How to Track AI Traffic in GA4 (And What It Still Misses)

GA4 finally has a native AI Assistant channel, which means the five-minute setup every guide told you to do is now partly built in. It also means the number in front of you is easier to read and just as incomplete, because Google's own definition of that channel excludes the largest source of AI-influenced traffic most UK businesses have.

The Native AI Assistant Channel Changed the Setup, Not the Problem

For most of the past two years, tracking AI traffic in GA4 meant building your own channel group, because the default grouping had no concept of an AI assistant. Sessions from chatgpt.com or perplexity.ai landed in Referral alongside directory listings and forum links, and stayed invisible until you separated them yourself.

That has changed. GA4's default channel group now includes an AI Assistant channel, and it populates automatically on every property. Google's default channel group reference defines it as the channel by which users arrive from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok, and states directly that it excludes Google's AI Overviews and AI Mode. Under the bonnet, GA4 stamps the session medium as ai-assistant when the inbound referrer matches its list.

Read that exclusion again, because it is the whole story. Google has given you a tidy channel for third-party assistants and explicitly kept its own AI surfaces out of it. For most UK businesses, Google's AI surfaces are the larger source of AI-influenced traffic by some distance. So the arrival of a native channel makes the small, measurable slice easier to see, while leaving the big, unmeasurable slice exactly where it was: inside your organic numbers, unlabelled.

What the GA4 AI Assistant Channel Actually Covers

Before you build a report on it, be clear about its three boundaries. None of them are bugs. All three are documented or follow directly from how GA4 works.

• It only covers the assistants on Google's list. The published definition names ChatGPT, Gemini, Deepseek, Copilot and Grok. Anything not recognised still falls through to Referral. Across our client base Perplexity is usually the single largest AI referrer, and it is not among the names Google gives, so the channel can under-report the source that matters most to you.
• It is forward-only. The classification depends on a medium written at collection time, so it tells you nothing about the months before it went live. This is the opposite of how custom channel groups behave, and it catches people out.
• It excludes AI Overviews and AI Mode by design. Google says so in the definition itself. If you were hoping the native channel finally isolates AI Overview clicks, it does not, and no configuration change will make it.

The practical read: use the native channel as your quick baseline, and keep a custom channel group running underneath it for everything the list misses.

How to Track AI Traffic in GA4 Beyond the Default Channel

A custom channel group is still worth the five minutes, for two reasons the native channel cannot cover: you choose which platforms count, and it reclassifies historical sessions, because custom groups are applied when you run the report rather than baked in at collection.

1. Open Admin in your GA4 property.
2. Under the Data display column, click Channel groups.
3. Click Create new channel group and give it a clear name, such as "Default with AI".
4. Click Add new channel, name the channel AI, and set the condition to Source matches regex.
5. Paste in a pattern covering the assistants you care about, escaping the dots so they are read literally:
   chatgpt\.com|openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|copilot\.com|deepseek\.com|grok\.com|mistral\.ai|you\.com
6. Use Reorder to drag your AI channel above Referral, then save.

Step six is the one people miss, and it quietly breaks everything. GA4 evaluates channel rules from the top down and assigns each session to the first rule it matches. Leave the AI channel below Referral and every AI visit gets claimed by Referral before your rule is ever tested, so the channel sits there reporting zero and you conclude you have no AI traffic. Google's documentation on custom channel groups covers the ordering behaviour and the other conditions available.

Keep the pattern under review. New assistants appear regularly, and a regex written a year ago will silently under-report. A quarterly look down your Referral channel for unfamiliar domains is enough to stay current.

Building a GA4 Exploration for Deeper AI Analysis

Channels answer "how much". An Exploration answers the more useful questions: which assistant, landing on which page, and doing what once they arrive.

1. Click Explore in the left navigation and start a Blank exploration.
2. Add Session source and Landing page as dimensions, and Sessions, Engagement rate and your key conversion event as metrics.
3. Add a filter on Session source using matches regex and the same pattern as above. Filtering on source rather than the new medium keeps Perplexity and anything else off Google's list in the report.
4. Set a long date range. AI referral volumes are low enough that a seven-day window tells you nothing useful.

The landing page breakdown is the part worth your attention. It tells you which specific pages assistants are citing, which is a direct read on what your AI-visible content actually looks like. In our experience those pages are rarely the ones a business expects. Definitional explainers, pricing pages and detailed comparison content tend to pull far more AI citations than the polished commercial landing pages that get all the internal attention.

One caveat on interpreting the engagement metrics: AI-referred visitors often arrive better informed and further down the funnel, because the assistant has already answered their basic questions. Judging them on session duration against your organic average will usually undersell them. Conversion rate is the fairer comparison.

How to Track AI Overview and AI Mode Traffic in GA4

You cannot, and it is worth understanding exactly why so you stop looking for the setting. Google's AI surfaces live on google.com, so a click from an AI Overview or from AI Mode arrives in GA4 like any other Google click: source google, medium organic. There is no separate referrer, no parameter you can rely on long term, and now an explicit statement in Google's own channel definition that the AI Assistant channel does not include them.

This matters because those surfaces are, for most UK businesses, a far larger source of AI-influenced traffic than every third-party assistant combined. The most consequential slice of your AI traffic sits inside your organic numbers, unlabelled and unsegmentable.

There is no clean fix, but there are two partial ones. Search Console is the better instrument, because impressions and clicks from Google's AI experiences are reported within your existing Search performance data. You can watch for the classic signature: impressions holding steady or rising while clicks on the same queries fall away. That pattern is what AI answers absorbing your clicks looks like in aggregate. Second, some teams use Google Tag Manager to read URL parameters present on AI Overview clicks and fire a custom event into GA4. It can work, but treat it as a directional signal rather than a source of truth, because the parameters are undocumented and Google can change them without notice.

The Dark Traffic Problem That Hides Most AI Visits

The larger measurement challenge is everything GA4 does not capture at all. When a referrer header is missing, an AI-influenced visit lands in direct / (none), the same bucket that catches bookmarks, copy-pasted URLs and email clicks without UTM tags. There is no native way to separate those visits from genuinely direct ones, and the AI Assistant channel cannot help, because it needs a referrer to match against.

Picture the journey. A user asks Perplexity which UK digital marketing agency they should shortlist, reads the answer, then types a brand name straight into their browser. That session is indistinguishable from someone who already knew the address. The AI did the influencing, but the analytics record gives all the credit to direct traffic.

This is not a small error. Based on the scale of AI platform usage and typical zero-click-to-referral ratios, the actual AI-influenced traffic reaching most websites is likely five to ten times higher than referral data alone shows. If you brief a board or a client on AI's impact using only the GA4 referral row, you will materially undercount it.

A practical partial check: watch direct traffic to deep pages. Genuine direct traffic goes to your homepage, or to pages people bookmark. When you see direct sessions landing on a specific technical article three levels down in your site structure, something referred that visit and the referrer was stripped. It is not proof, but a rising trend in that segment is one of the few AI signals available inside GA4 itself.

AI Crawlers Are Not AI Visitors

One confusion worth clearing up, because it sends people looking for traffic that was never going to be there. GPTBot, ClaudeBot, PerplexityBot and Google-Extended are crawlers. They read your pages so a model can learn from or cite them. They are not sessions, and they are not what the AI Assistant channel counts.

In practice they will not appear in GA4 at all, because GA4 measures through JavaScript that most crawlers do not execute. So a page can be crawled heavily by AI systems and show nothing whatsoever in your analytics. That is expected behaviour, not a tracking fault, and it is the reason server logs earn a place in the stack below: they are the only place the crawl is visible.

The sequence matters for interpretation. Crawl comes first, citation second, referral click a distant third. If you only watch the referral number, you are watching the last and smallest step in a chain, months after the part you could have acted on.

How to Build a Complete AI Traffic Measurement Stack

Because no single tool captures the full picture, the answer is to layer several signals. Each one fills a gap the others leave open. Here are the six components we use when we set up AI visibility tracking for a client. The visibility side has six checks of its own, set out in our answer engine optimisation audit, and the two stacks are meant to run together.

1. The native AI Assistant channel, plus a custom group underneath

Start with the default channel for a free baseline, then run a custom channel group alongside it to catch Perplexity and anything else Google does not recognise. The custom group also reclassifies historical sessions, which the native channel cannot do. Neither will ever capture Google's own AI surfaces.

2. Search Console for Google's AI surfaces

Because AI Overview and AI Mode clicks are reported inside your ordinary Search performance data, Search Console is where you monitor them. Track query-level impressions against clicks over time. Impressions up with clicks flat or down is the clearest available fingerprint of AI answers intercepting your traffic.

3. Bing Webmaster Tools

Microsoft's surfaces, Copilot and Bing Chat, are more transparent about attribution than most. Bing Webmaster Tools gives clearer visibility into AI-driven impressions and clicks than you will get from Google's tooling for third-party AI traffic, and it is free to connect.

4. LLM mention and citation tracking

Tools in this category test whether your brand or content appears in language model answers to specific queries. This is arguably the most important signal in the stack, because it tells you whether AI is citing you even when no click follows. A citation without a click is still brand exposure, and it feeds future branded search. Before paying for one, read our breakdown of how AI citation tracking tools work, including the six things none of them can see.

5. Server log analysis for AI crawlers

AI systems crawl before they cite. Filtering your raw server logs for known AI user agents, such as GPTBot, ClaudeBot, PerplexityBot and Google-Extended, shows which pages the models are actually reading. Rising crawl activity on a page is an early indicator that it may start surfacing in answers, and it is invisible everywhere else. When we ran this filter over 30 days of our own logs, bots from six AI and search companies made more than 11,500 requests, and not one of them asked for our llms.txt file. The two commands we used, and what the counts showed, are in our llms.txt server log test.

6. AI position and share-of-voice tracking

Rank trackers are adding AI Overview and answer-citation monitoring. Watching your share of AI citations for target keywords gives you an upstream signal that will eventually drive referrals, well before the referral data itself becomes legible in GA4.

The Metric That Actually Matters

If you measure AI's impact only by counting referral sessions, you are measuring the wrong thing. The question is not how many clicks AI sent you last month. It is whether you are being mentioned when people ask AI about your category.

Share of voice in AI answers, the percentage of times your brand or content surfaces when users prompt an assistant on relevant topics, is the upstream metric that will shape organic traffic, branded search volume and direct visits over the next few years. It is harder to quantify than a session count, but it is the number that separates the businesses that stay visible from the ones that quietly disappear from the top of the search experience.

Google's own guidance backs this up. According to Google Search Central on succeeding in AI search, the same fundamentals that earn strong organic visibility, helpful and reliable content with clear structure, are what surface in AI experiences. Sites built on that foundation appear in the answers. Those that are not are increasingly invisible at the top of the page, regardless of their traditional rankings.

What This Means for Your SEO Strategy

Optimising for AI citations builds on the same foundation as traditional SEO: authoritative, specific, well-structured content that directly answers the questions your audience is asking. Content that earns AI citations tends to share a few traits. It makes clear factual claims, includes specific data points, and is structured so a language model can extract it accurately. Schema markup, FAQ sections and clean entity definitions all help.

None of that is a new tactic. It is the same discipline that earns featured snippets and People Also Ask placements, applied to a broader set of surfaces. If you want the strategic picture rather than the measurement one, our guides to AI search optimisation for UK SMBs and what AEO actually changes versus SEO cover the content side, and our breakdown of which content formats win in AI search shows where to focus first. If the measurement question is really a tracking question, the same principles apply to conversion data: our guide to the Meta Conversions API covers what server-side collection fixes when the browser stops passing signals. And if the measurement comes back showing you are not being cited anywhere, our answer engine optimisation service is the work that changes it. The generative-search half of the same problem sits with our generative engine optimisation service, and our read of the Google AI search optimisation guide sets out why Google treats both as SEO rather than as separate disciplines.

One reporting note if you present these numbers to anyone else. Set the expectation early that AI traffic figures are a directional floor rather than a precise count, and say so every time you show them. The arrival of a native channel makes this harder, not easier, because a tidy default label reads as authoritative and people stop asking what it excludes. Boards and clients anchor hard on the first number they see, and an AI channel that grows from 1.8% to 2.4% is not a story worth telling on its own. The pages being cited, and whether you are cited at all in your category, are.

The measurement tools are still catching up. But the businesses investing in AI search visibility now will be far better positioned as referral volumes grow and the attribution picture clears. If you want to know how your site appears in AI search, or how to build the content infrastructure that earns citations, our SEO team can run an AI visibility audit alongside a standard technical review.

Frequently Asked Questions