How to See AI Traffic in GA4 (ChatGPT, Perplexity and Gemini)
GA4 has no AI channel, so visits from ChatGPT, Perplexity and Gemini hide inside Referral and Direct. Here is how to build a report that shows them, and what the numbers really mean.
Someone on your team asks the obvious question: “Are we getting traffic from ChatGPT yet?”
You open GA4, look at Traffic acquisition, and find nothing that says AI anywhere. The visits are real. They are just scattered across Referral, Direct and Organic Search, where they quietly blend in with everything else.
GA4 has no built in AI channel. Nobody is going to add one for you. But you can build one in about ten minutes, and once it exists it works on your historical data too.
Here is how to do it, what the numbers actually mean, and the parts that no setup can fix.
Why AI traffic is invisible by default
GA4 sorts every session into a channel using its default channel group. That grouping was written for a web where people arrive from search engines, social networks and email. AI assistants are none of those things, so they fall through.
Three separate things happen:
Most assistants land in Referral. A link clicked inside ChatGPT or Perplexity passes a referrer like chatgpt.com or perplexity.ai. GA4 sees an unfamiliar domain and files it under Referral, next to forum links and directory listings.
A large share lands in Direct. Mobile apps and desktop apps often strip the referrer entirely. The session arrives with no source at all, so GA4 calls it Direct.
AI Overviews land in Organic Search. When Google answers inside the results page and someone clicks a citation, the referrer is still google. GA4 records it as google / organic, identical to a normal blue link click.
So the honest framing is this. You can isolate assistant traffic well. You cannot isolate AI answers inside Google search, and anyone selling you a GA4 setup that claims otherwise is guessing.
If Direct is already a mess for you, fix that first. I wrote about the usual causes in how to fix Direct and Not Set traffic.
The fast version: check before you build
Before you create anything permanent, take two minutes to see whether there is enough traffic to be worth reporting on.
Open Reports, then Acquisition, then Traffic acquisition.
Change the primary dimension to Session source.
Use the search box above the table and type chatgpt.
Repeat for perplexity, claude, gemini and copilot.
For most sites the totals are small right now. Do not let that put you off. The interesting part is almost never the session count.
Build an AI Assistants channel group
This is the proper fix. A custom channel group adds your own channel to GA4 without touching your tagging, and it applies to data you already collected, so you get history on day one.
Step by step
Go to Admin, then Data display, then Channel groups.
Click Create new channel group and name it something plain, like Default plus AI.
Click Add new channel and name the channel AI Assistants.
Set the condition to Source, then matches regex, and paste the pattern below.
Drag the AI Assistants channel to the top of the list, above Organic Search and Referral.
Save.
Order is the step people skip. GA4 evaluates channels top to bottom and stops at the first match. If AI Assistants sits below Referral, Referral claims the session first and your new channel stays empty forever.
The regex to paste
Keep it to the sources that actually send you clicks. A shorter pattern is easier to trust than a long one copied from a blog post:
chatgpt|openai|perplexity|claude|anthropic|gemini\.google|bard\.google|copilot|deepseek|grok|x\.ai|mistral|poe\.com|you\.com|phind
The assistants worth having in there today are ChatGPT, Perplexity, Claude, Gemini and Microsoft Copilot. The rest are small enough that you can add them the day they show up in your Referral report.
A few notes on that. Match on Source, not Page referrer, because Source is what the channel grouping actually reads. Do not include bing on its own or you will swallow ordinary Bing organic traffic. And escape the dots in gemini.google so the pattern does not match something unintended.
Google’s own reference for the feature is the channel groups documentation, which is worth a skim if you have never edited one. If you want the longer walkthrough with screenshots, I covered the whole feature in how to build custom channel groups in GA4.
Where to see it afterwards
Go back to Traffic acquisition and change the primary dimension to your new channel group. AI Assistants now appears as its own row, with history attached. The default channel group is untouched, so nothing else in your reporting shifts.
Build the exploration that actually answers the question
The channel group tells you how much. An exploration tells you whether it matters.
Create a Free form exploration with these settings:
Dimensions: Session source, Landing page plus query string, Device category.
Metrics: Sessions, Engaged sessions, Average engagement time, Key events, Total revenue.
Filter: Session source matches regex, using the same pattern from above.
Rows: Session source, then Landing page as a second row dimension.
Now you can see which assistant sends people, which pages they land on, and whether those people do anything. That last column is the one to read first.
If you want to watch the path all the way to a conversion, put the same filter on a funnel instead. The setup is in how to build a funnel exploration in GA4.
What the numbers usually say
Across the accounts I have looked at, AI assistant traffic behaves in a fairly consistent way.
Volume is small. Often one to four percent of sessions. Sometimes less.
Engagement is high. Engagement rate and time on page frequently beat organic search, because the assistant already filtered out the people who were never going to be interested.
Landing pages are deep. Not your homepage. Comparison pages, pricing pages, specific how to guides, documentation.
Intent is late stage. These visitors have already had the explanation. They came to check a detail or to see who you are.
That combination is why reporting on session count alone gives the wrong impression. A channel worth one percent of sessions and six percent of revenue is not a rounding error. Put revenue and key events next to sessions every time you show this to someone.
Getting it into a dashboard
Once the channel group exists, Looker Studio picks it up automatically as a dimension in the GA4 connector. Build a small page with four things and stop there: sessions by assistant over time, top landing pages, engagement rate compared with site average, and key events with revenue.
If you have the BigQuery export running, you can go further and query the raw referrer instead of the processed source, which catches assistants your regex missed. Setup for that is in how to set up the GA4 BigQuery export.
Catching what GA4 loses
Two gaps are worth closing if AI traffic matters to your business.
The referrer only exists on the first page
If someone arrives from an assistant, browses for a while, comes back three days later and then converts, GA4 attributes the conversion using its own attribution model and the original referrer may not be what you see on the lead.
The fix is the same one that fixes every attribution gap. Capture the first referrer in a cookie with Google Tag Manager, write it into a hidden form field, and store it on the record in your CRM. Then the sales team sees where the lead came from, not just GA4. I walk through the whole pattern in how to get accurate lead source data into your CRM.
If you want it visible inside GA4 as well, send the stored value as an event scoped custom dimension. That process is in how to set up GA4 custom dimensions.
Redirects eat the referrer
If an assistant links to a URL that redirects, some redirect setups drop the referrer on the way through and the session becomes Direct. Worth checking if your Direct number looks strange. Same root cause as UTM parameters being stripped by redirects.
What about AI Overviews and AI Mode
Short answer: you cannot separate them in GA4 today. Clicks from Google’s AI answers carry the same google / organic source as any other search click.
What you can do is watch the pattern from the other side. In Google Search Console, look for queries where impressions hold steady or rise while clicks fall. That gap is the signature of an answer being shown on the results page instead of a click coming to you.
It is indirect evidence, not a report. Treat it as a trend to monitor, not a number to put in a board deck.
Common mistakes
Editing the default channel group instead of creating a new one. You lose comparability with every historical report and with other properties. Always create a new group.
Putting the AI channel at the bottom of the list. It will never match anything. Top of the list, every time.
Filtering on Page referrer in the channel definition. Channel groups read Source. Use Page referrer in explorations, not here.
Including bing or google as bare terms. You will pull in ordinary organic search and the channel becomes meaningless.
Reporting sessions only. The whole story is in engagement and revenue. Sessions alone make a good channel look trivial.
Treating the number as complete. App traffic with no referrer is invisible. What you measure is a floor, not a total.
Frequently asked questions
Does a custom channel group change my historical data?
It does not change the data. It changes how the data is grouped at query time, and that grouping applies to sessions you already collected, so you get history immediately. Your default channel group keeps working exactly as before.
Why is my AI Assistants channel showing zero?
Almost always channel order. If a broader channel sits above it, that channel matches first and yours never gets a turn. Move it to the top and check again.
Should I use Source or Session source in the exploration?
Use Session source in explorations, because you are asking where a visit came from. In the channel group definition the field is simply called Source and refers to the same thing.
Can I track which assistant produced a sale?
Inside GA4, only within its attribution window and only when the session source survived. For a reliable answer, store the first referrer on the lead or order record in your own system. That version survives long sales cycles and does not depend on cookies.
Is AI traffic worth optimising for?
Measure it before you decide. If it converts above your site average, the sensible move is to make your comparison pages, pricing pages and specific answers clearer, since those are the pages assistants tend to cite. That is good for regular search too.
Do I need server side tracking for this?
No. Everything in this guide is a reporting change. Server side tracking helps with a different problem, which is signal loss, and it will not reveal a referrer the browser never sent.
The short version
GA4 will not show you AI traffic until you tell it what AI traffic is. Create a custom channel group, put it above Referral, filter on the handful of sources that matter, and look at engagement and revenue rather than session count.
Accept that the number is a floor. App traffic and AI Overviews will always sit outside it. A channel you can see and partly trust beats a channel you cannot see at all.
Want this set up properly?
If your reports are telling you a story your revenue disagrees with, that is usually fixable in days, not months. I build and repair tracking for ecommerce stores and lead generation businesses, so the numbers in your reports match the money in your bank account.
Have a look at what I do, or book a call and we can go through your setup together.
