Tracking Perplexity, Gemini, and Copilot Referral Traffic

Perplexity, Gemini, and Copilot each send referral traffic under their own domain — perplexity.ai, gemini.google.com, and copilot.microsoft.com — but they don’t behave the same way, and none of them tag links as consistently as you’d like. Perplexity passes a referrer reliably on the web and strips it in its mobile app; Gemini is steady on the web but easy to confuse with Google’s AI Overviews; Copilot has changed its referrer domain more than once. This is the companion to tracking ChatGPT referrals — same idea, one engine at a time, plus how to fold them all into a single AI-traffic segment.

If you only take one thing from this: build the segment engine by engine, but read it as one channel. Each engine’s quirks matter for setup; what you actually want to watch is total AI referral traffic and its trend — one stage of the GEO funnel that turns AI citations into clicks.

Per-engine referrer behavior

Here’s how the major non-ChatGPT engines show up when they send a referrer at all:

Two consequences fall out of that table. First, unlike ChatGPT — which now stamps many links with utm_source=chatgpt.com — these engines mostly don’t add UTM parameters, so the referrer is your primary signal. Second, each engine has a specific way of disappearing, which is worth understanding one at a time.

Perplexity

Perplexity is citation-first by design: every answer is built around linked sources, so users click through more than they do on chattier assistants. On the web, it passes perplexity.ai as the referrer reliably, which makes desktop Perplexity one of the cleaner AI channels to attribute. The catch is the mobile app and in-app browser, which frequently open external links without a referrer — those clicks land in Direct. If you want to understand why Perplexity sends the clicks it does, the citations driving them are covered in how Perplexity chooses citations.

Gemini

Gemini sends gemini.google.com (and the legacy bard.google.com) as a referrer, and it’s fairly consistent on the web. The trap here is conceptual, not technical: Google’s AI Overviews — the AI answers that appear inside Google Search results — are not Gemini referrals. A click from an AI Overview arrives as ordinary Google organic search traffic (google.com), not gemini.google.com. So “Gemini referral traffic” specifically means the standalone Gemini app or site; the AI answers woven into search are a different, larger, and mostly unattributable stream. Don’t expect your gemini.google.com row to capture your AI Overviews exposure — it won’t.

Copilot

Microsoft’s Copilot generally passes a referrer, and its move to expose referrer data made it more trackable than it used to be. The wrinkle is stability: Copilot’s referrer domain has shifted over time (bing.com/chat and copilot.microsoft.com have both appeared), so match on more than one string and keep an eye out for new ones rather than hard-coding a single domain.

Building one AI-traffic segment

The per-engine detail is for getting the setup right. Day to day, you want a single view of all AI referrals, so you can see the channel’s size and trend without hopping between rows.

Kitbase captures referrer as a first-class dimension — enriched server-side, available in every breakdown and filter, as described in the web analytics overview and the analytics dashboard guide. To build the segment, filter Referrer to match any of the AI domains:

perplexity.ai
gemini.google.com
bard.google.com
copilot.microsoft.com
bing.com/chat
chatgpt.com
chat.openai.com
claude.ai

Save that as your AI traffic segment. Now you can do the useful things: break it down by landing page to see which pages the engines send people to, scope a funnel to “sessions from AI engines” to see whether they convert, and watch the trend line as your AI Visibility and crawler numbers move. Because referrer is also a filter dimension, every event breakdown, journey, and retention view can be scoped to the AI segment or to a single engine within it.

Keep the list current. As new engines gain share and existing ones change domains (Copilot has, and others will), add their strings to the segment. Treat it as a living definition, not a set-and-forget filter.

The honest caveats

Everything above tracks the referrals you can see. As with ChatGPT, a large share of Perplexity, Gemini, and Copilot traffic arrives with no referrer and files itself as Direct:

• Mobile apps and in-app browsers strip referrers on external clicks — Perplexity’s app is a notable example.
• Copy-paste of a cited URL produces a Direct visit with no origin, every time.
• Changing referrer domains (Copilot) mean a segment that was complete last quarter can silently start missing traffic.

flowchart TD
A["Click from an<br/>AI engine"] --> B{"Referrer sent<br/>and matched?"}
B -->|"Web click,<br/>known domain"| C["In your AI segment ✓"]
B -->|"App / copy-paste /<br/>new domain"| D["Direct — dark traffic ✗"]
D -.-> E["Watch Direct + AI Visibility<br/>to infer the hidden share"]

So read the AI segment as a floor, not a total. Its trend is trustworthy even when its absolute number is understated — if the visible AI channel is climbing, the invisible one almost certainly is too. To reason about the hidden share, watch it alongside a rising, unexplained Direct segment (a classic AI dark traffic signature) and alongside AI Visibility, which measures whether the engines are citing you in the first place — the citations that cause these clicks, whether or not the referrer survives. Whether the clicks you do capture are worth chasing is answered in do AI referrals convert better.

FAQ

What referrer does Perplexity send? perplexity.ai, reliably on the web. In its mobile app and in-app browser the referrer is often stripped, so some Perplexity traffic still lands in Direct.

Is Gemini traffic the same as Google AI Overviews? No. gemini.google.com is the standalone Gemini app or site. AI Overviews inside Google Search send visitors as regular Google organic traffic (google.com), not as a Gemini referral — they’re a separate, mostly unattributable stream.

Why does Copilot sometimes show a different referrer? Copilot’s referrer domain has changed over time — bing.com/chat and copilot.microsoft.com have both been used. Match on multiple strings and monitor for new ones instead of relying on a single domain.

Do these engines add UTM parameters like ChatGPT? Not consistently. Unlike ChatGPT’s utm_source=chatgpt.com, Perplexity, Gemini, and Copilot mostly don’t tag links, so the referrer is your primary attribution signal for them.

How do I track all AI engines together? Build one saved segment that filters Referrer to match every AI domain (Perplexity, Gemini, Copilot, ChatGPT, Claude, and their aliases). Read it as a single channel, and keep the domain list updated as engines change.

Want one AI-traffic segment across every engine, with conversion and funnels attached — no cookies? Start your free trial — 7 days, no credit card required — and see what Perplexity, Gemini, and Copilot actually send you.