AI answers strip out the signals that told users how much to trust them
Answer engines return a conclusion in the same confident tone whether the evidence behind it was deep or almost absent, and the signals people once used to judge that evidence do not survive the trip. Duane Forrester calls this lost byproduct path metadata: the number of sources, whether they contradicted each other, and how long the search took all used to shape how firmly someone committed to what they found.
Two Wharton marketing professors, Shiri Melumad and Jin Ho Yun, ran seven experiments with 10,462 participants and published the results in PNAS Nexus in October 2025. People who learned about ordinary topics from an AI summary knew less afterwards than those given standard Google links, even when both groups saw identical facts, and the advice they wrote was sparser and less likely to be acted on. Adding live source links beside the summary did not help, because participants did not click them.
Pew Research Center tracked 900 US adults across 68,879 Google searches in March 2025 and found the same pattern in real browsing. When an AI summary appeared, people clicked a normal result on 8% of visits against 15% without one, and clicked a source cited inside the summary on about 1% of visits. They ended the session on 26% of pages with a summary against 16% without, though Pew describes this as association rather than proven cause.
For brands, the practical consequence is that a thin or wrong answer no longer gets repaired by the reader's onward journey. At a 1% source-click rate, the correction that used to happen when someone landed on the real page and swapped the wrong version for it does not fire, so a misrepresentation that was once temporary now stays in place.
Reported by Search Engine Journal ↗
Topics: AI Overviews, Citations, Measurement