How to Track ChatGPT Traffic in GA4 (and Find Out Which Companies It Sends You)

When someone clicks a link in ChatGPT’s web interface, the visit arrives at your site with a chatgpt.com referrer, and ChatGPT frequently appends a utm_source=chatgpt.com parameter to the URL as well. GA4 records the session, but its default channel groups have no "AI" category, so the session is filed under Referral, one row among your other referring sites, unless you go looking for it.

That is the whole problem this tutorial solves: the data is already in your property; GA4 just does not surface it as a category. Below are three levels of setup, from a thirty-second check to a permanent channel, followed by the caveats that keep the numbers honest and the step GA4 cannot do at all: telling you which companies that traffic represents.

One naming note: ChatGPT moved from chat.openai.com to chatgpt.com in 2024, so older guides reference the old domain. If your property has long-range data, include both in any filter.

Before building anything, confirm you actually receive AI referral traffic:

1. In GA4, go to Reports, then Acquisition, then Traffic acquisition.
2. Click the primary dimension dropdown (it defaults to Session default channel group) and switch it to Session source / medium.
3. In the search box above the table, type "chatgpt".
4. Repeat the search for "perplexity", "claude", "gemini", and "copilot" to see other AI sources.

If rows appear, note the exact source strings you see, character for character. You will need them in the next step, because GA4 regex matching is case-sensitive and vendors’ referrer strings vary.

For a reusable report across all AI sources, use GA4’s Explore section:

1. Go to Explore and create a blank exploration.
2. Import dimensions: Session source, and Landing page + query string. Import metrics: Sessions, Active users, Engagement rate, and Key events.
3. Set Session source as the row dimension and add your metrics as values.
4. Add a filter: Session source, condition "matches regex", with a pattern such as: chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com
5. Adjust the pattern to the exact source strings you found in Step 1, since GA4 regex is case-sensitive.

The Landing page + query string dimension earns its place: it shows which pages the assistants are citing, and any utm_source parameters ChatGPT appended survive in it. If your docs or comparison pages dominate the landing list, that tells you what the models think you are for.

Explorations are for analysis; channel groups put AI traffic on the same footing as Organic Search and Email in your standard reports. GA4 supports custom channel groups (Admin, then Data display, then Channel groups) where you can define a new channel:

1. In Admin, open Channel groups and create a new channel group (or copy the default one).
2. Add a channel named "AI" with the condition: Session source matches the same regex you validated in Step 2.
3. Drag the AI channel above Referral in the order, since channel rules apply top-down and Referral would otherwise capture these sessions first.
4. Save, then use this channel group as the dimension in Traffic acquisition reports.

Two housekeeping notes. Custom channel groups apply from creation onward in standard reports, so build this before you need the trend line. And revisit the regex quarterly: new assistants appear, and their referrer strings only show up in your Step 1 audit.

Whatever number this setup produces, it undercounts, for reasons worth understanding before anyone puts the chart in a board deck.

• App clicks can lose the referrer. Links opened from ChatGPT’s mobile or desktop apps may arrive with no referrer header and no UTM, landing in Direct, indistinguishable from a typed URL.
• Consent and blockers apply. Sessions where analytics storage is denied never reach the report at all.
• Regex drift. AI vendors change domains (chat.openai.com to chatgpt.com being the canonical example) and your filter silently goes stale.
• Answers without clicks. The largest share of AI influence never produces a session: the assistant answers, the user never visits. No analytics tool sees that.

You will find dramatic percentages online for how much AI traffic hides in Direct. We are deliberately not quoting one: the measured share varies by site and methodology, and this category already has enough unsourced numbers. The direction is what matters: GA4 shows you a verifiable minimum.

Everything above tells you that ChatGPT sent traffic, which pages it landed on, and whether it engaged. For a B2B team, the next question is the one that decides whether any of this matters: who was it? A hundred anonymous AI-referred sessions could be students, competitors, or five buying committees researching you. GA4, by design, cannot say.

Company-level visitor identification can, for the identifiable share of traffic. It matches the visitor’s network to a registered business via IP intelligence, and when that match is joined to the referrer, the anonymous row becomes an account-level fact: someone at a named company arrived from ChatGPT and read your pricing page.

The usual honesty applies: identification resolves office networks well and residential, mobile, and VPN traffic poorly, so you see a sample of your AI traffic, not all of it. But a named sample beats an anonymous total when the goal is pipeline rather than a chart.

PageFox does the joining for you. It identifies visitors at the company level, keeps the referrer context, and scores the resulting leads with transparent signals, so "chatgpt.com session" becomes "target-account visit worth a Slack alert." It runs in passive mode with no visible widget if you only want the intelligence layer, and the script install takes about five minutes next to your GA4 tag.

If you also run the chat widget, there is a second payoff: visitors arriving from an AI assistant tend to arrive mid-question, and a widget that answers from your own content, grounded with sources, is a natural next step for them, and the cleanest way for an AI-referred visitor to turn themselves into a named lead.

Set up the GA4 floor first; it is free and takes an afternoon. Then, when the trend line starts moving, add identification so the line comes with names on it.