SUPPLEMENTSFebruary 12, 2026 · 29:11

I Built a Custom GPT to Manage My Supplements

James Grant

James Grant

Co-host, HealthStacks

SupplementsCustom-gptBlood-testsPersonalizationChatGPT Custom GPTs

Show Notes

What You'll Learn

  • How to build a custom GPT that acts as your personal supplement advisor
  • A prompting technique that compensates for gaps in your own knowledge
  • Why stacking similar supplements can work against you (and how AI catches it)
  • How to design 8-12 week experimental review protocols with pass/fail criteria
  • When to default to your doctor and when AI is the better prep tool

Timestamps

  • 00:00 Introduction
  • 00:30 HealthStacks show intro
  • 01:06 Origin story -- the scatter-gun approach before AI
  • 02:10 Why experimentation loops matter
  • 03:21 Key principle: start with goals, not supplements
  • 04:02 Screen share: building the Custom GPT
  • 06:19 Three health goals explained
  • 08:25 System instructions deep-dive
  • 08:57 Prompting technique: "ask me clarifying questions"
  • 10:14 Getting started advice for beginners
  • 14:17 Case study: why James removed glycine
  • 17:10 "Top 5 interventions" live demo
  • 18:17 Role prompting: specialist vs. generalist
  • 21:34 Results walkthrough
  • 24:08 Building a full daily stack with dosing schedule
  • 27:04 The future of the supplement industry
  • 29:13 Failures: when the same AI gives conflicting advice
  • 30:34 Final stack review
  • 32:50 Wrap-up and takeaways

Key Takeaways

  • Start with goals, not products -- Define 2-3 clear health goals before asking AI about any supplement
  • "Ask me clarifying questions" is a prompting superpower -- Adding this single instruction to any prompt compensates for incomplete knowledge
  • AI catches overlap that humans miss -- James was stacking three or four supplements that all served the same stress-reduction purpose
  • One blood test is a starting point, not an answer -- Trend data across multiple tests over time matters far more
  • Build in review protocols -- Set 8-12 week experimental periods with specific pass/fail criteria tied to lab markers
  • Context shapes AI output -- The same AI gave conflicting ashwagandha advice in different sessions
  • Default to a doctor, use AI to prepare better questions

Resources Mentioned


This episode is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional before starting, stopping, or changing any supplement or treatment. AI tools are decision-support aids, not replacements for professional medical guidance.

Never Miss an Episode

Get new episodes and workflow breakdowns in your inbox.

Interviewed by

Will Read
Will Read
Co-host

Up Next

TRAINING

This AI Coach Knows Me Better Than My Trainer

Samuel Bosch, Founder, KeepOn