Docs/AI Traffic Detection

AI Traffic Detection

Metrivo identifies AI-search referrals using four signals: referrer headers, UTM parameters, user-agent strings, and session evidence. This page describes what is detectable today and what is not.

How detection works

When a visitor hits a tracked page, the Metrivo script captures the document referrer, current URL parameters, and a coarse user-agent fingerprint. The collector compares these against a list of known AI-search hosts and patterns. Confirmed referrals are labeled by platform. Partial signals are recorded as inferred AI-assisted sessions. Visits with no signals at all remain Unknown direct traffic.

When a payment webhook later matches one of these sessions, Metrivo attributes the revenue to the source it observed — without elevating the confidence beyond what the underlying session supports.

Detection matrix

Platform	Detection method	Confidence	Common limitation
ChatGPT	Referer headers from chat.openai.com and chatgpt.com, plus known UTM patterns when sources are cited.	High	When users open links inside the ChatGPT desktop or mobile app, the referrer may be stripped, so traffic can appear as direct.
Perplexity	Referer headers from perplexity.ai. Source citation clicks are detectable when the browser forwards the referrer.	High	Some Perplexity surfaces (in-app browsers, mobile clients) may not forward a referrer.
Claude	Referer headers from claude.ai. Some traffic carries identifying user-agent or query parameters.	Medium	Claude does not always surface clickable source links, and some clients strip the Referer header, so volume is uneven.
Gemini	Referer headers from gemini.google.com, plus Google AI Overview referrer patterns when available.	Medium	Google AI Overview traffic can be reported under standard Google referrers, making it hard to fully separate from organic search.
Copilot	Referer headers from copilot.microsoft.com and Bing Chat surfaces, plus known UTM patterns.	Medium	Copilot inside Edge or Windows may not always forward an external referrer.
Unknown AI	Heuristics on user agent, query parameters, and session context for AI clients we cannot name with confidence.	Low	Without a recognizable referrer or signature, traffic is labeled Unknown rather than attributed to a guessed platform.

Confidence labels describe how reliably Metrivo can attribute a session to that platform, not the quality of the user.

Definitions

Quote-ready definitions used across the Metrivo dashboard and docs:

Attributed revenue

Revenue from payments that Metrivo has linked to a specific traffic source using session, visitor, or hashed-email evidence.

Unattributed revenue

Revenue from payments where no session, visitor ID, or hashed email matched. Metrivo does not assign these to a guessed source.

Confirmed AI-search referrals

Sessions where the referrer, UTM, or user agent clearly identifies the AI platform (for example, perplexity.ai in the referer).

Inferred AI-assisted sessions

Sessions where partial signals (UTM tags, certain user-agent strings, or known AI client behavior) suggest AI involvement without a confirmed referrer.

Unknown direct traffic

Sessions that arrive with no referrer, no UTM, and no AI signature. Metrivo keeps these as direct/Unknown rather than re-tagging them as AI.

What Metrivo will not claim

Metrivo does not claim that every direct visit is hidden AI traffic. Unknown stays Unknown.
Metrivo does not promise to detect 100% of ChatGPT, Perplexity, Claude, Gemini, or Copilot referrals — many clients strip the Referer header.
Metrivo does not infer which exact AI conversation produced a visit. It only records the platform when evidence is present.
Metrivo does not attempt to identify the specific user behind an AI-assisted session.