Evaluating a Standardized ChatGPT Workflow for Data Extraction from Research Articles: A Multi-Project Study

This study evaluates a standardized, project-specific ChatGPT workflow for extracting structured information from research articles. It is a multi-participant implementation evaluation. Each participant will use the supplied Extraction, Review, and Evaluation skill to organize files, preserve evidence, validate outputs, create audit records, and calculate the prespecified results. The substantive extraction rules remain the participant's frozen coding prompt.

All work completed before this registration, including TEST1 calibration, pilot, feasibility, workflow-development, and previously viewed Ground Truth, will be described as prior work and will not enter the prospective primary cohort. A previously developed prompt may be reused only if none of the participant's designated formal papers contributed to its development or revision. No prior formal-test article, coding sheet, AI output, comparison, adjudication note, or performance summary may be supplied to ChatGPT for the new prospective evaluation.

The prospective cohort begins after this document is deposited and each participant has reserved formal papers, separated formal Ground Truth, and frozen the formal extraction system.