How to track AI search traffic in your analytics
Visits from ChatGPT, Perplexity, and AI Overviews often hide as direct traffic. Here is how to surface them in GA4 and what the data still can't tell you.
You optimized a page, it started getting cited by ChatGPT and Perplexity, and now you want to know if that citation sent anyone to your site. Most analytics setups can't answer that question cleanly. AI referral traffic hides in the gaps between how these tools link out and how web analytics tools were built to track links. Here is how to close that gap as much as it can be closed today, and where you have to accept a blind spot.
Why AI referrals look different from search clicks
A Google search click carries a referrer and often a set of URL parameters your analytics tool knows how to parse. An answer from an AI assistant is a different shape of event. Some tools link out with a clear referrer domain. Some open links in an in-app browser that strips or rewrites the referrer. Some answers get read and never clicked at all, so there is no session to measure in the first place. The result is that a chunk of AI-driven traffic lands in your analytics as direct traffic: no source, no medium, no story.
This isn't a bug you can patch on your own site. It's a structural limit of referrer-based tracking meeting a new kind of client. The fix is to get better at spotting the traffic that does carry a signal, and to stop expecting a complete picture from the traffic that doesn't.
Finding AI traffic in GA4
Start with the referral traffic you already have, before building anything custom. In GA4, go to the Traffic acquisition report and look at Session source / medium. Most AI tools that do pass a referrer will show up as a domain under "referral," separate from your organic search and direct buckets.
- chatgpt.com and openai.com for ChatGPT-driven visits
- perplexity.ai for Perplexity
- copilot.microsoft.com for Copilot
- gemini.google.com for Gemini
- www.bing.com combined with a chat context, which is harder to isolate from regular Bing search
Once you can see these domains, build a custom channel group (or a saved exploration) that buckets them under an "AI referral" label instead of leaving them scattered across generic referral rows. This is mechanical work, not a one-time setup: new assistants launch, and existing ones sometimes change how they link out, so revisit the list every quarter.
Tag your own links where you control them
If you publish content on other platforms and link back to your site, or if you can control how a partner's chatbot links to you, add UTM parameters. That single change turns an ambiguous referral into a clean, attributable session. You cannot force a third-party AI engine to preserve your UTM tags, but you can make sure every link you place yourself carries one.
What the data still can't tell you
Be honest with whoever you report this to: what you're building is a partial view, not a full attribution model. A session tagged as chatgpt.com tells you a visitor came from a link in a ChatGPT response. It does not tell you what question they asked, whether your page was quoted directly or just linked as a source, or how many people read an answer that cited you and never clicked at all. Those numbers aren't hidden somewhere waiting to be extracted with a better report; they mostly aren't captured anywhere on your end.
Server log analysis fills in one piece of this. Your logs will show crawler visits from the bots these companies use to fetch pages for their answers, which is a leading indicator that your content is in the pool a model can draw from. A crawl is not a citation and a citation is not a click, so treat log data, referral data, and citation testing (manually asking the tools questions in your topic area) as three separate signals, not one funnel.
Treat AI referral tracking as a sampling problem, not a measurement problem. You're seeing a slice, not the whole picture.
A simple process to run monthly
- Pull the AI referral channel group and note volume and top landing pages
- Check server logs for crawl frequency from known AI crawlers on those same pages
- Manually query ChatGPT, Perplexity, and Google AI Overviews on your core topics and record whether your site is cited
- Compare month over month, not day over day; this traffic is low-volume and noisy for most sites
None of this replaces your existing organic search reporting. It sits next to it, as a rough, improving instrument rather than a finished dashboard. The tools for measuring AI-driven traffic are still catching up to the traffic itself, and any vendor promising a complete, precise breakdown of AI referral volume today is overstating what referrer data can actually show.
Start small: set up the custom channel group this week, add UTMs to every link you control, and check your logs once a month for crawler activity on your most-cited pages. That gives you a real, if partial, read on whether your AEO work is landing, without pretending the picture is more complete than it is.