Eating healthier is something many people want to do, but turning nutrition advice into daily meals can be surprisingly difficult.
Most people know they should eat more vegetables, cut back on processed foods, and choose healthier options.
The challenge is figuring out how to make those changes without completely changing the foods they enjoy or spending more money.
Now, researchers at the University of California, Davis have developed an artificial intelligence system that may help solve this problem. Their study, published in PLOS Digital Health, suggests that making just one to three simple ingredient substitutions in a meal could significantly improve nutrition while also lowering costs.
The research was led by Trevor Chan and Professor Ilias Tagkopoulos. The team wanted to address a common problem in nutrition science.
While experts have developed clear dietary guidelines to reduce the risk of chronic illnesses such as heart disease, type 2 diabetes, and obesity, many people find these recommendations difficult to follow in real life. Diet plans often require major lifestyle changes, which can feel overwhelming and hard to maintain.
Instead of asking people to completely redesign their diets, the researchers explored whether small adjustments to familiar meals could make a meaningful difference.
To investigate this question, they analyzed information from the What We Eat in America survey, a large database that tracks the eating habits of Americans. The researchers examined 135,491 meals reported by 55,228 adults. These meals included breakfasts, lunches, and dinners representing a wide variety of eating patterns.
Using this information, the team trained a generative artificial intelligence model. The AI learned common meal patterns and how people typically combine foods. It was then asked to create realistic meals that remained similar to meals people already eat while moving closer to nutritional recommendations from the United States Department of Agriculture (USDA).
The results were encouraging. Compared with actual meals from the same dietary patterns, the AI-generated meals came 47 percent closer to USDA nutrition targets. Importantly, the meals still resembled foods people commonly eat, rather than becoming unrealistic or highly restrictive.
The researchers then took the process one step further. The AI examined meals and suggested one, two, or three ingredient substitutions that could improve nutritional quality even more.
These small changes produced significant results. The nutritional quality of meals improved by around 10 percent. At the same time, estimated meal costs fell by between 22 percent and 34 percent.
Many of the suggested substitutions were relatively simple. The AI frequently recommended adding vegetables, beans, or legumes while replacing highly processed foods or ingredients containing large amounts of sodium. Rather than eliminating favorite meals, the system focused on modifying them in practical ways.
One of the most interesting findings was that healthier eating did not necessarily require dramatic changes. In many cases, replacing only one or two ingredients moved meals substantially closer to recommended dietary standards.
The researchers also compared their specialized model with the general-purpose AI system GPT-4o. They found that their nutrition-focused model produced meals that more closely matched USDA recommendations, particularly regarding major nutrients such as protein, carbohydrates, and fats.
The study highlights the growing role artificial intelligence may play in public health. AI systems can analyze enormous amounts of dietary information and identify patterns that would be difficult for humans to recognize.
In the future, similar tools could be incorporated into mobile apps, meal-planning programs, healthcare services, and community nutrition initiatives.
However, the researchers caution that their work remains an early step. The study was entirely computer-based and did not involve real people making the suggested changes.
Scientists still need to determine whether individuals would actually follow the recommendations and whether the proposed substitutions would remain acceptable in terms of taste, convenience, and long-term adherence.
Another important consideration is that food choices are influenced by culture, family traditions, availability, personal preferences, and budget. While AI can suggest healthier options, successful dietary improvements will likely require approaches that take these factors into account.
Overall, the study provides promising evidence that healthier eating may be more achievable than many people think. Rather than requiring complete dietary overhauls, small and practical ingredient substitutions may offer meaningful improvements in nutrition while reducing food costs.
The findings suggest that artificial intelligence could become a useful tool for helping people make healthier choices without giving up the meals they already enjoy.
In reviewing the study, one of its greatest strengths is its use of a very large national dietary database and its focus on realistic meal modifications. The findings support the idea that small changes can have measurable benefits.
However, because the results have not yet been tested in real-world settings, additional research involving actual users will be necessary before firm conclusions can be drawn about long-term effectiveness.
Source: University of California, Davis.
