When was the last time you used an AI tool in your practice? Did it go well? Or did it give lackluster results, leading you to quickly abandon it as a viable solution?
According to Dr. Sunjya Schweig, many functional medicine practitioners have tried using general AI tools (like standard ChatGPT) for medical brainstorming and have walked away unimpressed. They received generic, textbook answers or shallow advice that didn’t fit the complex realities of patient care.
But, says Dr. Schweig, often the issue isn't the AI—it's how we utilize it. General AI is like a brilliant resident who has memorized every medical textbook but has zero real-world context on how you practice.
But that doesn’t mean AI is worthless. It just means you have to know how to use it.
Demystifying LLMs
Before we dig into how AI tools can help you in your practice, we need to understand how they work—and specifically large language models (LLMs).
LLMs don’t pull information from a rigid, deterministic database, like an Access file or SQL database does. Rather, they work through probabilistic next-word prediction similar to the predictive text function on your cell phone.
As Dr. Schweig explains, LLMs “take in as much information as they can find—basically, the entire web and millions of books. And then they predict the most likely next word based on patterns from everything they've read.”
This probabilistic next-word prediction mechanism is incredibly powerful and helps make AI a valuable tool for you as a functional medicine practitioner. But it’s a tool that comes with both strengths and weaknesses.
On the one hand, an LLM can generate encyclopedic outputs that synthesize and apply all of the latest medical research to your situation at the click of a button. On the other hand, hallucinations can occur when the model's predictive logic prioritizes word probability and plausible formatting over factual accuracy. (In other words, an LLM will confidently lie to you.)
It’s this two-edged sword that we need to learn to handle well if we want AI to help us be more efficient and effective in caring for our patients.
So, how can you go about using AI tools in a way that maximizes their benefits and minimizes their drawbacks? According to Dr. Schweig, the answer lies in utilizing a “clinical stack.”
Most clinicians who use AI at all use only 20% of what it can do. This isn't laziness; it's a training gap.
Without structured learning, practitioners default to generic prompts, single-platform dependence, and passive tool consumption.
But you don’t have to let a lack of AI fluency hold you back.
In the newest Kharrazian Institute Master Class, “Clinical Applications of AI,” Dr. Sunjya Schweig distills what works and what fails when AI meets functional medicine practice.
You'll learn:
- How to build custom AI systems that encode your clinical expertise.
- How to master prompts that generate professional-grade outputs.
- How to maintain legal and ethical guardrails while using AI.
This is not motivational content. It's a working system taught by someone using it daily in clinical practice. Click here to learn more.
Using the Right AI Tools (aka the “Clinical Stack”)
A clinical stack is simply a combination of AI tools used in tandem. When utilized well, a stack of AI tools will outperform a single AI model, since each tool within the stack will specialize in its area(s) of strength.
Five components of a clinical stack
The possibilities for what to add to your stack are myriad, but Dr. Schweig suggests that a complete clinical stack has five distinct functional lanes:
- General LLMs - General workhorses for multi-system reasoning, data analysis, and patient-facing communication drafts.
- Clinical research tools - Specialized, medical-grade platforms trained directly on peer-reviewed literature.
- Ambient scribes - Medical listening tools that process conversations during visits and automatically generate structured notes.
- Knowledge management - Tools that allow you to upload your own curated sources to create an AI assistant grounded strictly in the information you provide.
- Specialized utility tools - Tools for voice dictation, presentation building, meeting capture, etc.
Dr. Schweig’s clinical stack “starter kit”
And if you’re still feeling overwhelmed, Dr. Schweig suggests starting with this specific set of tools:
- ChatGPT - Best for building custom GPTs
- Claude - Ideal for long-context reasoning, multi-document synthesis, and building reusable skills
- OpenEvidence - Go-to tool for free, HIPAA-compliant, evidence-linked clinical decision support
- Perplexity - Best for real-time web search and current guideline lookups
- Heidi - Preferred ambient scribe due to its deep customizability for functional medicine templates
- Gemini Notebook - Essential for personal knowledge management
- Wispr Flow - System-wide voice dictation to avoid manual typing
- Gamma - AI presentation builder to rapidly generate slide decks
Setting up your clinical stack is the first step toward making AI work in your clinical setting. But that’s only half the battle. The information and instructions you provide to your AI tools are just as critical.
Within the next ten years, most routine clinical and administrative tasks will be AI-assisted—which means if you are not already thinking about how to use AI in your practice, you are behind the curve.
But you don’t have to stay there.
When you register for the Kharrazian Institute Master Class, “Clinical Applications of AI,” you’ll learn how to use AI to gain efficiencies in your practice while still providing quality care.
Clinicians who don't build fluency now won't lose their jobs to AI; they'll lose ground to someone like them who learned faster.
The Anatomy of a High-Quality AI Prompt
In the 1980s, a common computing phrase was “garbage in, garbage out.” While computers may have gotten orders of magnitude more powerful since then, the principle remains true. The simple reality is that the quality of the input has a direct correlation to the quality of the output.
In order to get the most out of an AI agent, therefore, you must provide it with high-quality information and specific, well-thought-out prompts. After all, it doesn’t know what you want—you have to tell it.
The anatomy of a well-written AI prompt—one likely to give you a useful response—includes these components:
- Role - Tell the AI exactly who it is (e.g., “You are an expert functional medicine physician specializing in gut-brain axis disorders…”).
- Context - Provide detailed, anonymized patient parameters and clinical history. Accurate, thorough information is key.
- Task - Assign a highly specific, bounded task (e.g., “Generate a differential diagnostic list of three potential root causes, ranked by likelihood, with peer-reviewed mechanisms explained”) as opposed to a general question or explanation.
And if you want to go a step further, consider adding self-checks like the following:
- Ask the AI to cite its sources.
- Require the AI to justify its decisions.
- Instruct the AI to assign a confidence value from zero to 10 (with zero being not confident at all about its answer and 10 being extremely confident).
Adopting a clinical stack and implementing these simple prompting tips will drastically improve the way you use AI in your practice. But remember that mistakes and hallucinations can still creep in. The answer, of course, is to always validate everything an AI tool tells you.
At the end of the day, you are responsible legally and ethically for patient care. Your clinical stack can greatly help with administration, research, and other time-consuming tasks, but the final decision on what to do with the information it gives you must firmly reside with you.
Clinical Applications of AI: A Competency-Based Master Class for Integrative and Functional Medicine Practitioners taught by Sunjya Schweig, MD
With this post, we have laid out a basic framework for you to start using AI in your practice. If you want to go deeper, consider attending our latest KI Master Class, “Clinical Applications of AI,” taught by Sunjya Schweig, MD. When you take this course, you'll learn:
- How to prompt with precision using the AIM (Ask, Inform, Map) framework
- How to integrate ambient scribing into workflows that save 30+ minutes daily
- How to build custom AI systems that encode your clinical expertise
- How to maintain patient safety through HIPAA compliance and error vigilance
The curriculum is built on clinical evidence, case studies from practitioners already using these tools, and iterative testing across multiple platforms.
Don’t get left behind in the AI game; the time to level up your skills is now. Click here to register today.
Frequently Asked Questions
Is it safe to give AI tools patient information?
Yes, but only if you're careful. Never enter identifiable patient details into general AI tools unless the platform is HIPAA-compliant. Always anonymize patient data before typing it into any prompt, and remember that you're still legally and ethically responsible for how that information is used and protected.
Why does ChatGPT give me generic or unhelpful answers?
General AI tools work by predicting the most likely next word based on patterns from everything they've read. Without specific instructions, they default to broad, textbook-style answers. The fix isn't a better tool—it's a better prompt. Give the AI a clear role, detailed context, and a specific task, and the quality of the response improves dramatically.
Do I need to use multiple AI tools, or is one enough?
One tool can help, but a "clinical stack" of several specialized tools will outperform any single AI model.
How do I know if an AI's answer is trustworthy?
By always validating the information yourself. AI can speed up your research and administrative work, but you make the final clinical call.
Where can I learn a step-by-step system for using AI in my practice?
If you want more than the basics covered here, the Kharrazian Institute Master Class, "Clinical Applications of AI," taught by Sunjya Schweig, MD, walks you through a full framework—from prompting techniques to building custom AI systems that reflect your own clinical expertise. It's built for practitioners who want real, usable skills rather than a general overview. Click here to register today.
