Large Language Models (LLM) are increasingly used as research companions, yet guidance for effective and responsible integration into scholarly workflows remains under-specified. This paper presents a three-stage, human-in-the-loop framework supporting (i) proposal development, (ii) research design, and (iii) manuscript preparation. Each stage operationalizes a toolbox of prompting and verification techniques—Context-Grounded Prompting, Graph-of-Thought, Self-Consistency, Contrastive Chain-of-Thought, and Chain-of-Verification—paired with explicit human checkpoints. The framework is instantiated in a case study on mitigating intersectional bias in hiring models, with detailed datasets, objectives, and evaluation metrics, and an exploratory pilot in which Master’s students apply the pipeline to thesis proposals. Results indicate that structured prompting improves traceability and broadens the set of considered alternatives, while verification steps curb overconfident errors. Prompts, artifacts, and rubrics are made available to support replication. The paper concludes with guidance on ethical deployment and limitations, emphasizing the primacy of human judgment in validating claims and shaping scholarly contributions.
Guérard et al. (Tue,) studied this question.