Introduction

As the production of knowledge synthesis research accelerates in the field of exercise science, so has the proliferation of generative AI (genAI) academic search engines that claim to speed up the process of information retrieval—an often time-consuming and resource-intensive slog that requires specialized skills. These tools are potentially an attractive option for researchers looking to save time and resources, but it remains uncertain whether they can match the rigor of traditional search methods. Undermind is a genAI academic search engine that aims to help users “research with unprecedented breadth and depth” and uses iterative semantic search technology to search faster than a human without compromising thoroughness. This study asks: Does Undermind match or improve upon the real-world search practices reported in exercise science knowledge synthesis studies?

Methods

This study calculated the precision and recall of the academic AI search engine Undermind against a random sample of search methods reported in 50 systematic reviews published in exercise science journals between 2015-2020. Each review question was searched the same way twice in Undermind to assess replicability. An analysis was conducted to explore patterns in the performance of Undermind.

Results

Analysis is ongoing and results will be added.

Discussion & Conclusion to be added.