Calibrated SignalHard science. Honest signal.

Editorial Process

How I Use AI

I use AI. Here is exactly how, exactly where, and what it never touches.

I have been using AI since graduate school in 2017, back when we called it machine learning. It is a tool I have used professionally for years, and I would rather show you my process than let you guess at it.

How an article actually gets made

  1. 01

    I choose the topic

    It comes from something I saw on shift, something in my own labs and scans, or a longevity claim that is suddenly everywhere and deserves to be taken apart. Nothing else picks it.

  2. 02

    I trace the claim back to the paper

    A claim goes around on a podcast, a viral post, or a supplement label. I go find the actual study underneath it, which is very often not the study people think they are citing.

  3. 03

    I read the primary literature myself

    The paper itself, not a summary of it. Study design, sample size, endpoints, who paid for it, and what it does not show. If I have not read it, it does not get cited.

  4. 04

    I dictate the argument

    Out loud, in my own words: what is solid, what is not, what order it goes in, and where I am genuinely uncertain.

  5. 05

    AI drafts from that outline

    It also helps me brainstorm framings and surface related papers with nuances I might otherwise miss.

  6. 06

    I verify and cite every source

    Every claim traces back to a real, named study with a PMID you can check. If I cannot verify it, it does not run.

  7. 07

    I edit until the judgment is mine, then I sign it

    The framing, the emphasis, and the conclusions are mine to defend. My name goes on it, and so does my license.

What AI does here

  • Drafts from an outline I dictated
  • Brainstorms angles and structures
  • Helps me surface relevant literature
  • Organizes sources and formats citations

What AI never does here

  • Decides what I believe or what I conclude
  • Decides what is true
  • Supplies a fact, a number, or a citation I have not personally checked against the source
  • Replaces clinical judgment

AI cannot be held accountable for errors. My medical license can.

What none of this changes

  • Every claim is cited to primary literature, with a PMID you can check.
  • Nothing is fabricated. If a study does not say it, I do not write it.
  • Uncertainty gets flagged as uncertainty.
  • No sponsors, no affiliate-driven recommendations, and no supplement line. See Disclosures.
  • When I get something wrong, I correct it publicly. See Corrections.

If you would rather judge the work by whether the citations hold up, that is the standard I would choose too. The framework I use to weigh them is on How I Evaluate Evidence.

Hard science. Honest signal. No sponsors.