A new study published in the open-access journal PLOS Digital Health suggests that artificial intelligence (AI) can help people eat healthier and spend less by recommending just one to three ingredient swaps in their meals. Trevor Chan and Ilias Tagkopoulos conducted the research at the University of California, Davis.
Although dietary guidelines for preventing chronic conditions such as diabetes and cardiovascular disease are well established, many people struggle to apply them to everyday meals. Existing dietary tools often require extensive changes, making healthy eating difficult to sustain.
To address this challenge, the researchers analyzed 135,491 meals reported by 55,228 adults in the What We Eat in America study. Using these data, they trained a generative AI model to create realistic breakfast, lunch, and dinner meals that reflected common eating patterns while improving nutritional quality.
The AI-generated meals were 47% closer to U.S. Department of Agriculture (USDA) nutritional targets than comparable real-world meals, while remaining similar in flavour profiles and meal types. When the model recommended one to three ingredient substitutions, nutritional quality improved by about 10%, and estimated meal costs fell by 22% to 34%.
Common recommendations included adding vegetables or legumes and replacing highly processed or high-sodium foods with healthier alternatives. The specialized AI model also outperformed GPT-4o in generating meals that more closely aligned with USDA macronutrient guidelines.
While the findings have not yet been tested with real users, the researchers believe the approach could help translate dietary guidelines into practical, affordable meal improvements. Their results suggest that healthier eating does not require a complete overhaul of familiar meals; in many cases, a few targeted ingredient swaps may be enough to make meals both more nutritious and more cost-effective while preserving the foods people enjoy.
More information: Trevor Chan et al, Translating dietary standards into healthy meals with few-ingredient substitutions, PLOS Digital Health. DOI: 10.1371/journal.pdig.0001367
Journal information: PLOS Digital Health Provided by PLOS
