- AI assistants already send measurable referral traffic, but GA4 does not group it for you — the visits arrive under raw hostnames like chatgpt.com and perplexity.ai scattered across your referral report.
- A channel group plus an exploration that filters referral sources to the known AI hosts turns that scattered data into one readable 'AI assistants' number you can track over time.
- This measures the click-through after an answer, not whether you were cited — so pair it with answer-level visibility tracking to see the full picture, since many AI mentions never produce a click at all.
AI assistants are already sending you traffic. When ChatGPT, Perplexity or Gemini cites your page and the reader clicks through, that visit lands in your analytics like any other — except GA4 does not tell you it came from an AI assistant. Instead it scatters those visits under raw hostnames, buries some in other channels, and leaves you with no single number for "how much traffic are the AI tools sending me." This guide fixes that, so you can see the channel clearly, and then it draws the honest line around what the number actually means.
The fix is not a plugin or a paid add-on; it is a bit of configuration you do once. But before the how, it is worth being precise about the what, because a misread AI-traffic number leads to exactly the wrong conclusions — both the false panic of "AI sends us nothing" and the false comfort of "we're clearly being cited." Both are avoidable once you know what GA4 is and is not measuring.
Why GA4 hides your AI traffic
GA4 sorts incoming visits into channels — Organic Search, Direct, Referral, and so on — using rules it ships with. Those rules predate the assistants, so there is no "AI Assistants" channel and no logic that recognises chatgpt.com or perplexity.ai as a category. The result is that a click from an AI tool is treated as a plain referral from whatever hostname sent it. One day's AI traffic might be spread across chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, each a separate line in your referral report, none of them summed, and some potentially mis-bucketed elsewhere depending on how the link was formed.
So the data exists and is reasonably accurate at the per-hostname level; what is missing is the grouping. Your job is simply to teach GA4 that this set of hostnames belongs together and deserves to be read as one channel. Do that and a previously invisible source becomes a line you can watch rise or fall like any other.
Setting it up: two approaches
There are two practical ways to see AI traffic as a group in GA4, and they are complementary. One is quick and needs no configuration; the other is durable and shows up everywhere.
The quick way: an exploration with a filter. In GA4's Explore section, build a free-form exploration using "Session source" as the dimension and your sessions and conversions as metrics, then add a filter that matches the known AI hostnames — session source containing chatgpt, perplexity, gemini, copilot, or claude. That single report sums all of them into a readable table you can save and revisit. It is the fastest way to get a number today, and it is reversible, so it is a safe first step.
The durable way: a custom channel group. GA4 lets you define a custom channel group with your own rules. Create one that routes sessions whose source matches your AI hostnames into a new channel named "AI Assistants." Once defined, that channel appears in your standard acquisition reports alongside Organic and Referral, so the grouping is applied everywhere automatically rather than living in one saved exploration. This is the version worth setting up properly, because it makes AI traffic a permanent, first-class line in your reporting.
Keep your hostname list in one place. Whichever approach you use, write down the exact list of AI hostnames you are matching and treat it as a living document. New assistants launch, domains change, and a referral you do not recognise today may be tomorrow's major source. Reviewing the list every few months keeps your "AI Assistants" number honest rather than slowly drifting out of date.
What the hostnames are, and keeping the list current
A good starting set covers the assistants most likely to send you traffic: chatgpt.com and its older chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. Depending on your audience you may also see smaller or regional assistants, and you should add any you spot in your own referral report that clearly originate from an AI tool. The point is to match on the stable part of each hostname so variations still get caught, and to revisit the set periodically rather than treating it as fixed.
One subtlety worth knowing: not every AI mention produces a referral your analytics can see. Some assistants open links in ways that strip or alter the referrer, and some surfaces — particularly answers rendered inside a search results page — may attribute the click to the search engine rather than to an AI source. So your grouped number is a solid lower bound on AI-driven clicks, not a perfectly complete census. That is fine for tracking a trend, as long as you remember it is a floor rather than the whole truth.
A worked walkthrough: your first AI traffic report
To make this concrete, here is the quick exploration end to end, the way you would actually do it. Open GA4 and go to Explore, then start a blank free-form exploration. Set the dimension to "Session source" and bring in "Sessions" and whichever conversion or key-event metric you care about — a sign-up, a trial start, a contact. At this point you are looking at every source, which is noise. So you add a filter: session source matches a pattern containing chatgpt, perplexity, gemini, copilot or claude. Now the table collapses to just your AI sources, summed into a handful of rows, with sessions and conversions beside each.
The first time you run it, two things usually surprise people. One is that there is more AI traffic than expected — it was simply hidden by being split across hostnames. The other is that the conversion quality is often strong: these visitors asked a question, got an answer that named you, and chose to come and learn more, so they tend to arrive warmer than a cold search click. Save the exploration with a clear name so you can return to it, and you now have a baseline you can compare against next month. That is the entire quick path, and it costs about ten minutes.
When you later build the custom channel group, the experience changes from "a report I open" to "a line that is always there." Your standard acquisition reports begin showing "AI Assistants" next to Organic Search and Direct, which means anyone who looks at your traffic sees it without needing to know the exploration exists. For a team, that visibility is the real win: AI stops being a thing one analyst tracks in a saved report and becomes a channel the whole business can see and discuss.
Watch for AI traffic hiding in Direct
There is a quieter measurement trap worth naming: some AI-driven visits never carry an AI hostname at all, and land in Direct instead. This happens when an assistant's app or interface opens your link in a way that drops the referrer, or when someone reads an answer, does not click, and later types your name into their browser because the answer put you on their radar. The first case is invisible to your hostname filter; the second is genuinely attributable to AI but will forever look like Direct or Organic, because the visit that AI influenced happened later and through another door.
You cannot fully untangle this, and you should not try to force it, but you can account for it. A useful habit is to watch your Direct and branded-search traffic for lifts that coincide with rises in your citation visibility. If your presence in AI answers climbs and, a little later, your branded search and direct visits climb too, it is reasonable to read some of that as the downstream effect of AI mentions that never produced a tracked click. It is not precise attribution, but it keeps you from undercounting AI's real influence, which — as the next section explains — is the central risk in measuring this channel.
The limitation you must not forget
Here is the single most important thing in this guide, and the thing most "track AI traffic" advice leaves out: referral traffic measures clicks, not citations. The entire design of an AI answer is to satisfy the user inside the answer. Very often a model will quote your page, name your brand, and resolve the person's question completely — and they never click, because they got what they needed. That is a successful citation that produces zero referral traffic. If you judge your AI visibility solely by the GA4 number, you will systematically undercount your actual presence in AI answers, sometimes badly.
A low referral number is not proof you are not cited. It can mean you are rarely mentioned, or it can mean you are mentioned often but the answers are self-contained enough that readers do not click. These are opposite situations with opposite responses, and referral data alone cannot tell them apart. Reading "not cited" from a low click count is one of the most common and costly mistakes in AI measurement.
This is why serious AI visibility measurement has two halves. The GA4 referral number tells you about the clicks you earn when an answer sends someone to you — a real and useful signal, especially its trend over time. But to know whether you are actually being cited, you have to look at the answers themselves: run your buyer questions through the assistants and record which ones mention you, whether they link you, and who gets cited instead. That answer-level view is what fills the gap the click data cannot. Used together, the two give you both the visibility (are we in the answers) and the conversion (do those answers send us traffic).
How to read the two numbers together
Once you have both, the combinations are genuinely informative. High citation visibility with low referral traffic tells you the assistants trust you but their answers are satisfying readers without a click — often a sign your content is being summarised well, and a cue to make sure the answer leaves a reason to visit. High referral traffic relative to your citation visibility suggests the mentions you do get are compelling enough to pull clicks, which is a good position to scale. Low on both is the clearest signal that the visibility work itself needs attention, starting with the fundamentals in our playbook for getting cited by AI engines.
Watching the trends is what makes this actionable. A rising AI-assistants channel after you shipped a batch of citability improvements is direct evidence the work is paying off in real visits. A flat referral line alongside rising citations tells you the opposite — you are winning the mention but not the click — which points you at your titles, your opening lines, and whether your answers give a reader a concrete reason to come to the source. Neither insight is available from a single number, which is the whole argument for measuring both.
Turning the number into decisions
A metric you only look at is a waste; the point of grouping AI traffic is to change what you do. Once the channel is visible, three practical decisions open up. The first is prioritisation: segment your AI referral traffic by landing page and you learn which of your pages the assistants actually send people to. Those pages are your proven AI performers, and they tell you what kind of content earns the click — which is the template for what to make more of. The second is content triage: pages that get cited but rarely send a click (visible in the gap between your citation tracking and this referral report) may need a sharper reason-to-visit, while pages that get neither need the citability work itself.
The third decision is reporting honesty. When someone asks "is this AI stuff actually doing anything," you can now answer with a trend line instead of a shrug. Even a small but steadily rising AI-assistants channel is a real, defensible answer, and it reframes the conversation from speculation to evidence. That matters internally, because the work of improving AI visibility competes for time against everything else, and a visible, growing number is what justifies continuing to invest in it. Without the measurement, AI visibility work is an act of faith; with it, it is a channel with a trajectory you can point to.
It is worth setting a cadence rather than checking obsessively. Because the absolute numbers are still modest for most sites, daily readings are noise. A monthly look at the channel — total sessions, conversions, top landing pages, and the trend against prior months — is the right rhythm while AI is an emerging source. Pair that monthly referral read with your monthly citation-visibility read, and you have a complete, low-effort AI reporting habit that catches real movement without drowning you in daily fluctuation.
Getting started this week
You do not need to do all of it at once. A sensible order: first build the quick exploration so you can see your current AI referral traffic today and establish a baseline. Then, when you have a spare half hour, set up the custom channel group so the grouping becomes permanent and shows in your standard reports. Finally, close the measurement gap by adding answer-level visibility tracking, so you are no longer inferring citations from clicks but seeing them directly. Each step is useful on its own, and together they turn AI from an invisible, unmeasured channel into one you can report on and improve like any other.
See the half GA4 cannot show you. GA4 tells you about the clicks; AutoSEO Stack tells you about the citations — it runs your buyer questions across ChatGPT, Perplexity, Gemini and Google's AI surfaces every week and shows exactly which answers mention you and which cite a competitor. Check your AI visibility free and pair it with your referral data for the full picture.
AI referral traffic is real, measurable, and almost always high-intent, but GA4 will not hand it to you grouped — you have to build the channel yourself. Do that, and you gain a clear view of the visits AI answers send you. Just hold the limitation firmly in mind: the click count is the conversion half of the story, not the visibility half. Measure both the clicks you earn and the citations you win, read them together, and you move from guessing about AI's impact to tracking it with the same rigour you bring to every other channel.
Frequently asked questions
Why is my AI traffic not showing up in GA4 by default?
Because GA4 has no built-in 'AI assistants' channel. Clicks from ChatGPT, Perplexity, Gemini and the others arrive as ordinary referrals under their raw hostnames — chatgpt.com, perplexity.ai, gemini.google.com and so on — and get scattered through your referral report or lumped into other channels. The traffic is there; GA4 just does not label it as a group until you tell it to.
Which hostnames should I include as AI sources?
The main ones to start with are chatgpt.com (and chat.openai.com), perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. Keep the list as a living thing — new assistants and changed domains appear over time, so review it every few months and add sources you see in your referral report that clearly come from an AI tool.
Does this tell me whether AI is citing my content?
No, and this is the key limitation. Referral traffic only counts people who clicked through to your site after an answer. Many AI answers mention or quote you without the reader ever clicking, so a low referral number does not mean you are not being cited — it may mean you are being cited but the answer satisfied the user without a visit. To see citations themselves you need answer-level visibility tracking, which is a different measurement.
Is AI referral traffic worth tracking if the numbers are still small?
Yes, because the trend matters more than the absolute number right now. Even modest AI referral traffic tends to be high-intent — these are people who asked a question, got an answer that mentioned you, and chose to come and learn more. Watching that number grow over time tells you whether your AI visibility work is translating into real visits, which is exactly the feedback loop you want while the channel is still emerging.