Use of LLM assistants as support in the Project’s Formulation process: A Systematic Mapping

Introduction: Project formulation requires defining objectives, alternatives, resources, and evaluation criteria in increasingly complex contexts. In Colombia, despite the support of the Adjusted General Methodology (MGA), technical and operational barriers persist. Generative Artificial Intelligence, particularly large language models (LLMs), has emerged to support writing, synthesis, and planning, but its use in project formulation still lacks consolidated evidence. Objective: To identify and classify how LLMs have been applied to project formulation, their benefits and limitations, the evaluation methods used, application domains, and the existence of LLM or GenAI tools that support the MGA. Method: Systematic mapping following Petersen. Research questions were defined with PICOC, together with search strings and inclusion and exclusion criteria. IEEE Xplore, Scopus, ScienceDirect, Web of Science, and Taylor & Francis were queried from 2020 to 2025. A PRISMA diagram summarized identification, screening, and selection. Of 372 records, 33 studies met the criteria. Results: GPT-3.5 and GPT-4 predominate. The most evidenced tasks include effort and time estimation, plan and schedule generation, ideation and design, and early risk management. Support was also observed for requirements, work breakdown, and resource allocation. Evaluation combined human comparisons, expert review, automatic metrics with other LLMs, surveys or interviews, and objective measures. No reports were found of LLM or GenAI tools to support the MGA. Conclusions: LLMs speed up cognitive tasks and improve clarity and creativity, but they require human validation, better contextualization, explainability, data protection, and technical integration. Trends point to hybrid approaches with retrieval augmented generation (RAG), multitask agents, and multimodal capabilities. The absence of reported experiences for the MGA reveals a strategic opportunity for research and development.