Evaluating renewable energy development policies: A SWOT-based multi-criteria decision analysis comparison of domain experts and large language models

Renewable energy development in emerging countries involves complex strategic decisions with long-term economic, environmental, and social implications, making the reliable evaluation of policy responses essential. Such evaluations have traditionally relied on domain expert knowledge, which is often time-consuming and resource-intensive. At the same time, Large Language Models (LLMs) are increasingly explored as potential decision-support tools due to their wide accessibility and ability to rapidly generate structured judgments. This study proposes an expert-driven SWOT-based Multi-Criteria Decision Analysis (MCDA) framework for evaluating renewable energy policy responses in Ghana and introduces a systematic comparison between domain expert judgments and LLM-generated assessments within the same decision-making structure using five selected LLMs: GPT 5.3-mini, Flash 3.0, Grok 4.2, Sonar, and Sonnet 4.6. The analysis is conducted at two stages of the MCDA pipeline: 1) criteria weighting using the Ranking Comparison (RANCOM) method and 2) policy ranking using the Measurement of Alternatives and Ranking according to Compromise Solution (MARCOS) method with aggregation via Compromise Fuzzy Ranking (CFR) into a group consensus ranking. Additionally, repeated LLM evaluations and the Risk-Informed Decision Making (RIDM) method are incorporated to assess output stability and the implications of using LLMs as supportive components in expert-driven decision processes. The results indicate a high level of agreement between LLM-generated and expert-derived criteria weights, while more pronounced differences emerge in decision matrices and resulting policy rankings. Domain experts identified increasing electricity demand as the most critical criterion, while national consensus on renewable energy implementation emerged as the most preferred policy response. LLM-based evaluations partially reproduce these priorities, although with variations in ranking structure depending on the model. The findings suggest that LLMs can effectively approximate expert judgments in selected stages of MCDA, particularly criteria weighting, while their use in complete decision pipelines requires expert validation. These results provide methodological insights into where LLMs can support expert-driven MCDA and where human expertise remains essential to ensure reliable policy recommendations.