How to Track AI Referral Traffic
How to tell an AI referral from ordinary referral traffic in GA4, which domains to watch for, and why the number you get is still a floor, not a ceiling.
An AI referral is a session whose referrer is an assistant domain — chatgpt.com, perplexity.ai, gemini.google.com and the like — logged after someone clicked a citation inside a conversation. GA4 does not label these automatically; you build a custom channel or segment to see them as their own line instead of buried inside "Referral."
That line is one piece of the picture this cluster's SEO reporting guide walks through end to end, from raw channel data to a number a client actually reads.
On this page
What is the difference between a referral and a crawl?
A crawl is a bot — GPTBot, PerplexityBot, ClaudeBot and their siblings — fetching your page's HTML, usually to build a retrieval index or gather training data [1]. It never touches your analytics, because no browser ran, no script fired, and no session was ever created. If you are looking for crawl activity, that is a server log question, not a GA4 question.
A referral is what happens next, sometimes days later: a person reading an assistant's answer clicks a citation and lands on your site with a normal browser session, carrying a referrer header that names the assistant. That session is exactly as trackable as any other referral, provided you go looking for it specifically.
The two get confused constantly because the same conversation — "why is my ChatGPT traffic up" — can mean either one, and the fix is different depending on which it is. A crawl question is answered by checking robots.txt rules and server logs for the right user agent. A referral question is answered inside GA4, and it is the one this guide is about.
| AI crawl | AI referral |
|---|---|
| A bot fetching a page | A person clicking a citation |
| Visible in server logs and robots directives | Visible in GA4 as a session with a referrer |
| Happens whether or not you are ever cited | Only happens after a citation appears |
| No browser, no script, no GA4 event | Ordinary session, just an unusual channel |
Which referrer domains should you actually watch for?
The list is short today and will not stay that way. At minimum: chatgpt.com and chat.openai.com for ChatGPT, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Each assistant can also route a click through a redirect or shortened URL rather than a plain domain referrer, so treat this as a starting list to review quarterly, not a fixed set to configure once and forget.
Desktop browser traffic is the reliable part. A native mobile app is a different story: several assistant apps open links in an in-app browser that strips the referrer header entirely, and GA4 then has nothing to attribute the session to except "Direct." That session did come from an AI citation, it is just invisible as one, and no channel group you build will recover it after the fact. It is one more reason the number you can see is a floor rather than the whole picture.
Some assistants have started appending their own tracking parameters to outbound links — a utm_source naming the assistant, for instance — which is actually good news where it happens, because a UTM parameter survives an in-app browser in a way a bare referrer header does not. Check your raw traffic for query parameters you did not add yourself before assuming every untagged "Direct" session is a dead end.
Pull the current referrer list from raw traffic
Look at GA4's unfiltered referral traffic for domains that do not match your existing channel rules — new assistants show up here before anyone adds them to a guide.
Build a custom channel group
Group the known assistant domains into one channel so they show up on their own report row instead of scattered across "Referral" [3].
Save it as an exploration or segment too
A channel group changes future reports; a segment lets you apply the same definition retroactively to traffic you already collected.
Recheck the domain list every quarter
New assistants and new redirect patterns appear faster than any single guide, this one included, can keep up with.
Why will this number always undercount the real effect?
Because most assistant answers are read and acted on without a click. Someone asks a question, gets a complete answer with your brand named in it, and never opens a new tab — that reader still formed an opinion about you, and no analytics tool anywhere will ever log it. Referral tracking measures the fraction of AI-influenced readers who happened to click through, not the fraction who were influenced.
Add to that the in-app "Direct" leakage above and the redirect links that some assistants use, and you get a number that is real but consistently low. Do not let a small number read as "AI traffic isn't happening for us yet." It is a common early misread, and it usually leads to the wrong next step — waiting, instead of checking a citation rate that would show the effect this referral count cannot fully see.
How does this compare with ordinary organic and paid traffic?
It sits alongside them, not inside them. GA4 for organic traffic already has its own reporting quirks — sessions that never quite match Search Console, a channel definition that groups more than it separates — and AI referral tracking inherits every one of those plus its own: a shorter list of known domains, more redirect noise, and a much smaller current volume that makes month-to-month swings look dramatic when they are really just a small base number moving.
Do not fold AI referral sessions into the Organic Search row to make the report simpler. Keeping the channel separate is the only way to see whether it is actually growing, and a client asking "are people finding us through ChatGPT yet" deserves an answer pulled from its own line, not an estimate buried inside a bigger number.
There is a second reason to keep it separate: the two channels respond to different work. Organic Search moves with technical SEO and content depth. AI referrals move with whether third-party pages — review sites, comparison posts, professional profiles — mention you by name, which is a different task with a different timeline. Merging the two channels in a report merges two different stories about what caused the change, and whoever reads the report loses the ability to tell them apart.
The mistake I see most is treating this like a vanity metric because the volume starts small. It starts small on every channel. The point of tracking it now is having twelve months of trend data by the time the volume stops being small.
What do you do with an AI referral once you have identified one?
Follow it the same way you would follow any other qualified lead: into the CRM, tagged with its actual source rather than folded into a generic "referral" bucket [4]. Most CRMs already capture an original-source property on first contact; the work is making sure your channel definition and the CRM's source value agree on what to call an assistant referral, so a sales conversation and a marketing report are describing the same thing. Tracking SEO leads in a CRM covers the mechanics of that handoff — the short version is that the channel data GA4 gives you is only useful as far as the first click; whether that visitor became a lead is a question only your CRM can answer.
Report the two numbers side by side: sessions from AI referrals, and leads or conversions that came from them. A rising session count with zero downstream leads is worth investigating before it is worth celebrating, and SEO KPIs by stage has the fuller list of which number matters at which point in that funnel.
One practical note on volume: because the base number is still small for most sites, resist reporting it as a percentage change month to month. Going from three sessions to seven is a "133% increase" that means almost nothing on its own and will read as either misleading or absurd to anyone who checks the raw count. Report the actual number, plain, alongside the trend line over several months — that is the version that still makes sense a year from now, once the base number has grown enough for a percentage to mean something.
Questions people also ask
What counts as an AI referral?
A session that arrives with a referrer domain belonging to an AI assistant — chatgpt.com, perplexity.ai, gemini.google.com and similar — because someone clicked a citation inside a conversation. It is a click, not a crawl.
Does GA4 track ChatGPT traffic automatically?
It lands somewhere, usually Referral or Unassigned, but GA4 has no built-in "AI assistants" channel. You have to build a custom channel group or a segment yourself to see it as its own line.
Is AI referral traffic the same as being crawled by an AI bot?
No. A crawl is a bot fetching your page, often for training or for building a retrieval index, and it never shows up in GA4 because no browser ran your tracking script. A referral is a human clicking through afterward.
Why does my AI referral number look too small to be real?
Because a meaningful share of assistant answers are read and acted on without a click at all — someone gets the answer inside the chat and never visits your site. Referral tracking only ever counts the fraction that did click.
Which is more useful to report: referral count or citation rate?
Both, for different audiences. Referral count is the number a revenue conversation wants. Citation rate — how often you are named across a fixed set of prompts — is the leading indicator that predicts where the referral count is headed.
Sources
- Overview of OpenAI crawlers — OpenAI (checked 23 September 2026)
- AI features and your website — Google Search Central (checked 23 September 2026)
- Custom channel groups — Google Analytics Help (checked 23 September 2026)
- Understand Original and Latest traffic source properties — HubSpot Knowledge Base (checked 23 September 2026)
Find out whether AI assistants are already sending you traffic
Book a free 30-minute consultation with Ibrahem and Jeremy. We look at your site while we talk, and you leave with a prioritised plan.
Who wrote this
Jeremy ZavaCo-founder
Leads growth and client partnerships.
Works on: CRO, SEO
Reviewed before publishing by Ibrahem Helal, Co-founder.
- SEO reporting: what a good report showsA guide to SEO reporting: what a monthly report must include, which KPIs matter at each stage, and how to tie organic search to real revenue.
- GA4 for Organic Traffic: A Practical GuideHow to isolate organic search in GA4, why its numbers never match Search Console, and what to actually report each month from the data.
- How to Track SEO Leads in Your CRMHow to tag a lead's original source as organic search, handle multi-touch visitors, and report SEO performance in terms your CRM already tracks.
- SEO KPIs to Track by Funnel StageWhich SEO KPIs actually matter at each stage of the funnel, from first impression to closed revenue, and which ones are just noise dressed up as data.