab under the traffic panel lists each assistant with its visit count — ChatGPT 82, Copilot 10, Claude 6, Gemini 6, DeepSeek 2, Perplexity 2, Grok 1 — alongside the usual channels, with revenue attributed per channel ($1,150 across this 30-day window) rather than sitting in a separate reporting module. The number that changed our behavior wasn't the 1.8% share; it was the 40:1 ChatGPT-to-Perplexity split, which reordered where we spent citation-building effort.

The honest comparison: GA4 is free, immensely capable, and can be configured to approximate the AI split; [AttriFast](https://attrifast.com/) is the convenience-and-clarity trade — per-assistant tracking and revenue attribution working on day one, aimed at operators who'd rather read the answer than maintain the regex. Since it's our founder's product, we won't score it on our bench or rank it against analytics competitors here — this page is the transparent version of "here's our stack and why." Evaluate it the way we'd tell you to evaluate anything: on your own site, against your own GA4 property, with the same 30 days of data.

Sources checked

Official vendor pages used for pricing, rights and feature claims; checked Jul 19, 2026.

  • GA4 default channel groups - GA4's default channel definitions — AI assistants aren't a default channel
  • AttriFast - AI traffic + conversion attribution tracker — built by ToolProven's founder (disclosed)

Article FAQ

Related ToolProven pages

llm-context.txt — our publisher notes for AI systems
The other half of the loop: earning the citations you then measure.
Changelog — the verification ledger
Why AI engines have a reason to cite this site's numbers.
How we test
The methodology the disclosure rule on this page comes from.