A new Scrunch study tracking millions of searches, AI conversations, and website visits found that AI referrals accounted for just 1.1% of visits to news publishers following conversations with AI assistants. Most visits arrived through direct navigation or traditional search.
Scrunch was careful about the interpretation: the results do not show AI creating new demand, only that news-related AI conversations often coincide with continued news reading.
This is the load-bearing point. A referrer captures only the last click before a visit. It cannot reconstruct a path that runs AI conversation → search → direct visit days later. A reader who first encountered a brand in an AI answer and typed the URL a week later registers as direct traffic.
Measuring AI's impact through referral traffic alone guarantees an undercount.
News searches with an AI Overview produced publisher clicks about 20% of the time, versus roughly 30% for searches without one. Read alone, that looks like a 10-point loss caused by the Overview.
But comparing the same query with and without an AI Overview shrank the gap to about 2 percentage points. Most of the difference came from which queries trigger Overviews, not from the Overviews themselves.
During the study period, AI Overviews appeared on roughly one in four news searches, most often on utility-style topics like weather, market prices, and explainers. Sports searches — which had the highest publisher click-through rates — rarely triggered one.
When an AI assistant volunteered a publisher the reader had not named in the prompt, visits to that publisher rose 10.6 percentage points the next day and nearly 20 points over the following week, compared with similar publishers not mentioned. Scrunch describes this as a routing signal rather than evidence of causation.
The distribution is the problem.
The gap widens between being mentioned and being visited. A name in the answer does not convert on its own.
For a layered measurement structure, see A 5-Layer Framework for Measuring AI Search Performance; for content requirements that earn mentions, see LinkedIn Publishes an AI Citation Playbook.
About the data: the study analyzed millions of searches, AI conversations, and publisher visits from a privacy-safe, opt-in panel between February and June 2026. It measured page visits — not subscriptions, revenue, or time spent.
No. Referrers record only the last click before a visit, so paths running from AI conversation to search to a later direct visit are invisible. Referral data structurally undercounts the effect.
On raw averages, 20% versus 30%. But comparing the same query with and without an Overview narrowed the gap to about 2 percentage points — Overviews mostly appeared on utility queries that already generated few clicks.
When an assistant volunteered a publisher the reader had not named, visits rose 10.6 percentage points the next day and nearly 20 points over the following week. Scrunch calls this a routing signal, not causation.
Major publishers took 82% of volunteered mentions and 97% of follow-through visits. Mid-sized publishers held about 17% of mentions but a much smaller share of resulting traffic.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
An E-E-A-T checker was never automatable because there is no API for it. Here is how an AI coding agent read Google's quality guidance, crawled a page sample, and produced a scored audit — plus where it stops.
A GEO audit is not checking whether your brand appears in an AI answer. Here is a nine-point GEO checklist split into four on-site items, four off-site, and one measurement layer.
LinkedIn, YouTube, Reddit, and TikTok are building systems to identify low-value AI content. Here are the limits of watermarking and what content still gets distributed.