A journalist has detailed the results of a two-week experiment using OpenAI's ChatGPT to track and analyse her dietary habits, with the goal of building muscle and losing fat. The experiment, assessed by a registered dietitian, highlighted the AI's utility in identifying eating patterns but also exposed its limitations in providing nuanced, emotionally intelligent advice.
Julia Pugachevsky, a reporter for Business Insider, conducted the trial after growing frustrated with traditional food-tracking applications. She inputted daily logs of all food, drink, exercise calories burned, step counts, and resting metabolic rate data into a single ChatGPT conversation.
AI Provides Actionable Protein Guidance
Based on prior advice from a Life Time trainer to consume 90-100 grams of protein on workout days, Pugachevsky struggled to meet her targets, particularly on rest days. ChatGPT advised aiming for at least 80 grams of protein on non-training days, a goal she found "slightly more doable." The AI suggested protein-rich foods like Greek yogurt and, notably, never recommended meat products, having seemingly inferred her pescatarian diet from the logged data.
Shannon O'Meara, a registered dietitian at Orlando Health who reviewed the experiment, stated the protein advice was sound but emphasised the importance of goal-setting. "If you want to hit a certain calorie or protein goal, just make sure those goals are sound," O'Meara said, highlighting that objectives should ideally be set by a healthcare or fitness professional.
Pattern Recognition Emerges as Key Strength
One of the most significant benefits Pugachevsky reported was ChatGPT's rapid identification of her two distinct dietary "modes." On active days, she consumed protein shakes and pre-prepared meals and abstained from alcohol. On rest days, she typically ate less protein, walked for exercise, and indulged in dinners out with drinks and desserts.
The AI suggested small tweaks on rest days, such as increasing step count or slightly reducing high-calorie additions like cheese or peanut butter. O'Meara affirmed this approach, noting that recognising consistent patterns of exceeding calorie goals makes course-correction easier. However, she cautioned that AI might recommend cuts based on a pattern rather than a one-off event, and stressed the importance of ensuring calorie intake isn't too restrictive.
Emotional Intelligence and Tone Found Lacking
Pugachevsky reported that the AI's tone could feel "overwhelming" and "demoralising," particularly when it suggested cutting back on foods like goat cheese after a strenuous workout. She noted ChatGPT often couched critical feedback with "faux cheerfulness," which felt "condescending."
"AI is essentially a robot, it's going to give you that robot response," O'Meara said. She contrasted this with a dietitian's approach, which would consider the emotional and social benefits of a meal out and might suggest compromises, like opting for a mocktail, rather than outright restrictions.
Experiment Yields Results but Highlights Trade-offs
By the end of the two weeks, Pugachevsky reported becoming better at choosing protein-rich meals and believed the experiment helped her lose one pound. She was impressed by the ease of tracking and the clarity of the action steps provided.
However, she concluded that the process made her feel like "a computer speaking to another computer," reductively quantifying her life into metrics. The experience underscored a trade-off between data-driven optimisation and a more intuitive, less restrictive approach to diet and lifestyle.