An AI Diet Assistant with Gemini: Possibilities and Limitations

Can artificial intelligence act as a personal assistant for everyday food choices? This experience with Google’s Gemini model explores that question not simply by creating a meal plan, but by responding to needs that change throughout the day. Starting with a discussion of goals, lifestyle and food preferences, the process develops into an interactive assistant that evaluates meals, suggests alternatives and answers questions about food. The value lies less in a one-off answer than in an ongoing dialogue.

Personalisation starts with the context provided to the assistant. In this experience, goals, activity levels, likes and dislikes were considered together to create a meal plan that accounted for training and rest days. This approach illustrates how recommendations can reflect a user’s daily routine rather than offering everyone the same plan. However, an answer that appears personalised cannot be assumed to be medically appropriate. An AI-generated plan should not be viewed as a prescription that replaces professional assessment.

Visual interaction was one of the practical aspects of the experience. Meal photos were used to generate approximate calorie, protein, carbohydrate and fat values, and these estimates were added to the daily log. This created a recording process that began with a photo rather than requiring every meal to be described in text alone. However, an image cannot reliably show portion weight or every ingredient added to a dish. Results should therefore be treated as estimates, not measurements, and supplemented with information on quantities and ingredients whenever possible.

Daily tracking went beyond listing previous meals. In this experience, options for subsequent meals were suggested based on recorded intake and the daily calorie target. Questions about whether a product was suitable, accompanied by a photo taken in a shop or restaurant, were also answered with alternative suggestions. The key distinction is the information on which a recommendation is based: an incomplete meal log or an inaccurate portion estimate can affect the assessment of what remains of the daily target. This process should not be confused with an independent, continuously operating health monitoring system.

Asking about a food’s composition, comparing meal options or reconsidering a daily plan are among the accessible uses of AI assistants. At X Mind Solutions, we view this example as a use case showing how context-rich interaction can support everyday decisions. When it comes to nutrition, however, convenience and reliability must be assessed separately. Professional guidance should remain the foundation, particularly in cases involving medical conditions, allergies or specific dietary needs; the assistant should serve as a tool for organising information and preparing for decisions.