If you want to track AI traffic in GA4, the setup is much easier than it was a year ago.
Google Analytics now has a dedicated AI Assistant channel.
When GA4 recognises a referral from an AI assistant, it can automatically classify the visit using:
- Medium:
ai-assistant - Campaign:
(ai-assistant) - Channel:
AI Assistant
Google introduced this classification on 13 May 2026. Google Help
That means you no longer need to hunt through one giant Referral bucket just to find every visit from ChatGPT, Gemini or Claude.
But there are still three reasons to set your reporting up properly.
First, not every AI-originated visit will preserve enough referral information to be classified correctly.
Second, Google says its AI Assistant channel excludes traffic from Google AI Overviews and AI Mode. Google Help
Third, counting AI sessions does not tell you whether those visitors generate enquiries, revenue or qualified pipeline.
The useful questions are:
How much AI traffic are we getting?
Which assistants are sending it?
What happens after those visitors land?
Table of Contents
- What changed in GA4 in 2026?
- The fastest way to find AI traffic
- How to see which assistant sent the visit
- How ChatGPT referral tracking works
- When to build a custom AI channel group
- The GA4 Exploration worth saving
- Which landing pages attract AI traffic?
- What should you measure beyond sessions?
- What is a good share of AI traffic?
- Why some AI visits will still be missing
- Can you see the prompt that generated the click?
- How to report AI traffic internally
- Frequently Asked Questions
What changed in GA4 in 2026?
Until recently, most AI-traffic tutorials told you to look inside Referral and search for domains associated with:
- ChatGPT
- Perplexity
- Claude
- Gemini
- Copilot
That advice made sense at the time.
It is no longer the best starting point.
Google Analytics now recognises AI-assistant traffic as a separate channel.
When a recognised AI referrer is detected, GA4 can assign:
| GA4 Field | Value |
|---|---|
| Medium | ai-assistant |
| Campaign | (ai-assistant) |
| Default channel group | AI Assistant |
What changed in GA4 in 2026?
Until recently, most AI-traffic tutorials told you to look inside Referral and search for domains associated with:
- ChatGPT
- Perplexity
- Claude
- Gemini
- Copilot
That advice made sense at the time.
It is no longer the best starting point.
Google Analytics now recognises AI-assistant traffic as a separate channel.
When a recognised AI referrer is detected, GA4 can assign:
| GA4 Field | Value |
|---|---|
| Medium | ai-assistant |
| Campaign | (ai-assistant) |
| Default channel group | AI Assistant |
Google describes the channel as traffic coming from AI-assistant sources and notes that it excludes Google AI Overviews and AI Mode. Google Help
That makes the reporting model cleaner.
Source tells you where the visit came from.
Medium categorises the acquisition type.
Channel group gives you the wider bucket.
You want all three.
The fastest way to find AI traffic
Start here:
GA4 → Reports → Acquisition → Traffic acquisition
Use:
Session default channel group
Then look for:
AI Assistant
If there is measurable AI-referral activity, that row gives you the sessions GA4 has placed into the native AI channel.
That is your quickest baseline.
You do not need a regex to begin.
How to see which AI assistant sent the visit
The AI Assistant channel tells you the traffic category.
It does not tell you enough about the individual sources.
To break the data down:
- Open Traffic acquisition
- Find or filter AI Assistant
- Add a secondary dimension
- Select Session source or Session source / medium
You may then see sources associated with platforms such as:
- ChatGPT
- Gemini
- Claude
- Copilot
- Perplexity
- other assistants recognised by your setup
The exact values matter more than a fixed list because referrer patterns can change.
Use the traffic your property is actually receiving.
ChatGPT referral tracking is unusually clear
OpenAI currently makes ChatGPT referral traffic comparatively easy to identify.
Its publisher guidance says that ChatGPT automatically includes:
utm_source=chatgpt.com
in referral URLs from ChatGPT Search.
That allows publishers to measure inbound ChatGPT traffic using analytics platforms such as Google Analytics. OpenAI Help Center
So if GA4 records a source associated with:
chatgpt.com
you have a strong attribution signal.
OpenAI also says that publishers who want their site content eligible to appear in ChatGPT Search should allow OAI-SearchBot to access it. OpenAI Developers
That gives you a simple chain:
Crawler access
→ Page can be discovered
→ Page is surfaced or cited
→ User clicks
→ Referral reaches GA4
For the crawler part, link naturally to which AI crawlers you should allow in robots.txt
When to build a custom AI Assistant channel
For most sites, use GA4’s native AI Assistant channel first.
A custom channel group becomes useful when:
- you want your own source definitions
- you want to include additional assistants
- you need a consistent historical reporting method
- you want a separate board-reporting definition
- you need to compare your own classification with Google’s native one
Google itself provides a custom AI-assistant channel example. Google Help
Open Channel Groups
Go to:
You need Editor-level property access or above to create or edit channel groups.
Google Analytics: Custom channel groupsCreate a Group
Choose:
Give it a clear name such as:
Add a Channel
Create:
Use the relevant traffic-source field and select matches regex.
Google currently gives this example:
^.*ai|.*\.openai.*|.*chatgpt.*|.*gemini.*|.*gpt.*|.*copilot.*|.*perplexity.*|.*google.*bard.*|.*bard.*google.*|.*bard.*|.*.*gemini.*google.*$Reorder the Channel
Place:
above:
GA4 assigns traffic to the first matching channel in the list, so ordering matters.
Depending on your reporting definition, you may also choose to place the AI channel above Organic Search.
Google Analytics: Channel ordering guidanceSave and Test
Return to:
Select your custom channel group as the report dimension and check whether your AI-assistant traffic is being classified as expected.
Then select your custom channel group as the primary or secondary dimension.
Google says custom channel groups can also be used in Explorations and can be applied to historical reporting data. Google Help
The GA4 Exploration worth saving
The standard Traffic acquisition report is good for a quick check.
For ongoing GEO reporting, create one Free-form Exploration.
Go to:
Explore → Free form
Add these dimensions:
- Session default channel group
- Session source
- Session source / medium
- Landing page + query string
Then add:
- Sessions
- Engaged sessions
- Engagement rate
- Key events
- Session key event rate
- Revenue, where relevant
Add the filter:
Session default channel group = AI Assistant
Then use:
Session source
followed by:
Landing page + query string
as the report rows.
Now you have something much more useful than:
ChatGPT sent 42 visits.
You can answer:
Which assistants send traffic?
Which pages do they send users to?
Do those users engage?
Do they enquire?
Which landing pages are AI assistants sending traffic to?
This is one of the most commercially useful parts of the report.
AI assistants may send users to:
- blog articles
- service pages
- case studies
- guides
- research
- comparison pages
- FAQ pages
- the homepage
That matters because volume alone can mislead.
Compare:
300 AI sessions → blog posts → 0 enquiries
with:
40 AI sessions → service pages → 5 enquiries
The second traffic source is much smaller.
It may also be much more commercially useful.
What should you measure beyond sessions?
Do not stop at:
AI assistants sent us 87 visits.
Measure what those sessions did.
Sessions
How much measurable AI-referral traffic arrived?
Share of sessions
What percentage of all website sessions came through the AI Assistant channel?
Landing pages
Which pages attracted the clicks?
Engagement rate
Did AI visitors stay and interact?
Key events
Did they:
- submit a form
- call
- book a consultation
- download an important document
- complete another useful commercial action
Session key event rate
What percentage of AI sessions included at least one key event?
Revenue or qualified enquiries
This is where the reporting becomes useful to a board or commercial team.
A high engagement rate can be encouraging.
A qualified enquiry is stronger evidence.
What is a good level of AI referral traffic?
There is no useful universal benchmark.
A commercial cleaning company, SaaS product, dental clinic and recruitment firm will have very different buyer behaviour.
Do not tell every client:
AI should be 5% of your traffic.
Instead, calculate two ratios.
Why some AI visits will still be missing
GA4’s AI Assistant channel improves reporting.
It does not make attribution perfect.
A measurable referral depends on enough source information surviving the click.
Traffic can lose attribution because of:
- redirects
- privacy controls
- embedded browsers
- app behaviour
- missing referrer data
- URL handling
- intermediary systems
When GA4 cannot determine the source clearly, a visit may be attributed elsewhere, including Direct.
So the correct interpretation is:
GA4 shows measurable AI referral traffic, not necessarily every user influenced by an AI assistant.
That distinction is important.
Google AI Overviews are a separate measurement problem
Google’s own documentation says that the AI Assistant default channel excludes:
- Google AI Overviews
- Google AI Mode Google Help
That means you cannot look at the AI Assistant row and assume it represents every AI-influenced visit.
This is one reason GEO reporting needs more than web analytics.
GA4 tells you:
Who clicked from a measurable traffic source?
It does not tell you:
How many people saw your brand in an AI answer and never visited your site?
For that side of the measurement, link to how to run your own AI visibility audit
Can you see the prompt that generated the click?
Usually not.
GA4 may show:
- source
- medium
- channel
- landing page
- campaign information
It does not receive the user’s private ChatGPT or Claude conversation as a normal analytics dimension.
OpenAI currently documents:
utm_source=chatgpt.com
as referral attribution from ChatGPT Search.
It does not say the user’s original prompt is passed to the publisher. OpenAI Help Center
So if a buyer asks:
Which facilities management companies should I consider in London?
and then clicks your page from ChatGPT, you may be able to measure the ChatGPT referral.
You should not expect GA4 to show the private question that triggered it.
How to report AI traffic to a board
Do not give leadership thirty referral domains.
Give them a concise commercial view.
AI Assistant traffic
Sessions: 126
Share of website sessions: 1.8%
Month-on-month: +34%
Commercial performance
Key events: 9
Session key event rate: 7.1%
Organic Search key event rate: 4.8%
Main sources
- ChatGPT
- Perplexity
- Claude
Main landing pages
- AI visibility guide
- Facilities management case study
- GEO service page
Interpretation
AI referral traffic remains a small acquisition channel, but visitors currently show a higher-than-site-average key event rate. Continue monitoring before treating the difference as a stable trend.
That last sentence matters.
A small sample can produce dramatic percentages.
Do not turn:
12 visits and 2 enquiries
into:
AI traffic converts at 17%
without explaining the size of the sample.
Frequently Asked Questions About AI Traffic in GA4
Why is my AI traffic so low?
There are several possible reasons.
Your audience may not use AI assistants heavily for your category.
Assistants may answer the query without generating an outbound click.
Your website may not appear frequently.
Some visits may lose referral information.
And Google’s AI Overviews and AI Mode are excluded from the native AI Assistant channel.
Google Analytics: Default channel groupsDoes AI referral traffic convert better?
There is no universal answer.
Compare your own:
- Key events
- Key event rate
- Revenue
- Enquiry quality
against channels such as Organic Search and Referral.
The useful question is not whether AI traffic converts better globally.
It is whether it converts better for your business.
Can I see the prompt that sent the user?
Normally no.
Referral analytics identify the source of the visit, not the private conversation behind it.
OpenAI currently adds utm_source=chatgpt.com for referral tracking from ChatGPT Search, but it does not document
passing the user’s prompt to publishers.
Do I still need a custom regex in 2026?
Not necessarily.
GA4 now has a native AI Assistant channel.
Use that first.
A custom channel group is useful when you want to:
- Include additional sources
- Maintain your own reporting definition
- Compare historical patterns
- Build custom internal dashboards
Google provides an official AI-assistant regex example for this purpose.
Google Analytics: Custom channel groupsIs ChatGPT traffic still classed as Referral?
Recognised AI traffic can now be assigned:
ai-assistant
as the medium and placed into GA4’s:
AI Assistant
channel.
Older tutorials may still describe ChatGPT as ordinary Referral traffic because that was the earlier reporting setup.
Google Analytics: What’s newCan GA4 track Google AI Overview clicks separately?
Not through the AI Assistant channel.
Google explicitly excludes AI Overviews and AI Mode from that channel.
Google Analytics: Default channel groupsShould I report AI sessions or users?
For channel reporting, start with sessions.
Then add:
- Engaged sessions
- Key events
- Key event rate
- Revenue
- Qualified enquiries
Users can still be useful, but session-level reporting is clearer when you are analysing individual acquisition visits.
AI Traffic Is Easy to Count. Useful AI Traffic Is What Matters.
In 2026, the starting point is simple.
Go to:
Reports → Acquisition → Traffic acquisition
Find:
AI Assistant
Then break the traffic down by:
Session source / medium
From there, measure:
- sessions
- share of traffic
- landing pages
- engagement
- key events
- session key event rate
- enquiries
- revenue
Use custom regex only when you need additional control.
Save an Exploration if you want repeatable reporting.
And keep the limitations visible.
GA4 can measure attributable outbound visits.
It cannot tell you every time an AI system mentioned your business.
It cannot recover every click where referral information was lost.
And it cannot show you the user’s private prompt.
That is why GEO measurement needs two views:
Visibility: are assistants naming or citing us?
and:
Behaviour: when people click, what do they do?
You need both.
Get your free AI visibility audit
Run your own AI visibility audit
