You can track AI referral traffic in GA4, and it will show you a minority of your actual AI visibility. The setup takes ten minutes and is worth doing. What matters more is knowing which traffic it can never capture, because the gap is large enough that acting on the visible number alone will mislead you.
This guide covers the configuration, then the honest limits, then the two sources that fill part of the gap.
What GA4 can see
Assistants that send a referrer show up as ordinary referral traffic, identifiable by hostname.
| Source | Hostname | Reliability |
|---|---|---|
| ChatGPT web | chatgpt.com |
Good — has passed a utm_source since mid-2025 |
| Perplexity | perplexity.ai |
Good |
| Copilot | copilot.microsoft.com |
Reasonable |
| Claude | claude.ai |
Inconsistent |
| Gemini | gemini.google.com |
Inconsistent |
| Meta AI | meta.ai |
Inconsistent |
These are web interfaces only. The same assistant used through its mobile app usually sends nothing.
The GA4 setup
Five steps. Do them in this order — the channel group is worthless until the data is separated.
Step 1 — Check what is already there
Reports → Acquisition → Traffic acquisition, set the dimension to Session source / medium, and search for chatgpt, perplexity, claude, copilot and gemini.
Most sites find some traffic already recorded. This tells you the baseline before you configure anything.
Step 2 — Create a custom channel group
Admin → Data display → Channel groups → Create new channel group. Name it so it is obvious later — "AI Assistants" rather than "AI".
Step 3 — Add the rule
Add a channel called AI Assistants, with the condition Source matches regex:
chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|meta\.ai|you\.com|phind\.com
Drag it above Organic Search and Referral. Channel groups evaluate in order, so a rule sitting below Referral never fires.
Step 4 — Create an exploration you will actually open
Explore → Blank, with Session source as the dimension and sessions, engaged sessions and conversions as metrics, filtered to your new channel. Save it.
A channel group nobody looks at is configuration rather than measurement.
Step 5 — Add Search Console for the Google side
GA4 will never separate AI Overview clicks. Search Console will not give you those clicks either — but since 31 August 2026 it does break out impressions for AI Overviews, AI Mode and generative AI in Discover, split from ordinary Search.
That is the only first-party data on Google's AI surfaces that exists. Check it alongside GA4 rather than instead of it, and see the section below for what it does and does not contain.
Two cautions on the setup. Custom channel groups apply from creation and do not fully backfill, so your history stays in the old grouping. And keep the regex in one documented place, because this list grows and an undocumented regex becomes unmaintainable within a year.
What you will never see
This is the section that matters more than the configuration, and it is missing from almost every guide on this query.
Mobile assistant apps. The ChatGPT and Copilot apps strip the referrer. Those visits land in Direct, mixed with bookmarks and typed URLs. On a consumer audience this is likely the single largest category of AI-driven traffic, and it is unattributable.
Google AI Mode. It applies noreferrer deliberately. No client-side analytics tool can see it, and no configuration changes that.
AI Overviews clicks. They arrive as ordinary Google organic and cannot be separated from any other organic click, in GA4 or anywhere else. Search Console now shows you the impressions — it does not show you which of your organic clicks came from an Overview.
Zero-click citations. The majority case. Your brand is named, the reader gets their answer, nobody visits. There is no event to record, so this generates no data anywhere.
Voice assistants. No click, no session.
The consequence to internalise: your AI referral number is a sample of unknown and variable size, not a measurement. Treating a rise or fall in it as a trend in AI visibility is reading noise, particularly if your audience skews mobile.
The two sources that fill part of the gap
Search Console, for Google's surfaces
This changed recently and most guidance has not caught up. Google began breaking out AI performance in the UK on 3 June 2026 and completed the global rollout on 31 August 2026.
What you get: impressions for AI Overviews, AI Mode and generative AI in Discover, separated from ordinary Search, filterable by page, country, device and date.
What you do not get:
- No clicks. Not for any AI surface.
- No click-through rate, because there are no clicks to divide by.
- No query data at any level.
- No per-link position — every link inside an Overview shares one position slot.
So the honest description is a visibility report, not a traffic report. You can now see that you appeared in an AI answer. You cannot see whether it sent you anyone.
The comparison worth watching: AI impressions rising while your overall Search clicks for the same pages stay flat. That is the signature of being cited without being visited, and until August 2026 there was no way to observe it at all.
Note also that this is a breakout of existing data rather than new data — those impressions were always inside your Search Console totals. Your historical totals did not change; they were just never itemised.
Server logs, for retrieval crawlers
Bots do not run JavaScript, so nothing in GA4 will ever show you a crawler. Your access logs will.
Grep for OAI-SearchBot, PerplexityBot, ClaudeBot, Claude-SearchBot, ChatGPT-User, Perplexity-User and GPTBot. What you learn:
- Whether assistants can reach you at all — a robots.txt mistake shows up here immediately
- Which pages they fetch repeatedly, which is a proxy for what they find useful
- Whether user-triggered fetches are happening, meaning real people are asking assistants about your pages
The crawler taxonomy and what each bot means is in llms.txt and AI crawlers.
This is the most under-used data source in the whole category, and it is sitting on your server already.
Reading it without fooling yourself
Four rules.
Compare like with like. AI referral volume is small on most sites. A jump from 40 to 80 sessions is a doubling and is also 40 sessions. Look at direction over months, not week to week.
Watch engagement, not just volume. AI referrals frequently arrive better informed — the assistant has already filtered — so engagement rate and conversion rate matter more than session count. A small, high-converting channel is worth more than the volume suggests.
Treat Direct as partly AI. If Direct rises without a campaign to explain it, some of that is mobile assistant traffic. You cannot prove it and you should not pretend the category does not exist.
Do not buy a visibility score to fix this. Tools in this category prompt assistants repeatedly and count how often you appear. That is a survey of the tool's own prompts, not a measurement of your traffic, and it answers a different question from the one you started with. It can be genuinely useful for tracking brand mentions — just do not confuse it with attribution.
Frequently asked questions
Can you track ChatGPT traffic in Google Analytics?
Partly. ChatGPT's web interface passes a referrer and has included a utm_source since mid-2025, so those sessions appear as referral traffic from chatgpt.com. The mobile app strips the referrer, so app-driven visits land in Direct and cannot be separated.
How do I see AI traffic in GA4?
Create a custom channel group with a source regex matching the assistant hostnames — chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com — and order it above Organic Search and Referral so it evaluates first. Channel groups apply from creation and do not fully backfill.
Why does AI traffic show as direct?
Because mobile assistant apps do not send a referrer header. The session arrives with no origin information and GA4 files it under Direct alongside bookmarks and typed URLs. No configuration recovers it.
Can I track AI Overviews traffic?
Partly, and only since 31 August 2026. Search Console now reports AI Overviews impressions separately from ordinary Search, filterable by page, country and device. It reports no clicks and no click-through rate for any AI surface, and clicks from Overviews still arrive as ordinary organic in GA4 — so you can see that you appeared, not whether anyone came.
Why is Google AI Mode traffic invisible?
Google applies a noreferrer attribute to those links deliberately, so the browser sends no referrer. Client-side analytics tools cannot see what the browser does not send. Search Console reports impressions and clicks for AI Mode, which is the only available window.
How do I know if AI assistants are crawling my site?
Check server access logs for user agents including OAI-SearchBot, PerplexityBot, ClaudeBot and ChatGPT-User. Bots do not execute JavaScript, so analytics will never show them. Logs also reveal immediately whether a robots.txt rule is blocking retrieval crawlers.
Are AI visibility tracking tools worth it?
They answer a different question. Most prompt assistants repeatedly and count how often your brand appears, which measures the tool's prompt set rather than your traffic. Useful for tracking brand mentions over time; not a substitute for attribution, and not a measurement of visits.
How much AI traffic should I expect?
There is no meaningful benchmark, because the visible portion depends on your audience's device mix and which assistants they use — both of which vary enormously. The useful comparison is your own trend over months, alongside Search Console impressions and crawler activity in your logs.
What to do next
Before configuring anything, search your existing GA4 traffic acquisition report for chatgpt and perplexity. Most sites already have data there and have never looked at it.
Then grep your server logs for OAI-SearchBot and PerplexityBot. That takes a minute and tells you something no analytics tool can.
Related guides
- SEO when AI answers the question first — what you are trying to measure
- llms.txt and AI crawlers: block them or not? — what the log entries mean
- Entity SEO: why named things beat keywords now — the work that changes the number
- Content formats that still get clicks from AI — what earns the click when there is one
- Search Console: the 5 reports that matter — the other half of the picture
Free: The 60-Minute Email Authentication Fix
A no-fluff checklist to set up SPF, DKIM & DMARC correctly and pass Gmail & Yahoo's sender requirements.
Muhammad Basim has worked in digital marketing since 2013, focused on email deliverability and AI-assisted content production. He is the author of The Email Deliverability Playbook and The Email Copywriting Playbook.
Related Articles
Email Deliverability: Why Authenticated Emails Still Land in Spam
Email deliverability is whether your message reaches the inbox rather than the spam folder, and it is decided by three factors: technical infrastructure, list quality, and sending behaviour. Authentication is one item inside the first factor. Getting it right is necessary, and it is nowhere near sufficient. That gap explains the most common complaint in […]
Linkable Assets: What Actually Earns Links
A link is a citation, and a citation requires that someone writing about your subject needed you to make their point. That is the whole mechanism, and it is the reason most "linkable content" earns nothing. The test, before you build anything: Could a writer covering this topic finish their sentence without referencing you? If […]
Link Outreach Emails That Get Replies
Outreach fails for exactly two reasons, and only one of them gets written about. The first is that the email was not worth replying to. That is the reason every guide addresses, and the advice — personalise, be brief, offer value — is correct and insufficient. The second is that the email never arrived. Nobody […]