Preprint
Read abstract
Using large language models (LLMs) in research requires decisions about which work to offload and where sustained attention and increased friction can improve the execution of the user's intentions. This guide offers students and researchers a practical method for using LLMs to organize that attention while retaining responsibility for their reasoning and decisions. I describe an auto-interview practice in which an LLM asks one critical, pointed, and specific question at a time. My colleagues and I used this practice to develop the study design and analysis plan for the Stage 1 submission of a registered report. Working through the questions together helped us limit digressions, explain our judgments, and develop a shared understanding of the proposed study. A worked example shows how answering one specific question prompted us to reconsider which data would define our participant groups. Examining our own reasoning as we answered the question made the planning process an exercise in metacognition. The guide explains how to choose a suitable working environment, supply relevant context, and evaluate both individual replies and the developing conversation. It also shows how to extract reusable instructions and checks from repeated successful work, distinguishing what carries forward from the context each task requires. Documenting and sharing these methods allows other researchers to examine the judgments behind the work and adapt useful practices. The aim is to make deliberate engagement with research decisions a repeatable part of working with LLMs.