Tracking ChatGPT referrals

Lindsay Smith · Published 2026-08 · Evidence: operator documentation

Of all AI-search outcomes, one is directly countable with tools you already run: clicks from ChatGPT arrive labelled. OpenAI appends utm_source=chatgpt.com to applicable referral links. Official recommendation

Setting it up

You install nothing; the tag arrives on the URL. In your analytics, segment sessions where the landing URL carries utm_source=chatgpt.com. A saved report or channel definition takes minutes. Watch three things: which pages receive the clicks, what those visitors do next, and whether either changes over months. The landing pages reveal the most, because AI referrals tend to arrive deep on specific, evidence-heavy pages rather than on the homepage.

Reading the numbers honestly

• Expect small counts. AI answers resolve many questions without a click, and a handful of referrals can still represent real, high-intent interest. Judge quality, in enquiries and purchases, rather than volume.
• Absence is ambiguous. Zero referrals may mean you are never cited, or cited without clicks. Pairing referral data with prompt testing separates the two; nothing else does.
• Coverage is partial. The tag covers applicable ChatGPT referral links. Other AI surfaces have their own habits, and some send no referrer at all. This instrument measures one pipe and measures it well.

Why it still matters

This is the cleanest link in the measurement chain: an AI system recommended you, a human acted, and both facts sit in your own data. When someone claims AI search sends no real customers, or claims it will replace all your traffic, this number checks both stories against your business.