How to Use Artificial Intelligence to Get Healthy: A Practical Guide
Artificial intelligence is one of those topics that tends to divide people into two camps: those who embrace it enthusiastically and those who want nothing to do with it. Everyone else lives somewhere in the middle, curious but unsure how to use it in a meaningful way.
In my practice, I use artificial intelligence regularly. Not as a decision maker, and certainly not as a replacement for medical expertise, but as a tool to support ideas, structure, and planning. I also teach patients how to use it, because when used correctly, it can be helpful for things like building meal plans or creating workout routines.
But there is an important distinction that needs to be clear from the start. Artificial intelligence is not an expert. It is a tool. It generates responses based on patterns in data, and those responses are only as good as the information you give it and the way you ask your question. It can make mistakes. Sometimes obvious ones.
Used carefully, it can be useful. Used blindly, it can mislead. So this week, I want to talk about how to use AI to support health goals and how to avoid the pitfalls.
AI Platforms and Their Uses
There are multiple artificial intelligence platforms available, each marketed slightly differently.
ChatGPT is an OpenAI product designed for conversational responses, content generation, and general problem solving. It is the most widely used platform and the one I rely on because it is the one with which I am most familiar.
Google Gemini integrates with Google’s search infrastructure to provide answers with real-time context, making it useful when someone is interested in current information.
Microsoft Copilot is embedded within Microsoft products, making it a natural fit for people who already work within the Microsoft ecosystem.
Claude is best known for its writing and copy editing capabilities.
There are several other platforms as well, but these are the most well-known.
Why Expert Input Comes First
One of the most common mistakes I see is people assuming that artificial intelligence is an authority or content expert. It is not. It is a computer program.
There is a growing body of literature showing that AI systems can produce incorrect, incomplete, or fabricated information, particularly when asked to summarize research or provide clinical guidance.1 This is not a rare occurrence. It is a known limitation.
I’ve experienced this with literature searches. Platforms have provided me with summaries and data from papers that don’t even exist.2 Because I have subject matter expertise, I can identify when something seems off. People without expertise in the content they are searching will not detect this because the AI outputs are designed to sound cogent, confident, and conversational.3,4 This is why expert input must come first.
If you are trying to lose weight, manage a chronic condition, or improve your metabolic health, you need a clear, evidence-based framework before you ever open an AI platform. Once that foundation is established, AI can help operationalize the plan. It cannot reliably build the plan for you.
How to Talk to AI: the RTCO Framework
Once you have the expert information, the next step is learning how to communicate with AI effectively. This is where most people struggle, not because the technology is difficult, but because they do not know how to structure the question.
Most platforms draw from vast amounts of data across the internet. The more specific and more structured your question, the more you strategically narrow the search and produce relevant, accurate output. While there are several prompting frameworks, I prefer and teach is the RTCO model: Role, Task, Context, Output.
Role
The role prompt tells the AI model what it is. You are assigning it a job title. The more specific you are, the better. For example, if you want data on food choices, you don’t want just “a nutritionist”. You want a registered dietician with 20 years of experience who specializes in weight management and metabolic health and who works with adult patients. Once the role is defined, the model knows where to focus the search and find the most relevant information.
Task
The task is what you are asking the model to do. This is the specific job you are assigning. Are you asking it to create something? Summarize something? Revise something? Be direct and specific. Sticking with the above example, the task would be: “Create a 7-day meal plan with 3 full meals and 2 snacks.” “Help me with food” is not a helpful task.
Context
This is one of the most important components and the one most people skip. Context is where you provide all of the relevant details and share expert data. In our 7-day meal plan example, context would include the caloric targets, macronutrient goals, dietary restrictions or food allergies, cooking ability, time constraints, where to obtain the food data, and relevant health conditions. The more detail you provide, the more useful the output.
Output
Tell AI what you want the end product to look like. Do you want a table? A bulleted list? A downloadable file? A simple response in the chat? Specifying the format saves time and ensures the response is usable.
One Final Tip
Before you submit your prompt, ask the model if it has any additional questions to complete the task with 95% confidence. This allows models to fill in any remaining gaps, so you are not going back and forth trying to correct the output.
These models work best with the first 2 or 3 prompts. After three turns, the model loses the focus of your original RTCO instructions.5 Adding “with 95% confidence” prevents two things. The first is the back-and-forth multi-turn conversation that reduces accuracy.6 The second is over-confidence. When you add the level of confidence you desire, you force the model to invoke a self-correction so that it does not produce an overly confident, inaccurate output.7
A Practical Example
The easiest way to demonstrate how I use this in practice is to use a fictional patient as an example.
Sofie is a 47-year-old female who wants to improve her overall health. She has prediabetes, elevated lipids, and is overweight. She is unsure what to eat and unsure what kind of exercise she needs. After a thorough interview and data collection, Sofie and I determine she needs to limit her calories to 1500 kcal/day. Her macronutrients are 100 grams of protein, 25 grams of fiber, and no more than 10 grams of saturated fat per day in order to meet her health goals. Her body composition reveals an elevated visceral fat level and reduced muscle mass, particularly in her upper body.
At present, she does not exercise regularly because she is very busy with work, managing her household, and taking care of her kids. She has a helpful husband, but he travels a lot for work. She is not interested in outside childcare options. She does not smoke, has a glass of wine every night, and will occasionally use THC gummies to help her fall asleep. She uses a food delivery service several nights per week. She would like to cook more but is unsure how to make healthy meals and is concerned about the time commitment. She has no food restrictions or allergies but does not like onions. She has two boys, ages 12 and 13, who are not picky eaters.
(I promise Sofie is not anyone in particular. I took some of the most common issues I see across many patients and combined them into one fictional person. Sofie also happens to be the name I had picked out for my second child, who turned out to be a boy.)
While I could demonstrate several different prompts for Sofie, I will focus on one for nutrition and meal planning for the sake of length of this post.
Role: You are a registered dietician with 20 years of experience, and you specialize in weight management and metabolic health.
Task: Produce a 7-day meal plan that includes 3 meals and 2 snacks per day.
Context: Total daily calories cannot exceed 1500 kcal on average and should achieve 100 grams of protein, 25 grams of fiber, and no more than 10 grams of saturated fat per day. The meals cannot take more than 45 minutes to prepare. They should follow the MyPlate portion guidelines. Use the USDA FoodData Central database for caloric and macronutrient data of foods. Avoid onions in all meals.
Output: Produce an Excel file with the days of the week across the top row and the meal and snack type in the first column.
What other questions do you have to complete the above task with 95% confidence?
This prompt will give Sofie a practical starting point for a 7-day meal plan. She can then fine-tune the suggested meals to her and her family’s preferences and use additional prompts to request recipes.
Sofie would also benefit from a Zone 2 exercise plan that helps her make exercise a regular part of her life. She ultimately wants to get to 500 MET minutes of physical activity per week. A separate set of prompts can help her map out how to achieve that goal gradually, within the real constraints of her life.
While the outputs will not be perfect, they provide an excellent starting point and help translate clinical recommendations into something practical and actionable.
The Limitations You Need to Understand
While AI is very helpful, it has real limitations and understanding them is essential.
AI does not “understand” anything. It does not verify information in real time. It predicts language based on patterns in data. This is fundamentally different from clinical reasoning, and confusing the two can cause harm.8
AI can deliver incorrect information with complete confidence. It fabricates information, a phenomenon commonly referred to as “hallucinations.” It can oversimplify complex situations in ways that sound reasonable but are not. If you are not in a position to critically evaluate what it produces, you should not rely on it for decisions that affect your health.9
There is also a behavioral limitation worth acknowledging. AI can create the illusion of progress. Generating a great meal plan does not mean you will follow it. Creating a detailed workout schedule does not mean it will be executed. The plan is only as valuable as the action behind it.
There is also an environmental cost that rarely gets mentioned in conversations about AI. These platforms run on massive data centers that consume extraordinary amounts of electricity and water. Water is used to cool the servers, and the energy demands are significant enough that several major AI companies have had to revisit their carbon footprint as usage has increased.10 This does not mean you should not use AI. But it is worth being aware that every query has an environmental impact, and the collective impact of AI at scale is a legitimate concern.
There is also risk of dependency. Instead of building the skills to think through a problem, some people default to asking AI for every decision.11 That is not the goal. The goal is to use it as a tool that supports action, not one that replaces thinking.
Finally, I want to return to something I said at the beginning of this post. Yes, I use AI regularly and teach my patients how to use it. But I also catch its mistakes regularly because I have the training to recognize them. Most patients do not have that advantage. This is not an argument against using AI. It is an argument for always starting with expert guidance first.
Bottom Line
Artificial intelligence can be a useful tool in health, but only when it is used the right way.
It should never replace expert guidance. It should never be the sole source for a medical decision. What it can do is take a well-defined plan and turn it into something practical and actionable, which is valuable for anyone just beginning their health journey.
But it is still just a tool and just like any tool, it is only as good as the person using it. The quality you get out of out of it is dependent on what you put in.
Disclaimer: Even though I’m a doctor, I’m not your doctor—and reading this blog does not establish a doctor–patient relationship. This information is intended for general educational purposes only and should not be taken as personalized medical advice. Always speak with your own healthcare provider before making decisions about your health.
References
- Takita H, Kabata D, Walston SL, et al. A systematic review and meta-analysis of diagnostic performance comparison between generative AI and physicians. NPJ Digit Med. Mar 22 2025;8(1):175. doi:10.1038/s41746-025-01543-z
- Ji Z, Lee N, Frieske R, et al. Survey of hallucination in natural language generation. ACM computing surveys. 2023;55(12):1-38.
- Sikha VK, Siramgari D, Korada L. Mastering prompt engineering: Optimizing interaction with generative AI agents. Journal of Engineering and Applied Sciences Technology SRC/JEAST-E117 DOI: doi org/1047363/JEAST/2023 (5) E117 J Eng App Sci Technol. 2023;5(6):2-8.
- Bender EM, Gebru T, McMillan-Major A, Shmitchell S. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 2021:610-623.
- Liu NF, Lin K, Hewitt J, et al. Lost in the middle: How language models use long contexts. Transactions of the association for computational linguistics. 2024;12:157-173.
- Laban P, Hayashi H, Zhou Y, Neville J. Llms get lost in multi-turn conversation. arXiv preprint arXiv:250506120. 2025;
- Press O, Zhang M, Min S, Schmidt L, Smith NA, Lewis M. Measuring and narrowing the compositionality gap in language models. 2023:5687-5711.
- Shao A. New sources of inaccuracy? A conceptual framework for studying AI hallucinations. Harvard Kennedy School Misinformation Review. 2025;
- Omar M, Sorin V, Collins JD, et al. Multi-model assurance analysis showing large language models are highly vulnerable to adversarial hallucination attacks during clinical decision support. Communications Medicine. 2025;5(1):330.
- Bashir N, Donti P, Cuff J, et al. The climate and sustainability implications of generative AI. 2024;
- Gerlich M. AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies. 2025;15(1):6.