What is AI referral traffic?

AI referral traffic arrives when an assistant answers a question and the user clicks through to one of the pages it cited or recommended. The click might come from a source link under a Perplexity answer or a citation beside a ChatGPT web search result. It is still a small share of most sites' traffic, but it behaves differently from organic search because the visitor has already read a synthesised answer before landing on your page.

You can usually spot it in analytics by referrer hostname. ChatGPT sessions tend to arrive with chatgpt.com as the referrer (older sessions used chat.openai.com), Perplexity with perplexity.ai, Copilot with copilot.microsoft.com, Gemini with gemini.google.com and Claude with claude.ai, though some sessions carry no referrer at all and land in direct traffic. ChatGPT also appends a utm_source=chatgpt.com parameter to many outbound links, which makes those sessions easier to isolate. The catch is that GA4's default channel groupings generally file all of this under Referral or Unassigned, so you need a custom channel group or an exploration filtered to those hostnames. Clicks from Google's AI Overviews are the blind spot: they are recorded as ordinary google.com organic traffic and Google does not currently break them out.

The reason this traffic tends to convert well is qualification. By the time someone clicks through from an assistant, much of the comparison work has already happened inside the conversation: the visitor has asked follow-up questions and narrowed their options before choosing your link. That is a very different visitor from someone skimming a page of blue links. Volumes are lower than classic organic search, sometimes dramatically so, but teams that segment AI referrals often report stronger engagement and better conversion per session.

AI referral traffic is a lagging indicator. You only receive the click if an assistant named or cited your brand in the first place, which makes visibility the thing to work on. Assistants also get brand details wrong more often than teams tend to assume: in Discoverable's July 2026 test of 30 brands, 27 of 30 were misdescribed when the assistant answered from memory and 13 of 30 still contained errors with live web search enabled. Discoverable's free AI visibility checker runs without a login and shows how an AI assistant currently describes and recommends your brand, and paid plans re-run the check weekly with a score trend, so you can watch whether the visibility behind this traffic is moving.