Search Engine Land reported this week that a new Scrunch study puts the AI referral share of news publisher visits at 1.1%, a figure that reads at first glance like evidence AI answers are not actually hurting publisher traffic. The two claims are not interchangeable. A tiny referral share is consistent with an AI answer satisfying the reader before a visit ever happens, which is a different mechanism than an AI answer sending the reader to a competitor’s link instead.
Scrunch, the vendor behind the study, has not published any independent replication of its findings. Its analysis drew on an opt-in panel, screened for privacy, that it followed from February through June 2026, spanning millions of search events, AI conversations and site visits. Within the specific window Scrunch measured, visits occurring during the week after a reader’s news-related AI conversation, roughly three-quarters arrived through direct navigation, about 9% arrived through a traditional search engine, and 1.1% carried a tag identifying an AI product as the referrer. That is the precise population the 1.1% figure describes. It is not a measure of what share of all publisher traffic, across every reason someone might land on a page, comes from AI.
A referrer is generated only when a click carries a traceable link back to the AI product a reader used. It cannot register when a chatbot simply answers the question inside the conversation and the reader never opens a browser tab. A low referral share is exactly what that substitution pattern would produce, not evidence against it. Scrunch’s own methodology note reinforces the limit: referral logs register only the final click before someone lands on a page, not an AI conversation that happened several steps earlier in that reader’s path.
Scrunch did find a behavioral link between AI conversations and continued news reading. People who’d just discussed news with an AI assistant showed up on publisher sites at a rate 20.5 points higher over the next week than people whose AI conversation covered something other than news, a gap Scrunch calculated by comparing each reader’s own browsing history over time. Scrunch cautioned that this pattern does not show AI conversations created new demand for news. It shows only that a news-related AI conversation tended to coincide with continued reading, a correlation, not a causal claim about traffic generation.
Google’s AI Overviews complicate any simple story about traffic loss, in a direction critics might not expect. News searches displaying an AI Overview produced a publisher click around 20% of the time, versus close to 30% on searches without one, roughly a ten-point gap. Once the analysis held the query itself constant, checking pages that did and did not show the feature for identical searches, the difference dropped to about two percentage points. AI Overviews turned up disproportionately on utility queries, weather, market prices, quick explainers, the kinds of searches that already sent readers to publishers less often. Sports queries, the best performers for publisher clicks in the data set, almost never showed an AI Overview at all. Across the full study period, an AI Overview showed up on roughly a quarter of all news searches.
Being named without being asked for still moves behavior, even without a click. When an AI assistant volunteered a publisher’s name that the reader hadn’t typed into the prompt, next-day traffic to that publisher rose about 10.6 percentage points, and the gap widened to nearly 20 points across the following week, relative to comparable publishers that went unmentioned. Scrunch labeled the pattern a signal about routing behavior, stopping short of calling it proof that the mention itself caused the visit. Large, established publishers captured almost all of that effect: they made up 82% of the publisher names AI assistants offered unprompted and 97% of the visits that followed those mentions. Mid-sized publishers accounted for close to 17% of the mentions but converted them into a much smaller slice of the resulting visits.
None of this means AI answers are neutral for publishers, and none of it proves they are devastating either. It means referral logs cannot settle the question alone, because a reader who gets a satisfying answer inside the chat leaves no referrer behind to count. Search teams should start tracking how often their brand gets volunteered by AI assistants on queries readers did not type themselves, since that unprompted-mention rate predicted next-week visits far more strongly in this data than any AI-referrer share did.
Search Engine Land reported the Scrunch findings on August 13, 2026, in an article by Danny Goodwin.
