← ClickRadius Institute

Tracking AI Referral Traffic in Google Analytics

ClickRadius Institute · April 21, 2026

The first thing anyone does after hearing that AI engines send traffic is open analytics to look for it — and the first thing they discover is that it is harder to find than it should be. AI referral traffic in GA4 is real, it is growing, and it is also systematically undercounted by the tool that is supposed to measure it. This guide is the practical, technical version: the exact source domains to watch, how to build a channel group that isolates them, the regular expressions that hold up, the UTM hygiene that reduces leakage, and the honest limits of what analytics can and cannot see. Getting this right matters because AI referrals are the one layer of AI-search performance that produces a literal, defensible tracked session — the trailing evidence that anchors the softer measures.

How AI referrals actually appear in GA4

When a user clicks a source link inside an AI answer and the engine passes a referrer, GA4 records a session whose source is the AI product's domain and whose default channel, unhelpfully, is usually Referral — the same bucket as any other website linking to you. It does not arrive pre-labeled as “AI.” That means the raw signal is present but unaggregated: ChatGPT sessions, Perplexity sessions, and Gemini sessions sit scattered among your ordinary referrers until you deliberately group them.

The domains worth watching, as of this writing, are:

This list moves. Engines add domains, rename products, and change how they render source links, so treat any list as a snapshot to be revisited quarterly rather than a permanent configuration.

Building an AI channel in GA4

You have two practical routes, and serious measurement uses both.

Route 1: a custom channel group

In GA4's admin, create a custom channel group and add a channel named AI Assistants ordered above the default Referral channel. Define it by matching the session source against a regular expression covering the hosts above, for example: chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|claude\.ai|grok\.com|x\.ai|copilot\.microsoft\.com. Ordering matters — GA4 evaluates channel rules top-down and assigns the first match, so the AI channel must sit above generic Referral or your AI sessions will be swallowed by it.

Route 2: an exploration segment

Channel groups apply going forward and cannot always be backdated cleanly, so for historical analysis build a session or user segment in Explorations using the same source regex. This lets you compare AI-sourced sessions against your other channels on engagement rate, conversions, and revenue over any date range you have data for. The segment is also where you answer the questions that matter commercially: do AI-referred visitors convert at a different rate, land on different pages, and carry different intent than search or social visitors?

A common finding worth anticipating: AI-referred sessions are frequently lower in raw volume but higher in intent, because the user arrived after an engine effectively pre-qualified them with a recommendation. Do not judge the channel on session count alone; judge it on conversion quality.

The dark-traffic problem, stated honestly

Here is the part most tracking guides skip. The number you build with the steps above is a floor, not a full count, and the gap between the floor and reality can be large. Two forces suppress it.

First, zero-click answers. A rising share of AI interactions — industry estimates put zero-click searches near half of all searches heading into 2026, and higher still inside AI answer surfaces — end with the user's need met and no click issued. No click means no session, which means nothing to track. The influence happened entirely inside the engine.

Second, referrer loss. Of the clicks that do occur, some arrive with a stripped or empty referrer. This happens when an engine omits the referring header, when traffic comes from a mobile app rather than a browser tab, or when a privacy setting or intermediate redirect drops the referrer. GA4 files these sessions under Direct or Unassigned, not under your AI channel, even though an AI answer caused them.

Analytics measures the clicks that survive to your server. In AI search, most of the influence never becomes a click, and some of the clicks that do arrive shed the evidence of where they came from. The referral report is the visible tip; the mechanism is mostly below the waterline.— ClickRadius Institute

The practical consequence: never present GA4 AI referrals as the size of your AI-search performance. Present them as the trackable minimum, and triangulate with the other layers.

UTM hygiene and the limits of tagging

UTM parameters are the classic fix for referral ambiguity, and they help — but AI search constrains them. You can only tag links you control. When an engine autonomously renders a citation to your homepage, there is no opportunity to append a UTM; the engine generates the link, not you. Where UTMs genuinely help is on the surfaces you do control that feed AI: the profiles, directory listings, and syndicated content that engines crawl and sometimes link through. Tag those consistently so that when they are the path an engine takes, the visit self-identifies.

Two hygiene rules prevent avoidable loss regardless of AI. Keep redirects short — every hop is a chance to drop the referrer. And audit your own outbound and cross-domain links so that internal movement does not overwrite the original AI source in the session. Sloppy link plumbing turns trackable AI sessions into direct traffic before you ever see them.

One more source of distortion is worth a deliberate check: bot and crawler traffic. The same AI engines that cite you also crawl you, and crawler hits are not human visits — GA4's standard bot filtering catches much of this, but not all of it, and an unfiltered spike of crawler activity can masquerade as a referral surge that never corresponded to a real reader. Before you report a jump in AI-adjacent traffic, sanity-check it against engagement signals: real human sessions from an AI recommendation show normal engagement time and page depth, while crawler noise does not. This is also why server log analysis is a useful complement to GA4 for the AI era — logs show you which AI crawlers are reaching your pages at all, which is a prerequisite for being cited in the first place and a diagnostic that analytics alone cannot provide. Reading the referral report alongside your logs separates the two questions that matter: are the engines crawling me, and are the people they influence coming through.

What to actually report from GA4

Once the channel and segment exist, resist the urge to over-report a small number. A disciplined AI-referral report contains four things:

  1. AI channel sessions over time, shown as a trend, not a single figure — the slope is the story.
  2. Engagement and conversion rate of AI sessions versus your site average, to demonstrate intent quality.
  3. Top landing pages from AI sources, which reveal exactly which of your pages engines are citing and sending readers to — a direct input into your content roadmap.
  4. The honesty caveat, stated in the report itself: this is the trackable floor, and total AI influence is larger and measured elsewhere.

That last line is not a weakness in the report; it is what makes the rest of the report credible. According to the layered-measurement approach used across the Institute library, GA4 referrals are the trailing, trackable layer beneath citation share and citation-created signals like branded-search lift and self-reported attribution.

Connecting GA4 to the fuller picture

Because analytics sees only the surviving clicks, the complete measurement stack pairs GA4 with two things it cannot provide. The first is citation monitoring — sampling your priority questions across the five engines to see how often you are named and cited, which measures the influence that never becomes a click. The second is self-reported attribution — training your intake to log when a customer says an AI recommended you, which recovers the revenue impact that analytics filed under direct. GA4 tells you about the visitors who clicked; citation monitoring tells you about the answers that recommended you; intake data tells you about the buyers who acted without clicking. All three together are the measurement; any one alone is a fragment.

The teams that measure AI search well stop asking analytics to do a job it structurally cannot do. GA4 counts the clicks; it was never built to count the recommendations, and in an answer engine the recommendation is the product.— ClickRadius Institute

Frequently asked questions

Which referral sources indicate AI traffic in GA4?

Watch for the AI product domains that pass a referrer: chatgpt.com and chat.openai.com for ChatGPT, perplexity.ai for Perplexity, gemini.google.com for Gemini, claude.ai for Claude, grok.com and x.ai for Grok, and copilot.microsoft.com or bing.com for Copilot. Build a custom channel group or an exploration segment matching these hosts with a regular expression so they roll up into one AI channel. Coverage is partial: some engines strip the referrer, and app-based sessions often arrive with no referrer, so the referral report is a floor on AI influence, not a full count.

Why is my AI referral traffic lower than the influence I can see in AI answers?

Because most AI influence is dark to analytics. A large and rising share of AI answers resolve the question without a click, so no session is created at all. Of the sessions that do occur, some arrive with a stripped or blank referrer and land in direct or unassigned rather than in your AI channel. The referral number therefore undercounts real influence, sometimes dramatically. Treat GA4 AI referrals as the trackable floor and triangulate with citation-share monitoring, branded-search lift, and self-reported attribution captured at intake.

How do I keep AI referral traffic from being miscategorized as direct traffic?

You cannot fully prevent it, because the miscategorization happens when the engine omits the referrer before the visit reaches you, which is outside your control. What you can do is reduce avoidable loss: keep tracking on your outbound and internal links clean, avoid redirect chains that drop the referrer, and where you can influence a link an engine renders, add UTM parameters so the visit self-identifies. Then monitor direct and unassigned traffic for correlated lift when your citation share rises, since a portion of that increase is unattributed AI influence.

Analytics is only one layer. See how many of your priority questions the AI engines actually answer with your brand by running a free AI Readiness Score, and explore full five-engine citation monitoring on the pricing page.

Related from the Institute