For many B2B sites, that traffic is still small but growing fast. And with AI tools gaining popularity and potentially increasingly impacting how customers make buying decisions, it will be a question you need to try to answer. 

This article walks you through the GA4 setup for AI traffic analysis in five steps, explains the three things GA4 can't see, and uses data from B2B websites to determine whether or not the traffic is any good. 

Referrals, crawlers and visibility: the three types of AI traffic

The phrase "AI traffic" is often used to describe three entirely separate things, with only one referring to an actual human visit to your website. Here’s a quick breakdown:

In this post, we’re going to talk about AI referrals, or the human visits that happen after a user clicks on a link in the output of a tool like Claude, Gemini, or ChatGPT. If you’re looking for more information about other tools to track AI traffic, check out our guide on llm traffic analytics tools.

How AI assistants show up in GA4

Google Analytics 4 (GA4) is a web analytics tool that files visits from AI assistants into a default channel called “AI Assistant.” The channel includes visits from multiple sources, including: 

• ChatGPT

• Gemini

• Deepseek

• Copilot 

• Grok 

GA4’s AI assistant reporting specifically excludes Google's AI Overviews and AI Mode, which are part of Google Search and sit in Organic Search instead. Google added the channel on 13 May 2026, so compare a date range before and after that date before you read anything into a trend. 

To find this tab:

• Go to Reports

• Go to Acquisition

• Go to Traffic acquisition

• Find the AI Assistant row

Before the new channel existed, visits from AI assistants landed in the Referral tab. During that time, general marketing advice encouraged marketers to use a custom channel group with a regex rule to separate them out. If that’s the method you were using, it still works but may not have identical data with the new default group due to slight reporting differences. 

The differences come down to which sources each method recognizes. Google maintains the default channel's source list and can change it at any time, so a new assistant could appear in the default group before you've added it to your regex. The reverse can happen, too, and if your regex includes a source Google hasn't added yet, your custom group counts it as AI traffic while the default group files it under Referral. 

How to track AI referral traffic in GA4

Step 1: Open the AI Assistant channel

Go to Reports, then Acquisition, then Traffic acquisition. Set the primary dimension to Session default channel group and find the AI Assistant row. 

To get a quick overview of what’s happened in the past 90 days, you can look at: 

• Sessions: Your total volume of AI-referred visits.

• Engaged sessions: The subset that lasted more than 10 seconds, triggered a key event, or included more than one pageview.

• Key events: Visits that resulted in an action you defined, like a form fill or a demo-page visit.

Step 2: Split it by assistant

Not all AI assistants perform equally. Different assistants send different volumes and, as the quality section below shows, different quality traffic. You'll want to see the breakdown rather than the total.

Add Session source as a secondary dimension. The AI Assistant row breaks into separate lines based on the specific AI tool that sent the traffic your way.

Step 3: Add a custom channel group if you need one

The default AI Assistant channel covers the sources Google recognizes, but it might miss an assistant you care about. A custom channel group also lets you report AI traffic consistently across older and newer data.

Go to Admin, then Data display, then Channel groups, then create a new group. Add a channel, name it something like AI Assistants, and set the rule to match Session source against a regex. A reasonable starting list at the time of writing:

chatgpt\.com|perplexity\.ai|gemini\.google\.com|claude\.ai|copilot\.microsoft\.com|chat\.deepseek\.com|grok\.com

Place the channel above Referral in the list so it catches the match first. New assistants appear regularly, and old domains redirect. The older chat.openai.com now redirects to chatgpt.com, for example, so you don't need both. Review the list before you save.

Step 4: See which pages AI visitors land on

AI assistants tend to link to deep pages rather than the homepage. They may surface pricing pages, comparison articles, or documentation to help customers find answers to specific questions. Knowing which pages they're sending visitors to tells you which content the assistants are relying on when they cite you and what potential customers are asking about. 

Stay in the Traffic acquisition report, filter to the AI Assistant channel, and add Landing page as a secondary dimension. That's a tendency, not a rule, so let your own data confirm it.

Step 5: Compare AI visits with organic search

Organic search is the natural benchmark. Putting AI traffic next to it tells you whether AI-referred visitors engage and convert at a similar rate, or whether there's a gap worth investigating.

Put the AI Assistant and Organic Search channels side by side and compare two columns: engagement rate and key events per session. A fair comparison needs volume, so wait until you have at least a few hundred AI sessions before reading anything into the difference. If your AI channel is still in double digits, track it monthly and revisit next quarter. If you're asking which channel gets the credit, the Traffic acquisition report credits a key event to the channel of the session it happened in, so a visitor who first came from ChatGPT and converted later through a branded search counts under Organic Search.

GA4 compares visits, not who the visitors were. Whether AI visitors are from companies you actually sell to is a different question, and the next two sections pick it up.

AI traffic and what GA4 can’t see

Even after you've set everything up, GA4 is very likely undercounting your AI traffic. There are three gaps that cause this to happen, and unfortunately none of them are fixable inside GA4 alone. Knowing where the blind spots are keeps you from treating the numbers as the complete picture.

AI visits hiding in direct traffic

Some visits from AI assistants arrive with no referrer. This happens when someone copies a link from an AI app rather than clicking it, or when the app strips the referrer header. GA4 files those visits as Direct. 

Watch for direct traffic rising on deep pages nobody would type into a browser, such as a comparison article or a technical doc. That's a signal and not necessarily proof, but it's the best indicator you have that some AI-referred visits are hiding in your direct bucket.

AI Overviews and AI Mode buried in organic

Clicks from Google's AI Overviews and AI Mode sit inside Organic Search. GA4 can't separate them from a regular search click, and there's no filter or secondary dimension that will. Search Console's Generative AI performance report shows impressions in AI Overviews and AI Mode, but not clicks. 

You can see whether your pages appear in those features, but not how many visits they drove. For the search side of this problem, seo web analytics covers what Search Console can and can't tell you.

The answer that never becomes a visit

When an assistant answers a question using your content and the user gets what they need without clicking, no visit happens. That's visibility, not traffic, and it belongs to the third row of the table at the top of this article. Getting mentioned and cited by AI assistants is its own discipline that’s worth accounting for, but that’s another topic for another day.

Is AI traffic good traffic?

Knowing how much AI traffic you get is only useful if you also know whether it's the right traffic. While general advice can tell you AI traffic is growing, it’s hard to determine whether the companies it brings are the ones you sell to. That takes company-level data to answer.

Leadfeeder's ai traffic quality study found that "AI assistants send B2B websites three times the traffic they did 18 months ago, but that traffic is measurably less likely to match the site's ideal customer profile, about 3 points below organic search even after adjusting for where the visitors come from." In other words, the volume is real and growing, but the fit is weaker on average.

AI referrals rank with social on ICP fit, below both kinds of search. And the gap isn't uniform across assistants. Visitors referred by Claude fit 1.3x to 1.4x better than visitors referred by ChatGPT, an advantage concentrated in the largest B2B properties. That's why Step 2 above splits by assistant rather than lumping them together.

Keep in mind that "fits your ideal customer profile" means something specific. Your ICP is a description of the company most likely to buy from you and get value from your product, defined by attributes like industry, company size and location. The ICP ratio is the share of identified companies that fit that profile. When the study says AI traffic fits less well, it means a smaller share of the identified companies matched the site's ICP.

So what should you do with this? Simply put, don't assume AI traffic is better or worse on your site. These are averages, and averages hide variation by assistant, market and content type. Your site's AI traffic could be better or worse than the study's numbers. The only way to know is to check the ICP fit on your own traffic, which is what the next section walks through

How to check AI traffic quality on your site

GA4 can tell you how many AI-referred sessions you got, but it can't tell you whether those sessions came from companies you sell to. That takes company-level data.

Leadfeeder Web Insights adds that layer to your traffic reporting. It sits alongside GA4 rather than replacing it. GA4 counts sessions and users, Leadfeeder counts companies, and the two sets of numbers won't match because they count different units. 

Web Insights shows which channels, campaigns and pages bring the companies you want, so you can spend where the leads actually come from. The Traffic Sources report shows traffic by source and medium, and it can be filtered to AI assistants such as ChatGPT, Perplexity and Gemini. For those sources it shows how many companies came and the ICP ratio, never company names. Open the Traffic Sources report, filter to AI assistant sources, and compare the ICP ratio with organic search to see whether the study's finding holds on your site.

To see which companies arrived from an AI assistant, use B2B website visitor identification in Web Visitors. Filter by referrer, such as chatgpt.com or perplexity.ai, and you'll see each identified company by name and whether it fits your ICP.

The identification works through reverse IP lookup, which matches a visit's IP address to the organization behind it. It identifies organizations, not people.

Company-level data is where AI traffic analytics meets the broader practice of website intelligence, which looks at which organizations were behind the visits.

What to do with the numbers

Having the data is one thing, and reporting it in a way that leads to good decisions is another. 

AI traffic is easy to overstate because growth rates on a small base look dramatic, and it's easy to understate because the absolute numbers are still modest. Here's how to keep the reporting honest.

Report growth and share together. A channel that tripled year over year can still be a small share of total visits, and either number in isolation misleads. Put them side by side so the reader of your report gets scale and trajectory in one line. For agency analysts putting AI traffic into a client report for the first time, that pairing keeps the number honest: the client sees the trend without overstating the volume. For the format, web analytics report covers how to lay it out.

Watch the landing pages. The pages AI assistants send people to tell you which content the assistants rely on. If your pricing page or a key comparison article shows up, that content is doing work you won't see in a conversion report. If a page you consider low-priority keeps appearing, the assistants are treating it as authoritative, and that's worth knowing.

Let fit guide your reaction, not volume. Don't chase AI traffic on growth alone. A fast-growing channel with poor ICP fit costs you attention without returning pipeline. Run the quality check from the section above quarterly and use fit to decide how much energy this channel deserves.