Use of Publicly Available Artificial Intelligence Tools to Support Prosumer Decision-Making: The Case of Households with Photovoltaic Installations
Purpose: The aim of this article is to assess the usefulness of publicly available generative artificial intelligence tools in supporting prosumers in decisions related to energy and cost management. Particular attention is paid to the ability of the models to consider the structure and timing of energy consumption, individualize recommendations, and recognize limitations caused by incomplete input data. Methodology/Approach: The study was conducted as a comparative case study of two real households equipped with photovoltaic micro-installations with similar production parameters but different structures of major loads and different usage profiles. ChatGPT, Gemini, and Microsoft Copilot were examined in their publicly available versions, without paid plans, APIs, agents, or additional tools. Each system received identically structured datasets and the same prompt. The responses were assessed qualitatively in terms of data use, individualization, factual correctness, feasibility and prioritization of recommendations, recognition of missing information, and quality of justification. Findings: All tools correctly recognized the importance of the time mismatch between production and consumption and produced different recommendations for the two households. The largest differences concerned the prioritization of actions and the way the tools responded to incomplete data. Gemini more often filled data gaps with additional assumptions, which led, among other things, to an unjustified estimate of PHEV energy consumption and an imprecise interpretation of the settlement system. Copilot more often made its conclusions conditional on obtaining additional parameters. ChatGPT took an intermediate position, providing recommendations that could be made from the available data while also indicating the limits of their accuracy. None of the tools considered an additional investment to be clearly justified on the basis of the available information. Practical implications: Publicly available generative AI can support prosumers at the stage of initial analysis and identification of possible actions without the need to build a formal optimization model. However, recommendations that require precise technical, economic, or investment assessment should be verified using more detailed data and specialist methods. Originality/Value: The study does not assess a specialized energy system. Instead, it evaluates the practical usefulness of freely available conversational tools in the same clearly defined real-world decision problem. The value of the study also lies in comparing the flexibility of generative AI with the transparency and repeatability of classical multi-criteria methods.