AI Traffic in Google Analytics: How to Spot Visits Sent by ChatGPT and Perplexity

🇫🇷 Version française

You may already have visitors coming from ChatGPT or Perplexity in your reports — without knowing it. Most analytics tools file this traffic under "Direct," the catch-all bucket for visits with no identified source. Here's where to look to pull it out of the shadows.

Getting cited by an AI model is great. Knowing that citation actually sent you a visitor is better — and that's often where it breaks down. Unlike a classic Google link, a link clicked from a ChatGPT answer or a mobile app doesn't always pass the origin information Google Analytics expects. The result: that traffic exists, but it's invisible in your usual reports if you don't know where to look.

Why does AI traffic hide inside "Direct"?

Google Analytics classifies a visit based on the information the browser passes at the moment of the click. When that information is missing — because the link was opened from a native app like the ChatGPT mobile app, or because the referrer got stripped along the way — Analytics has nothing to attach the visit to, and defaults it to "Direct," the same bucket as someone typing your URL by hand.

This isn't a configuration mistake on your part: it's standard behavior for a relatively recent traffic source that the tool's default attribution rules don't fully recognize yet.

Which referrer domains should you look for to spot ChatGPT, Perplexity, Gemini, and Copilot?

Some AI traffic does pass a usable referrer, particularly on desktop or through the web versions of these tools. In your traffic acquisition report, look for these domains among your sources:

If none of these domains show up in your current traffic sources, that doesn't necessarily mean you have zero AI traffic: it may also mean it's currently buried in "Direct," or that it's still marginal in volume, which is common for most sites in 2026.

How do you isolate this traffic in your reports?

Once you've identified the domains, two simple habits are enough to start tracking this signal over time, without overhauling your analytics stack:

• Create a segment or a custom channel group that bundles these domains together, so they're tracked as their own category instead of scattered across your "other" sources.
• Watch this segment's trend over time rather than its absolute value at any given moment: the trend tells you whether your GEO work is starting to generate real traffic, not an isolated monthly number.

Stay cautious about interpretation: an unexplained rise in "Direct" traffic (no campaign, no obvious spike in brand awareness) is often, itself, an indirect clue of unattributed AI traffic — without being something you can state with certainty.

What this traffic doesn't replace

For the vast majority of sites, identifiable traffic coming from AI models remains well below classic SEO traffic. That's not a contradiction with the value of GEO: a good share of AI citations generate no click at all, since the answer is given directly to the user without them needing to visit your site. This traffic should be read as a complementary signal — proof that a citation sometimes turns into a visit — not as the single metric for your GEO performance.