Using ChatGPT or AI to build a personal protocol

modoodom said:

MTSpace said:

Has anyone turned to AI when putting together their protocol? Because I have premium subscriptions to Gemini and ChatGPT, I'm focusing on getting my baseline health markers sorted out first, ahead of adding GLP-1s, supplements, and other extras.

I've uploaded 15 years of bloodwork, symptoms, and notes from my current doctors.

What has your protocol given you that actually proved useful?
Click to expand...

Yes, and AI has genuinely reshaped how I handle all of this.

The most useful part is treating it as a clinical thinking partner rather than a chatbot. I built out a custom Claude skill with a persistent health profile that I keep updated. It covers:

  • Labs and biomarkers: hormones, lipid panels, metabolic markers, kidney/liver function, hematocrit
  • Wearable data: resting HR, HRV, sleep, steps, VO2 Max, active calories, blood pressure trends
  • Training history: complete resistance training logs plus running/cardio data exported from my apps
  • Current protocol: supplements, medications, start dates and dosing
  • Nutrition snapshot: protein sources, macros, timing

Since the skill persists, Claude already has my full context whenever I start a conversation. No more re-explaining myself every time. It reads the data, flags gaps I'd missed, and points out what I still lack before I make sensible calls on the next layer.

Your "baseline first" approach is spot on. That's the direction the AI pushed me in as well — don't chase advanced add-ons before the foundation is solid.

Happy to go into how I built it if that'd help.
Click to expand...
I can't speak to Claude since I haven't used it, but ChatGPT holds 15 years of my medical records, my medication list, the summaries my doctors give me after visits, data from every wearable I've used (Fitbit and Apple devices), nutrition logs, and 5-7 years of workouts. What surprises me is how much it loses track of. Sometimes it blends my father-in-law's blood markers from his heart attack into mine. Other times it mixes in my dad's post-stroke bloodwork, or my wife's cancer bloodwork. It's frustrating, because every analysis it gives me has to be checked.

Maybe switching to Claude is what I should do.
 
On my own local hardware, I run an agent powered by an uncensored model. When I ask, he never holds back from saying [[Q1]]. Even though my full medical history sits in his memory, my main use for him is decoding terminology I'm unfamiliar with, since I'm not a doctor. I'll request general guidelines and protocols, but before purchasing or injecting anything, I always verify the information.
 
Each AI I've tried places its own limits on the questions you're permitted to ask. Claude comes across as the strictest of the bunch, while ChatGPT holds back on certain data. As for DeepSeek and Mistral, they appear to have the least in the way of guardrails.



Since I pay for Claude, I do put it to work reviewing some of my data. After I began ADHD medication, my blood pressure came back elevated, so I set up a Google Doc that Claude could read. As time went on, I kept adding more information — my Garmin numbers, for instance, along with records of when I used a GLP-1 and the dose involved — and it has turned out to be pretty helpful.
 
Earlier today, ChatGPT gave me incorrect information about the weather twice for the area where I was backpacking. At first it said the days would be sunny and warm, then it changed to overnight rain and snow. I only found out the truth once I looked up the actual forecast, which showed totally different temperatures and only a small chance of rain. When I pointed this out, ChatGPT responded that it is simply an AI that makes mistakes. Because of this, I have become wary of AI when it quotes medical research data, applies that data to my own situation, and offers advice. On the other hand, it does put together a solid exercise plan for me.
 
MTSpace said:

Has anyone turned to AI when putting together their protocol? Because I have premium subscriptions to Gemini and ChatGPT, I'm focusing on getting my baseline health markers sorted out first, ahead of adding GLP-1s, supplements, and other extras.

I've uploaded 15 years of bloodwork, symptoms, and notes from my current doctors.

What has your protocol given you that actually proved useful?
Click to expand...

Hmm, a notebook and a pen
 
slimmingmouse65 said:

Earlier today, ChatGPT gave me incorrect information about the weather twice for the area where I was backpacking. At first it said the days would be sunny and warm, then it changed to overnight rain and snow. I only found out the truth once I looked up the actual forecast, which showed totally different temperatures and only a small chance of rain. When I pointed this out, ChatGPT responded that it is simply an AI that makes mistakes. Because of this, I have become wary of AI when it quotes medical research data, applies that data to my own situation, and offers advice. On the other hand, it does put together a solid exercise plan for me.
Click to expand...

Being careful is definitely the right call, and whenever something seems off I usually have it cite where it got the information. That said, the more time you spend with these tools, the clearer it becomes which tasks suit them and which ones don't.

For example, asking about the history of a hiking spot along with its plants and wildlife seems like a reasonable use. But for something like weather, which depends on live data, a general chatbot would be far from my first choice.

Exercise information is everywhere online, so I'd assume the training data on that topic is solid, and I do rely on it for that. My one gripe is that the answers can get more complicated than what I actually need.
 
I've put in some hours with ChatGTP, Claude, and Gemini. Each one brings strengths and weaknesses. My take:

For image generation, ChatGTP does a solid job, but it's a lying MF. I've watched it fabricate things just to avoid looking wrong or to conserve resources. Nothing it tells me gets my trust.

As an all-around model, Claude is likely the strongest. It's upfront about what it can't do and suggests alternatives. Data limits are easy to bump into even when you're paying, though I still consider it worthwhile. Where it falls short: it's the tightest of the three when it comes to medical topics.

When I need something off the beaten path medically, Gemini is what I reach for. It appears the most open to discussing peps and other options that aren't exactly legally prescribed. Nothing problematic (lying, etc.) has come up for me so far, and I've noticed I'm leaning on it more lately.
 
5byfive said:

I've put in some hours with ChatGTP, Claude, and Gemini. Each one brings strengths and weaknesses. My take:

For image generation, ChatGTP does a solid job, but it's a lying MF. I've watched it fabricate things just to avoid looking wrong or to conserve resources. Nothing it tells me gets my trust.

As an all-around model, Claude is likely the strongest. It's upfront about what it can't do and suggests alternatives. Data limits are easy to bump into even when you're paying, though I still consider it worthwhile. Where it falls short: it's the tightest of the three when it comes to medical topics.

When I need something off the beaten path medically, Gemini is what I reach for. It appears the most open to discussing peps and other options that aren't exactly legally prescribed. Nothing problematic (lying, etc.) has come up for me so far, and I've noticed I'm leaning on it more lately.
Click to expand...

Opus is appealing, but Sonnet handles plenty of tasks just fine while consuming far fewer tokens.

I wasn't aware Gemini had fewer guardrails....Deepseek had been my choice for that, so I'll try it out.
 
Back
Top