# Why your AI traffic is mostly invisible
Before you measure AI traffic, know what you are measuring: the visible sliver of a mostly invisible channel. When ChatGPT or Perplexity answers a question using your business, three things can happen. The customer acts on the answer without ever visiting your site. They visit, but the click arrives without a referrer - the note a browser passes along about where a visitor came from - and GA4 files it under Direct. Or they visit with the referrer intact, and you finally get a session you can attribute. Only the third case is measurable, and it is the smallest of the three.
The scale of the first case is the striking part. Per DataReportal's Digital 2026 Australia data, around 65% of Australian searches already end without a click on any website - and AI assistants are built to push that further, resolving the whole question in the answer box. The second case, the referrer-less click, comes from AI mobile apps, in-app browsers and copied-and-pasted links; those sessions sit in your Direct bucket right now, indistinguishable from someone typing your address.
So the honest frame for this post: the regex below will not show you your AI demand. It will show you the floor of it - the clicked-through fraction - accurately and consistently. That floor is worth measuring because of what it converts at, which we will get to. And measuring it properly is step one of the wider visibility work covered in the GEO guide for Australian businesses.
# What GA4's native AI Assistant channel does and does not cover
Since 13 May 2026, GA4 has shipped with an AI Assistant channel in the Default Channel Group - no setup required, and no control either. Google's release notes describe it plainly: traffic referred from "chatbots like ChatGPT, Gemini, and Claude" gets an ai-assistant medium value and groups under the new channel in your acquisition reports. If you have done nothing, you already have this.
The catch is the word "like". As Search Engine Journal's coverage notes, Google has not published the full list of recognised AI referrers, has not said whether Perplexity or Copilot are on it, and has not said how the list will be updated as new assistants launch. You cannot audit a list you cannot see. And the channel inherits the referrer problem from the last section: an AI visit that arrives without a referrer header still lands in Direct, native channel or not.
Sources: Google Analytics release notes, 13 May 2026; GA4 custom channel groups documentation.
Keep the native channel - it costs nothing and catches what it catches. But run your own custom channel group alongside it, because a measurement you cannot audit is a measurement you cannot trust, and the custom version has one further advantage the native channel cannot match: it applies to your historical data the moment you save it.
# The regex, copyable, with every current AI referrer
This is the pattern. Copy it as-is, or trim domains you do not care about. A regex is just a matching pattern: this one says "the session source contains any of these domains". Every domain on the list is a live AI assistant that passes a referrer when a user clicks through to your site.
What each piece is: chatgpt.com and the legacy chat.openai.com are ChatGPT. perplexity.ai is Perplexity. copilot.microsoft.com and the older edgeservices.bing.com are Microsoft Copilot. gemini.google.com and the legacy bard.google.com are Google's assistant. claude.ai is Claude, meta.ai is Meta AI, chat.deepseek.com is DeepSeek, chat.mistral.ai is Mistral's Le Chat, grok.com is Grok, and you.com and poe.com are smaller assistants that still send occasional clicks.
Google's own channel group documentation ships a sample AI regex, and it is worth reading as a cautionary tale: it includes patterns as loose as .*gpt.*, which would classify any source containing those three letters as AI traffic. The pattern above matches whole domains instead, so a newsletter called "gptweekly" does not pollute your channel. The docs' one piece of advice worth keeping verbatim: update your expression when the list of assistants you want to measure changes. Put a quarterly reminder on it - a new assistant is invisible to you until its domain is in the pattern.
# Setting it up as a custom channel group, step by step
Five minutes, no code on your website, and it works backwards through your existing data. Per Google's documentation, standard GA4 properties can hold two custom channel groups alongside the default one, and custom groups "can be applied to your reports retroactively" - meaning the history you could not see is visible the moment you finish this list.
Step 1: In GA4, open Admin, and under Data display choose Channel groups. Step 2: Click Create new channel group - GA4 starts you with a copy of the default group; name it something you will recognise, like "Channels + AI assistants". Step 3: Click Add new channel, name it "AI Assistants", and set its condition to Source → matches regex → paste the pattern from the section above. Step 4: Drag your new channel above Organic Search and Referral in the channel order. Order decides ties: if your AI channel sits below Referral, Referral claims the session first and your channel stays empty. Step 5: Save, then open any acquisition report and switch the primary dimension to your new channel group.
One honest note on scope: this classifies sessions, which is the right tool for the job. You may see guides suggesting looser source filters or referral exclusions instead - they solve different problems, and the channel group is the one that shows AI traffic as its own line next to Organic Search and Paid in every report you already use.
# What the numbers will look like (small, and that is correct)
Set your expectations before you open the report: the AI line will be tiny, and a tiny line is not a broken setup. The best public benchmark is Seer Interactive's seven-month study of a large site's GA4 data: AI assistants combined - ChatGPT, Perplexity, Gemini and Claude - sent about 11,000 sessions against 14 million from Google organic. That is 0.07%: for every ten thousand organic visits, seven from AI.
Within that sliver, concentration: ChatGPT sent 61% of the AI referrals and Perplexity 24%, with everything else sharing the remainder. Expect the same shape on an Australian small-business site, at smaller absolute numbers - single-digit sessions per week is a perfectly normal starting point. If your AI channel reads zero after a month, check the channel ordering from the setup section before concluding you are invisible; a channel below Referral in the order reads zero forever.
The reason to measure a channel this small is direction, not size. AI referral traffic is the fastest-shifting acquisition line in most analytics accounts right now, and the businesses that can see it move week by week are the ones that will notice - early - when it stops being small. You cannot manage the trend you never separated from Direct.
# The conversion-rate argument for why small still matters
The AI visitor arrives late in their decision, and it shows in the only number that pays invoices. In the same Seer Interactive benchmark, sessions referred by ChatGPT converted at 15.9% and Perplexity at 10.5%, against 1.76% for Google organic - roughly nine times and six times the organic rate.
Source: Seer Interactive - how traffic from ChatGPT converts.
The mechanism is intuitive once you picture the session that produced the click. The customer asked the assistant a question, compared the options inside the conversation, asked their follow-ups, and clicked through to exactly one or two sites at the end of it. The comparison work that normally happens across eight browser tabs happened before your analytics ever saw them. What lands on your site is closer to a referred customer than a browsing one - which is also why the pages those clicks land on should answer like a knowledgeable staff member, not a brochure. That is the citability work covered across the schema-for-AI-search post and the wider cluster.
For an Australian service business, the practical read: ten AI referrals a month at those rates is one or two genuine enquiries - from a channel most of your competitors have not noticed yet. Small is not the same as negligible when the rate is nine times organic.
# What you still cannot measure, honestly
The regex gives you clicked sessions. It cannot give you the three things most owners actually want to know. First: how often the assistants mention your business in answers nobody clicked. That is the majority of AI exposure, it happens entirely off your site, and no analytics tool sees it - the only way to observe it is to ask the assistants the questions your customers ask and record what comes back. Second: referrer-less clicks, which stay in Direct forever; treat every AI number in GA4 as a floor. Third: Google's AI Overviews, which are part of the search results page - a visitor who clicks through from one arrives as ordinary Organic Search, blended in with the classic blue links, per Google's own documentation on AI features in Search.
That first gap matters most, because the measured picture and the real picture can point in opposite directions. Ahrefs' research (covered in the ranking-versus-recommendation post) found only 12% of AI-cited URLs rank in Google's top 10 - being visible to Google and being cited by assistants are different contests, and your GA4 referral line only scores the moments the second contest ends in a click.
So run both halves. The regex from this post is the traffic half: five minutes, free, retroactive - do it today. The visibility half is checking what the assistants actually say about you: our AI Search Scorecard is the DIY version of that check, and the visibility audit is the done-for-you version that scores profile, website and AI surfaces together and hands back the fix list in priority order.
Read the full GEO guide for Australian businesses ›
One playbook like this per fortnight.
New pillars and tactical spokes from CalnetCorp's audit work with Australian service businesses.
Frequently asked questions
The regex shows the clicks. The audit shows the mentions.
GA4 can only count the AI visits that happen. The visibility audit checks what ChatGPT, Perplexity and Google's AI actually say about your business when customers ask - and hands back the fix list in priority order.