How to measure AI search traffic in GA4

GA4 does not have an AI traffic channel, and Google has shown no urgency about building one. Until it does, AI referrals sit inside your Referral bucket, mixed in with newsletters and partner links, and a meaningful chunk of AI-influenced traffic never gets attributed at all. Here's the setup that works, followed by the part most articles skip: how much of the picture this method actually gives you.

The regex, first

Paste this into any GA4 regex filter on Session source (or Session source / medium, if you prefer):

.*(chatgpt\.com|chat\.openai\.com|openai\.com|perplexity\.ai|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|bing\.com/chat|claude\.ai|anthropic\.com|you\.com|poe\.com|phind\.com|grok\.com|x\.ai|deepseek\.com|mistral\.ai|kagi\.com|arc\.net|meta\.ai|duckduckgo\.com/aichat).*

Notes on why it looks like that:

• GA4 uses RE2, and its matches regex operator wants a full match in several UI surfaces. Wrapping the group in .* on both ends makes it behave like a "contains" filter in every place you'll paste it. Keep the wrappers.
• Dots are escaped. chatgpt.com unescaped would also match chatgptXcom, which is unlikely to matter and free to fix.
• chat.openai.com is legacy but still shows up in older data, so keep it alongside chatgpt.com.
• copilot.microsoft.com and bing.com/chat are separate because Microsoft has moved Copilot around more than once.
• Drop entries you don't care about. A shorter regex evaluates faster and is easier to defend when someone asks what it does.

Sanity-check it against your own data before you trust it. Go to Reports → Acquisition → Traffic acquisition, switch the dimension to Session source, and search for chatgpt. If you see zero rows, either you have no AI traffic yet, or your data has a problem worth finding before you build reports on top of it.

Building the exploration

Reports → Explore → Free form. Then:

That gives you: which AI engines send traffic, which pages they land on, and whether those sessions convert. The landing page column is the one people underuse. It tells you exactly which pages the engines are citing, which is a direct read on what content is working and, by omission, what isn't.

If you want this in your standard reports rather than an exploration, build a custom channel group: Admin → Data display → Channel groups → Create new. Add a rule named AI Search with condition Session source matches regex <your regex>, and drag it above Referral in the ordering so it claims those sessions first. Rule order is evaluated top-down and the first match wins, which trips people up regularly.

One caveat on custom channel groups: they don't retroactively rewrite every report, so you'll have a seam in your history at the creation date. Build it now rather than in three months, when the seam will be more annoying.

What this setup cannot see

This is the section that separates a useful analyst from a dashboard-builder. Four blind spots, in rough order of how much they distort the picture.

Most AI reads never produce a click

The biggest one. Pew Research found that only 8% of users click a search result when an AI summary is shown, versus 15% without, and just 1% click a link inside the summary itself.

One percent. That's the conversion rate from "your brand was cited in an AI answer" to "GA4 records a session." Everything else, the 99%, is a buyer reading your brand name, forming an impression, and moving on without ever touching your analytics. That impression is real. It shapes shortlists. GA4 will never show it to you.

Ahrefs measured the same phenomenon from the other side: in a 300,000-keyword Google Search Console study, the #1 organic result sees 58% lower CTR when an AI Overview is present. Same rank, same query, most of the clicks gone.

So when your AI referral number comes back at 400 sessions a month and someone says "AI is irrelevant to us," the correct response is that 400 clicks at a 1% click-through implies an enormous number of AI reads you can't measure. GA4 is not a visibility instrument. It's a click instrument. Don't confuse the two.

AI Overviews traffic is invisible

A click from a Google AI Overview arrives as google / organic. It is indistinguishable from a click on the fifth blue link. Search Console doesn't separate it either. Given that AI Overviews now appear on roughly 48% of tracked queries and on 82% of B2B technology queries, that's a large and growing part of your search surface hiding inside a bucket you can't split.

The best available proxy: in Search Console, find queries where impressions are flat or rising while clicks are falling. That divergence is the signature of an Overview eating the click. It's inferential, not definitive, and it's still better than nothing.

In-app browsers and stripped referrers

Some AI clients open links in an in-app webview that doesn't pass a referrer, and some strip it deliberately. Those sessions land in GA4 as Direct. So a portion of your "direct traffic" spike is AI referral you'll never attribute. You can partially catch this by watching for direct-traffic anomalies on deep pages, since nobody types a URL like /blog/how-to-build-a-prompt-set from memory, but it's forensic work rather than reporting.

Assistant-mediated actions

A user asks an assistant to summarize your pricing page, and the assistant fetches it. Depending on the client and whether it executes JavaScript, that may fire your GA4 tag as a session with no human behind it, or it may not fire at all. Neither outcome is a meaningful number. Server log analysis, filtering for GPTBot, PerplexityBot, ClaudeBot and friends, is the only clean way to see AI crawler activity, and it lives outside GA4 entirely.

What the traffic you can see is worth

Small, but disproportionately valuable. Two independent measurements agree:

• Semrush found an AI-search visitor is 4.4x as valuable as an organic search visitor by conversion rate.
• Adobe Analytics found AI-referred traffic to US retail sites grew 138% year over year and converts 54% better than non-AI traffic.

The mechanism is obvious once you say it out loud: the AI already did the qualifying. It understood the question, filtered the options, and sent a pre-sold visitor. That person is further along than someone who clicked the third result on a broad query.

The volume is also moving. ChatGPT referral traffic rose 157.7% week-over-week after OpenAI put clickable brand links directly in answers in May 2026. One product change, a step function in referrals. Expect more of those, in both directions.

So report AI traffic on revenue per session and conversion rate, not on session count. If you benchmark it against organic volume, it will look like a rounding error and get defunded, right as it becomes the channel that decides your shortlist placement.

GA4 vs. visibility tracking: what each one answers

They're complementary, and neither is optional. GA4 tells you what the clicks did. Visibility tracking tells you what the 99% saw. Run the regex above for the clicks, and run your prompt set through the engines for the reads. Reporting only the first one is how you end up concluding that AI doesn't matter, in a market where 51% of B2B software buyers now start their research in a chatbot.

What to do next

1. Paste the regex into a Free Form exploration on Session source and set the date range to the last 90 days. Five minutes.
2. Build the custom channel group so AI Search appears in your standard acquisition reports from today forward. The seam only gets worse if you wait.
3. Look at the landing page column. Those are the pages the engines are citing. Note what's missing.
4. Report conversion rate and revenue per session, never raw sessions, when you present this internally. Volume framing kills the channel.
5. Pair it with a visibility scan so you're measuring the reads, not only the 1% of them that click. The two numbers together are a channel. Either one alone is a guess.