JMIR Publications

ABSTRACT

Background:

large language models are increasingly used to support literature review, analysis, and manuscript drafting. In health-AI research, evidence generation requires more than generic writing assistance: citation verification, researcher review, protocol-based writing, and submission gates. Existing AI writing systems have not described how these requirements can be embedded into a health research team’s daily workflow.

Objective:

this paper describes Gabo, a governed evidence-generation system for health-AI organizations, with emphasis on workflow design, issue-based state management, citation-audit checkpoints, and researcher review gates.

Methods:

we conducted a retrospective methodological description of two manuscript-production cases. The unit of analysis was the writing workflow rather than model performance. Case 1 describes a medical billing audit manuscript; Case 2 describes a clinical Automated Machine Learning (AutoML) methodology manuscript.

Results:

the system operated through six layers. Case 1 produced 22 manuscript versions across 12 dedicated writing tasks, with iterative researcher corrections throughout each section. Case 2 reached submission through 12 manuscript versions, incorporating accumulated learning from Case 1, including a technical review by the engineering team at version 10. A structured self-review identified nine system gaps; eight were corrected within the same development cycle.

Conclusions:

Gabo is described as a potentially replicable specialized AI writing system for health research evidence generation. The observed writing progression across two cases supports the feasibility of a researcher-governed, task-managed approach to scientific writing in health organizations. A structured formative review identified nine governance gaps; eight were corrected within the same development cycle, supporting the view that the issue-based governance model supports not only writing production but iterative system improvement. The framework illustrates the scientific evidence-generation workflow applied to two manuscripts and documents the sequential execution of the system as a formative assessment of its operational feasibility. Future studies should prospectively evaluate the quality of produced manuscripts, as well as the transferability, scalability, and adoption of the system in broader research settings. Clinical Trial: N/A