Can ChatGPT and other generative artificial intelligence (A.I.) large language models (LLMs) successfully recommend to homeowners specific energy efficiency retrofits that can best lower carbon emissions or provide the quickest return on their investment?
Turns out, LLMs are good at explaining in general terms how retrofit options could lower emissions, but they can’t really rank options against each other—particularly when considering a property’s specific features, local weather conditions and human consumption behaviors, according to a study released this month by researchers at Michigan State University.
As for which options could provide the earliest payback period for homeowners, LLMs only factored in upfront costs in their decisions—completely disregarding how energy cost savings from the retrofits could expedite ROI.
“While the emerging applications demonstrate the promise of LLM-based A.I. in supporting retrofit decisions, most models are trained on broad, general-purpose datasets rather than domain-specific knowledge,” the researchers wrote. “As a result, their capability to generate reliable energy retrofit recommendations that reflect diverse building characteristics, climate conditions and occupant behaviors remains unclear.”
The researchers prompted seven LLMs—ChatGPT, DeepSeek, Gemini, Grok, Llama and Claude—to compare 16 retrofit options for 10 different households that were detailed within the prompt. The task within the prompt was phrased this way:
“You are a house retrofit specialist. There are ten houses that need retrofitting. First, analyze the characteristics and differences of each house. Then, compare the cost and efficiency of 16 retrofit packages for each house. Finally, identify the retrofit package that achieves the greatest CO2 reduction and the one with the shortest payback year for each house.”
The researchers found that LLMs have the ability to produce effective retrofit decisions, although they struggle to pinpoint the best one. They are better at determining the technical context of retrofit options, i.e., discussing how they can maximize CO2 reduction, and not as good at evaluating the “sociotechnical” context, i.e., how retrofit options can minimize payback periods.
“This difference likely reflects the relative clarity and consistency of technical optimization objectives, which are more easily captured by model reasoning, whereas sociotechnical considerations involve trade-offs between economic, behavioral and contextual factors that may be more difficult for LLMs to interpret and prioritize accurately,” according to the report.
To make LLMs a more reliable decision-making source for energy retrofits of specific homes, their accuracy, consistency and contextual understanding must be improved, the researchers concluded.
About The Author
KUEHNER-HEBERT is a freelance writer based in Running Springs, Calif. She has more than three decades of journalism experience. Reach her at [email protected].
