The paradox of information abundance: Answers provided by popular information-seeking tools lead to differences in trust, memorability and desire for more information

Search engines and AI tools organise, process, and make the world’s information easily available. By leveraging learned statistical properties within natural language, AI tools have the potential to provide information in a format that is easy to process. The format potentially minimises cognitive load, maximising the information users successfully integrate within memory. We explore how information seekers evaluate information relative to output (ChatGPT and Google’s Featured Snippets) and query type (trivia, e.g., the largest city in Venezuela, and ‘how-to’ or instructions for actions, e.g., how to stand on a surfboard). Participants input a series of twenty queries into a search box and viewed the text output of each tool’s response to these queries. They then rated the trustworthiness, ease of understanding of each response, and their desire to seek additional information. The response was presented without the context of the search tool to ensure that any effects were not confounded by features of the display or preconceived biases associated with a given search tool. Participants then completed an unanticipated multiple-choice test on the material. Participants found the more detailed responses to how-to queries offered by ChatGPT more trustworthy and informative than the more direct and concise responses from Google. Paradoxically, Google answers to these how-to queries resulted in higher recognition scores at test. We discuss these results in relation to how stylistic aspects of generative AI can result in meta-cognitive failure based on the cognitive heuristics of fluency and the paradox of information abundance.