Whether It’s Searchable or Not: How Generative AI Affects Finding Diverse Information Online
From Searchable to Non-Searchable: Generative AI and Information Diversity in Online Information Seeking
HCI Today summarized the key points
- •This article analyzes research on how generative AI like ChatGPT changes the diversity of online information seeking.
- •The research team analyzed more than 230,000 human–ChatGPT conversation logs from 2023 to 2025, dividing questions by their searchability.
- •As a result, 79% of questions were non-searchable—difficult to answer directly with general search—and these questions covered broader topics and content.
- •Meanwhile, for searchable questions, ChatGPT’s answers had lower information diversity than Google search results across most topics.
- •It also shows that the more diverse the AI answers are, the more diverse the next questions become—demonstrating that generative AI can broaden exploration, but its answers may still become narrower.
This summary was generated by an AI editor based on HCI expert perspectives.
Why Read This from an HCI Perspective
This article shows generative AI not as a ‘smarter search engine,’ but as an interaction tool that changes the information-seeking experience. It’s important that it examines how users ask questions, what answers they receive, and how their subsequent exploration changes. In particular, the approach of separating searchable questions from non-searchable ones offers strong implications for both UX design and HCI research, because it empirically demonstrates that while AI can broaden the scope of exploration, it may also reduce information diversity.
CIT's Commentary
The core of this study is not the model’s raw performance, but how the ‘back-and-forth between questions and answers’ changes users’ exploration. The finding that 79% of questions are difficult to handle well with traditional search suggests that people are using AI as a wider space for imagination than a search box. Conversely, the fact that AI answers to the same questions are less diverse than search results shows that a comfortable conversation can actually narrow one’s view. Therefore, an important design challenge is not to stop at providing good answers, but to reveal system state and create intervention paths that let users re-enter exploration from different directions. If this balance breaks down in hybrid search or AI agents, convenience may increase while the breadth of exploration could shrink.
Questions to Consider While Reading
- Q.How can we detect—at the interface level—the moment when AI reduces answer diversity, and make that visible to users?
- Q.How should the criteria for distinguishing searchable from non-searchable questions be designed differently in product contexts?
- Q.If answer diversity affects the diversity of the next questions, what UX metrics should be designed to measure it?
This commentary was generated by an AI editor based on HCI expert perspectives.
Please refer to the original for accurate details.
Subscribe to Newsletter
Get the weekly HCI highlights delivered to your inbox every Friday.