---
title: "How to Track ChatGPT Referral Traffic in Your Analytics | Kitbase Blog"
description: "See the visitors ChatGPT sends you: how chatgpt.com referrals and utm_source=chatgpt.com appear in analytics, how to segment them, and why a large share arrives as untraceable Direct."
canonical: https://kitbase.dev/blog/track-chatgpt-referral-traffic/
---

**To track ChatGPT referral traffic, filter your analytics for the referrer `chatgpt.com` (and the legacy `chat.openai.com`), plus the `utm_source=chatgpt.com` tag ChatGPT now appends to many of its outbound links.** Sessions that arrive with either signal are people who clicked a link inside a ChatGPT answer. Group them into their own segment and you can finally see a channel that most dashboards bury under "Referral" or, worse, misfile as "Direct."

That last caveat is the whole story. ChatGPT referral traffic is real, growing, and high-intent — but a large fraction of it arrives with **no referrer at all**, which means referrer-based tracking captures only the visible slice. This guide covers how to segment the traffic you *can* see, exactly what the referrer data does and doesn't tell you, and how to account for the part that goes dark. It's one stage of the broader [GEO funnel](/blog/geo-funnel/): crawl, citation, then the referral click you're trying to measure here.

## How ChatGPT referrals show up

When someone clicks a citation or link inside a ChatGPT answer on the web, the browser sends your site a referrer, and ChatGPT increasingly stamps the destination URL with a campaign parameter. In practice you'll see two signals:

| Signal | Value | Where it appears |
|---|---|---|
| Referrer | `chatgpt.com` (or legacy `chat.openai.com`) | Referrer / traffic-source dimension |
| UTM source | `utm_source=chatgpt.com` | UTM source dimension (on links where ChatGPT adds it) |

The referrer is the reliable web signal; the UTM tag is a bonus that ChatGPT applies to *some* outbound links, not all of them. Either one is enough to attribute a session to ChatGPT — which is why you segment on both.

## Segmenting ChatGPT traffic in Kitbase

Kitbase captures both **referrer** and **UTM source** as first-class dimensions — enriched server-side and available in every breakdown and filter, as documented in the [analytics dashboard guide](https://docs.kitbase.dev/guide/analytics). Two moves surface your ChatGPT channel:

1. **Break down by Referrer.** Set the breakdown dimension to Referrer and `chatgpt.com` appears as its own row, with sessions, landing pages, and everything else you can pivot on. To include the older domain, filter Referrer to match `chatgpt.com` or `chat.openai.com`.
2. **Or filter by UTM source.** Set a filter of UTM source equals `chatgpt.com` to catch the links ChatGPT tagged — useful because a UTM survives some cases where the referrer is dropped.

From there, drill in. Which landing pages does ChatGPT send people to? (Usually the pages it cites — the ones your [crawler data](https://docs.kitbase.dev/crawler-detection) shows the bots read most.) Do those visitors trigger your key events? Because Kitbase's referrer and UTM fields are also filter dimensions, you can scope funnels, journeys, and event breakdowns to "sessions from ChatGPT" and compare their behavior against your other channels.

For a stable, always-on view, treat "referrer contains `chatgpt.com` OR utm_source is `chatgpt.com`" as your ChatGPT segment and save it. The companion post on [Perplexity, Gemini, and Copilot referrals](/blog/track-perplexity-gemini-referrals/) extends the same pattern into a single AI-traffic segment across every engine.

## What referrer data can and can't tell you

This is where honesty matters more than a tidy dashboard. Referrer-based attribution has a hard ceiling.

| Referrer data **can** tell you | Referrer data **can't** tell you |
|---|---|
| That a session came from ChatGPT's web interface | Which prompt the user typed |
| The landing page ChatGPT linked to | Whether you were *cited* as a source or merely *mentioned* |
| Behavior after the click (events, conversions) | Anything about sessions that arrived without a referrer |
| Volume trend of visible ChatGPT clicks | Your true total ChatGPT-driven traffic |

The two "can'ts" on the right are the important ones. Referrer data tells you a click happened; it says nothing about the *answer* that produced it — whether ChatGPT recommended you, cited you, or your competitor. That question belongs to a different dataset: [AI Visibility](https://docs.kitbase.dev/ai-visibility) measures how often the engines mention and cite your brand by querying them directly, which is the citation stage that *feeds* these referral clicks.

## The part that goes dark

The biggest limitation isn't in the table above — it's the traffic that never reaches it. A substantial share of ChatGPT referrals arrive with **no referrer**, and land in your reports as ordinary **Direct** traffic, indistinguishable from someone typing your URL. It happens for concrete reasons:

- **The ChatGPT mobile app** frequently opens external links without passing a referrer header. On mobile, the click is real but the origin is stripped.
- **Copy-paste.** When someone copies a URL out of an answer and pastes it into a browser, there's no referrer by definition — the visit is Direct.
- **In-app browsers and privacy settings** strip or shorten referrers in ways you don't control.

Industry analyses of AI referral traffic put the share arriving without a referrer anywhere from roughly a third to well over half of AI-driven sessions — the exact figure varies by audience and device mix, but the direction is consistent: **referrer-based tracking undercounts ChatGPT, sometimes badly.** This is the "[AI dark traffic](/blog/ai-dark-traffic-attribution/)" problem, and it's why a rising, unexplained Direct segment on a site that's gaining AI visibility is often ChatGPT traffic in disguise.

**One ChatGPT click, two fates in your analytics**

```mermaid
flowchart TD
  A["User clicks a link<br/>in a ChatGPT answer"] --> B{"How was it opened?"}
  B -->|"Desktop web"| C["Referrer: chatgpt.com<br/>(often utm_source too)"]
  C --> V["Attributed to ChatGPT ✓"]
  B -->|"Mobile app /<br/>copy-paste"| D["No referrer"]
  D --> W["Filed as Direct ✗<br/>(dark traffic)"]
```

### Reducing the blind spot

You can't fully close the gap, but you can shrink it and reason about it:

- **Watch Direct alongside your visible ChatGPT segment.** When crawler data and [AI Visibility](https://docs.kitbase.dev/ai-visibility) show you're gaining ground in ChatGPT, a correlated rise in Direct is a strong tell that dark AI traffic is growing too.
- **Use your own UTMs on links you control** (docs, newsletters, profiles) so that when ChatGPT cites *those* URLs, attribution survives even a stripped referrer.
- **Segment by landing page.** Direct traffic hitting deep, specific pages — a niche comparison or docs page, not your homepage — rarely comes from someone typing the URL. It's a fingerprint of AI and other referral sources that lost their referrer.

Note that platforms are slowly improving here — Google added a native AI-traffic channel to GA4 in 2026, and ChatGPT's UTM tagging is newer than the referrer itself — but all of it still depends on a signal surviving the click. The dark share is a property of how links are opened, not a bug any one tool can fully fix.

## FAQ

**How does ChatGPT traffic appear in analytics?**
As referrer `chatgpt.com` (or the legacy `chat.openai.com`), and — on many links — with `utm_source=chatgpt.com` attached. Segment on either signal to isolate the channel.

**Why is some ChatGPT traffic showing up as Direct?**
Because the referrer was stripped. The ChatGPT mobile app often opens links without a referrer, and copy-pasted URLs never have one. Those visits look identical to someone typing your address, so they fall into Direct.

**Does ChatGPT add UTM parameters to links?**
Yes, to many outbound links ChatGPT now appends `utm_source=chatgpt.com`. It's not on every link, so don't rely on it alone — pair it with the referrer.

**Can I see which prompt sent someone to my site?**
No. Referrer and UTM data confirm the click came from ChatGPT but reveal nothing about the question asked or whether you were cited versus mentioned. That requires querying the engines directly, which is what AI Visibility does.

**Is ChatGPT referral traffic worth tracking if it undercounts?**
Yes — the visible slice is still a real, high-intent channel and its trend is meaningful. Just read it as a floor, not a total, and pair it with crawler and visibility data to see the full picture. Whether those clicks convert is its own question, covered in [do AI referrals convert better](/blog/do-ai-referrals-convert-better/).

---

*Want to segment ChatGPT traffic and watch it convert — without a cookie banner? [Start your free trial](https://app.kitbase.dev/signup/) — 7 days, no credit card required — and break down every channel, including the AI ones, by referrer and UTM.*