When someone discusses the news with an assistant such as ChatGPT and opens a publisher directly a few days later, analytics usually records the source as direct. If the person searches for the outlet on Google, organic search gets the credit. A conversational AI touchpoint may have influenced the journey without appearing in the last-click report.
A study published by Scrunch on August 13, 2026 makes that measurement gap visible through news consumption. The company linked millions of searches, AI conversations, and publisher visits for the same people in a privacy-safe opt-in panel between February and June 2026. Only 1.1% of the publisher visits observed after news-related AI conversations carried a direct AI referrer. This does not mean AI influenced only 1.1% of visits. It means a last-click referrer captures very little of the preceding journey.
The practical news for publishers, content brands, and marketing teams in Türkiye is a change in measurement discipline. Treating the AI referral row in GA4 as total AI impact can exclude later direct and branded-search visits. Assigning every direct visit to AI would be equally misleading.
What did the study find?
The analysis compared weeks after a person's news-related AI conversation with weeks after their non-news conversations. People were 20.5 percentage points more likely to visit a news publisher in the week after discussing news with an assistant. The researchers do not present this as proof that AI created new demand. A news conversation can simply mark an existing moment of interest. The result shows that opening an assistant does not necessarily end the open-web journey and is associated with later news activity.
About three quarters of those visits arrived through direct navigation, while roughly 9% came through traditional search. Only 1.1% carried an AI referrer. Because referral data records the final transition, it cannot reconstruct a chain that starts with a conversation and continues through a later search or direct visit.
The study also examined publisher naming. When an assistant volunteered a publisher that the user had not included in the prompt, visits to that publisher rose 10.6 percentage points more by the next day than visits to comparable unnamed publishers. The gap reached about 19.7 points over a week. Scrunch classifies this as a strong routing signal, not causal proof. Model selection and prior reader familiarity are not fully separable.
This Fark Studio illustration presents direct, search, and other referral routes as parallel paths after an AI conversation. It is not user data or a strict funnel.
What does the AI Overviews result mean?
The Google AI Overviews section asks a different question. Publisher clicks occurred on about 20% of news searches with an AI Overview and about 30% without one. When the same query was observed both with and without an overview, however, the gap narrowed to roughly two percentage points. The full raw difference therefore cannot be assigned to the overview alone. Google shows AI Overviews more often on utility-style topics, such as weather, market prices, and explainers, where users were already less likely to click a publisher.
AI Overviews appeared in about one quarter of news searches in the Scrunch panel. The rate climbed from roughly one in six in February to more than four in ten in June. Sports searches produced comparatively strong publisher click-through while triggering AI Overviews less often. Query mix is therefore an important confounder.
This observational result does not invalidate the randomized field experiment conducted by Saharsh Agarwal and Ananya Sen with 1,065 US desktop Chrome users. Their paper, first posted in April and revised in July, found that when an AI Overview was actually shown, it reduced outbound organic clicks by 39.8% and increased zero-click searches by 34.5%. The studies use different designs and answer different questions. One tests the causal effect of showing an overview. The other examines selection effects in real news-query mixes and the wider journey after an AI conversation.
The visual compares abstract search layouts with and without a generative-answer layer. It is not a Google interface, a real CTR chart, or a product screenshot.
What is confirmed and what remains uncertain?
Direct AI referral data is clearly not a complete record of the journey. Google Search Console counts an external-link click inside AI Overviews or AI Mode as a click. On some surfaces, a link must be scrolled or expanded into view before an impression is recorded. These metrics describe visibility and clicks within Google, but they do not join a later branded search to an earlier AI exposure in one user journey.
Scrunch's public methodology does not disclose the total panel size, country distribution, exact assistant mix, or full list of publishers. The panel is opt-in and tilted toward people who use AI. It measures page visits rather than subscriptions, revenue, reading time, or news quality. Those limits make a Türkiye-specific conversion factor impossible.
Concentration also matters. Major publishers accounted for 82% of the names volunteered by assistants and 97% of the subsequent visits to named outlets. Being present in an answer does not guarantee that a mid-sized or niche publisher will be chosen at the same rate.
What should brands in Türkiye do now?
1. Monitor the AI referral row in GA4, but do not label it total AI impact. Use a bounded name such as direct AI referrals. 2. Review generative AI impressions and clicks in Search Console alongside branded queries and related content clusters. If the dedicated report is not available for the property, preserve trends in the overall performance data. 3. Create a stable prompt panel for important topics. Measure brand mentions, citations, answer position, and sentiment under consistent dates and market conditions. 4. Do not assign changes in direct or organic traffic to AI by default. Record campaigns, email, PR, seasonality, app use, and dark-social activity as competing explanations. 5. Compare AI visibility with branded search, new-user visits, and qualified outcomes over at least four to eight weeks. Avoid daily correlations at low volume. 6. Instead of trying to imitate the scale of major publishers, build topic-specific evidence, consistent brand presence, and clear authorship. A niche brand does not need to appear in every answer. It needs to be chosen at the right research or purchase moment. 7. Separate three layers in executive reporting: visibility, observable traffic, and commercial outcomes. Do not assume an increase automatically passes from one layer to the next.
This framework separates the last-click lens from a wider set of visibility and behavioural signals. It is not an attribution model or a real performance dashboard.
Where should teams wait?
Do not use 1.1% as a benchmark for every industry or for Türkiye. The study focuses on news publishers, and the public methodology does not provide country-level results. Ecommerce, travel, and B2B journeys may behave differently.
A single jump in branded search is also insufficient grounds for reallocating budget to AI visibility. Avoid causal claims until the observation window, control topics, and movements in other channels are documented. There is no reason to wait before correcting report labels, fixing the prompt panel, or annotating measurement breaks.
Fark Studio perspective
AI-search measurement has two unreliable extremes. One looks only at referrers and concludes there is no effect. The other treats every increase in direct and branded search as AI success. A more credible model keeps SEO visibility, digital marketing touchpoints, and performance marketing outcomes in separate layers, then reads their movement together over time.
If you want to connect AI-source visibility to real demand and conversion signals, plan a visibility and attribution audit with Fark Studio. The goal is not to give AI more credit. It is to expose the unknown area and improve decision quality.
Sources
Scrunch, Surviving the clickpocalypse: What millions of search events say about AI eating the news, August 13, 2026; panel methodology, post-AI visit routes, and publisher concentration.
Search Engine Land, Only 1.1% of news publisher visits come directly from AI: Study, August 13, 2026; independent industry summary of the research.
Saharsh Agarwal and Ananya Sen, The Impact of Google AI Overviews on Publisher Traffic and User Experience: Evidence from a Field Experiment, April 3, 2026, revised July 8, 2026; randomized field experiment on AI Overview effects.
Google Search Console Help, What are impressions, position, and clicks?, accessed August 13, 2026; click, impression, and position methodology for AI Overviews and AI Mode.
