If you’ve checked your Google Analytics 4 reports for visits from ChatGPT or Perplexity and come up nearly empty, you’re running into a default GA4 limitation: the traffic is very likely there, but GA4 doesn’t surface it on its own.
Around May 2026, Google reportedly rolled out a native AI Assistant channel to GA4 properties, which automatically separates out traffic from a set of recognized platforms. Coverage of the rollout reported that ChatGPT, Gemini, and Claude were included, but Google hasn’t published a complete list within its own help documentation.
This is a rolling update, so check your own GA4 property to confirm whether you have access to the update. Where it’s live, that update can be a helpful way to track traffic from AI search, but it only catches sessions that arrive with a clean referrer header attached.
A referrer header is a small piece of information a browser sends along when someone clicks a link, telling the destination site where the visitor came from.
A large share of AI-driven visits don’t carry one, and Perplexity in particular still lands in your generic referral traffic rather than getting its own bucket. The GA4’s native AI channel also doesn’t cover Google’s own AI Overviews or AI Mode. Clicks from those still get counted as Organic Search, so they won’t show up as AI traffic anywhere in GA4. If you want a fuller picture of third-party AI assistant traffic, you need to build your own custom segment alongside the native channel.
Why the native channel isn’t enough on its own
Some AI platforms don’t consistently send a referrer header, and some browsers strip it out before it ever reaches your analytics. When that happens, GA4 has nothing to sort the visit by, so it gets dumped into the direct traffic bucket with no label at all.
Google’s native AI Assistant channel solves part of this for the platforms it recognizes, as long as the referrer comes through intact and your property has the channel live. It doesn’t solve it for Perplexity, it doesn’t cover AI Overviews or AI Mode, and it doesn’t help with anything still arriving unlabeled. Building your own segment helps you catch what the native channel misses and build a more accurate picture of which LLMs are sending traffic your way.
Set up a custom segment to catch AI referral traffic
You’ll build this in GA4’s Explore section.
- Open Explore and start a free-form report. Log in to GA4, select Explore from the left-hand menu, and choose Free form as your exploration type.
- Create a new segment. In the Variables panel, find the Segments section and click the plus icon. Choose Session as your segment type, since you’re isolating visits rather than tracking a user across multiple sessions.
- Define your condition. Set the dimension to Session source and use “matches regex” as your operator. Enter a pattern that captures the AI platforms you want to track, for example:
perplexity|chatgpt|chat\.openai|claude\.ai|gemini\.google|copilot\.microsoft
Before you trust this pattern, check it against your own data. Pull up Session source / medium with no filter applied and see what’s actually there.
Two things to watch for. First, the same platform can show up under more than one medium. ChatGPT, for example, might appear as both a referral and under the native ai-assistant medium. Don’t assume a platform sent you nothing just because one row is missing.
Second, a domain that looks like an AI tool isn’t always one. Microsoft Copilot’s real domain is copilot.microsoft.com, not copilot.com, which is a separate, unrelated site. If you see a similar-looking domain in your data, confirm what it actually is before adding it to your pattern.
If GA4 offers a case-insensitive option for the match, use it. Referrer strings don’t always arrive in consistent casing, and a case-sensitive regex will quietly miss sessions that should have matched.
Update this list periodically. New platforms launch often, and the regex only catches what you’ve told it to look for
- Name and save the segment. Use something clear, like “AI Referral Traffic,” so it’s easy to find later.
- Add the segment to your report. Drag it into your exploration’s rows or columns, then add dimensions like landing page or session source/medium to see which platforms are sending visits and where they’re landing.
If you’ve also built a custom channel group for AI traffic, reorder it above Referral. A segment, which is what you just built, only affects how you view data inside Explore. A channel group is a separate GA4 object that changes how traffic gets classified across your standard reports. If you’ve set one of those up too, GA4 evaluates channel groups top to bottom, so your AI channel needs to sit above Referral or it’ll never get a chance to catch anything. If you’ve only built the segment in this guide, you can skip this step.
One thing to expect once you’re looking at the breakdown: the same platform can show up across more than one source/medium row. ChatGPT, for instance, often splits across a few variants depending on how the visit arrived, sometimes tagged as a referral, sometimes picked up by the native ai-assistant medium, sometimes landing as something else entirely. That’s not a sign your regex is broken. It just means one platform’s traffic doesn’t always consolidate into a single row, so check for a platform’s name across multiple rows before concluding it sent you nothing.
What to do with the data once you have it
Treat this number as a floor, not a ceiling. A segment built this way only captures sessions that pass a referrer header in the first place, so it will always undercount your actual AI-driven traffic, including the people who read an AI answer and later searched your brand name directly instead of clicking through.
Because of that undercount, the raw session total isn’t the number worth reporting on. What holds up regardless of how small or large that number looks is the comparison: how AI-referred visitors convert against your organic baseline. If that conversion rate is meaningfully different, that’s worth flagging to your team no matter how thin the volume is on its own.
Pair this with a look at your branded search trend in Search Console. When someone gets an answer from an AI tool and doesn’t click through, they’ll often search your brand name directly instead, the same way the segment we built here would never catch them. A branded search lift with no corresponding brand campaign behind it is a sign of AI visibility this segment alone won’t show you.
What Victorious Does With AI Referral Data
Building the segment is the easy part. Knowing what the numbers mean once you have them is where most teams stall out. We track AI-referred sessions against organic conversion rates, watch branded search for the lift that segments like this one can’t capture, and fold both into a transparent reporting system.
If you want help separating real AI-driven traffic from noise and turning that into data your team can act on, schedule a free consultation.
