Does AI traffic show up in analytics? Only 1.1% of post-AI news visits arrive with an AI referrer

What the study measured

The study, by Scrunch principal data scientist Michael Iannelli and published 2026-08-13, analyzes "a privacy-safe opt-in panel that links people's real AI conversations to the same people's anonymized web activity (i.e., Google searches and news-publisher visits) from February 2026 to June 2026" — millions of search events. Because the panel sees the conversation AND the later visit from the same person, it can measure influence that never carries a referrer. Disclosure worth knowing: Scrunch sells an AI-visibility platform, so the vendor has an interest in AI influence being real; the design, the numbers and the stated caveats are nonetheless the most direct measurement of the AI-to-visit path we have seen published. Search Engine Land covered the study the same day.

The numbers: AI moves visits, referrers miss them

• After a news-topic conversation with an assistant (ChatGPT, Perplexity, Gemini, Claude and others), readers are 20.5 percentage points more likely to visit a news site the following week than after a non-news conversation.
• But "only 1.1% of post-AI conversation news visits show up as direct AI referrals." The rest arrive by ordinary-looking routes: "Direct navigation accounts for about three quarters, traditional search for about 9%, and other referrals for most of the remainder."
• The study is careful with the causality: "That doesn't mean AI is the cause of 98.9% of news visits. It means referral reporting captures almost" none of the influence — in its words, "a referral dashboard can't tell you whether AI shaped a later search or direct visit."
• Concentration is stark: major publishers are 82% of the publisher names assistants volunteer (mid-sized 17%, niche under 1%) — but of the visits that land on a named publisher, 97% go to a major. After an assistant names them, majors see a 22.9-point lift in visits; mid-sized just 8.6.

The AI Overviews click gap mostly isn't the overlay

The study also re-examines the most-quoted number in the "clickpocalypse" narrative: news searches with an AI Overview convert to a publisher click about 20% of the time versus about 30% without one. Held to the same query — "the same query, seen sometimes with a generative overlay and sometimes without — the gap shrinks to about 2 percentage points. That's too small for us to tell apart from zero." In the study's framing, Google isn't so much shrinking clicks on a given search as choosing to put overlays on searches that rarely produced clicks anyway. That is a materially different diagnosis than the 10-point headline gap implies, and it echoes what Google's own randomized experiment claimed in July — this time from a third party's panel rather than Google's.

What this means if you care about AI visibility

This is news-publisher data — an opt-in, AI-engaged panel counting page visits, "not subscriptions, ad revenue, or how long anyone reads," as the study itself cautions — so extrapolate to other businesses with care. But two mechanics travel well. First: if AI-shaped visits overwhelmingly arrive as direct navigation, then referral-based measurement — the way nearly everyone currently gauges whether AI matters to their business — undercounts by orders of magnitude. AI referral rows are already tiny; this study says the influence behind them is roughly 90× the visible slice. The measurable asset is whether assistants NAME you when a customer asks, not whether a click carries a referrer. Second: being named is necessary but not sufficient — assistants volunteer mid-sized names 17% of the time, yet 97% of follow-through goes to majors. Familiarity compounds after the answer. For a small business the practical reading is the same one our data keeps producing: check whether AI engines actually name your business when asked like a customer would — because the visit that follows will look like it came from nowhere.