The Problem With "Just Try It"
Most people run their supplement experiments entirely by feel. They start taking something, tell themselves it's working after a week, and add it permanently to their stack. Or they feel nothing after two weeks and stop. Neither of these is an experiment — both are just guessing with extra steps.
The problem is that your memory is unreliable, your expectations create real physiological responses, and the world keeps changing around you while you're testing. Without a structured design, you're not getting answers. You're collecting impressions.
Here's how to do it differently.
Related: Try our Experiment Builder to test this yourself. Also worth reading: How Long to Run a Health Experiment: Duration Guide and our The Complete Guide to Supplement Tracking.
Start With a Hypothesis
An experiment without a hypothesis is just random tracking. Your hypothesis forces you to be specific before you start — which protects you from changing your interpretation of results after the fact.
A good supplement hypothesis has three parts: the intervention, the outcome, and the direction.
Example hypothesis: "Taking 600mg ashwagandha daily for 30 days will increase my average HRV by at least 5ms compared to my baseline."
This is specific, measurable, and defined before you start. You either meet the threshold or you don't.
Write it down. Don't let yourself revise it mid-experiment when the data doesn't look how you expected.
Choose ONE Variable
This is the rule most people violate. You cannot test two supplements simultaneously. You cannot change your sleep schedule while testing a supplement. You cannot start a new exercise protocol during your active phase.
If you change multiple things and see improvement, you'll never know which change caused it. If you see no improvement, you won't know whether the supplement failed or whether something else interfered.
One variable. Every time.
If you want to test multiple supplements, test them sequentially with washout periods between each trial. It takes longer, but the answers you get are actually worth having.
The Four-Phase Structure
Phase 1: Baseline (7–14 Days)
Do nothing new. Track everything you plan to track during the experiment, using the exact same methods and timing. This is your control data — the "before" you'll compare everything against.
During baseline, track:
- Your target outcomes (e.g., daily energy rating, HRV from your wearable, sleep quality score)
- Potential confounders (stress level, exercise type and duration, caffeine intake, alcohol)
- Any existing supplements you're already taking
Fourteen days is ideal. Seven is the minimum. The longer your baseline, the more reliable your comparison becomes.
Phase 2: Active Phase (21–30 Days)
Introduce only the supplement you're testing, at a consistent dose and timing. Track exactly the same metrics you tracked during baseline. Nothing else changes.
Most chronic-use supplements need at least three weeks to show full effects. Acute-acting compounds (like L-theanine or caffeine) show effects sooner, but you still need enough data points to distinguish a real effect from day-to-day variation.
Protocol template:
- Dose: [supplement] at [mg] at [time of day] with [food or fasted]
- Duration: 28 days
- Primary outcome: [your specific metric]
- Secondary outcomes: [2-3 additional metrics]
- Confounder log: Stress (1-5), alcohol (yes/no), exercise (type + duration)
Phase 3: Washout (Optional but Useful)
If you plan to retest, or if you want to confirm the effect was real, a washout period lets you return to baseline before the next trial. Washout length depends on the compound's half-life — more on this in a separate post on washout periods.
Phase 4: Analysis
Compare your active phase averages to your baseline averages. Look for:
| Signal | Threshold to Consider Meaningful |
|---|---|
| Subjective ratings (1-10) | 0.75 point change or greater |
| HRV | 5ms change or greater |
| Sleep quality score | 5-point change or greater |
| Energy (subjective) | Consistent 1-point change across ≥75% of active days |
Single-day spikes don't count. What matters is whether the average shifted and whether it stayed shifted.
Controlling for Confounders
Even with a single-variable design, things will happen during your experiment that could mask or mimic a supplement's effects. Travel, illness, work stress, a bad week at the gym — all of these can move your metrics independently of whatever you're testing.
Log them. Every day. A simple 30-second confounder check keeps your post-hoc analysis honest.
If you have an unusually stressful week during your active phase, note it. If you traveled across time zones, mark those days. When you're analyzing results, you can look at clean days versus confounded days separately.
A 5-point rating scale is enough for most confounder tracking. Rate your stress (1=low, 5=high), note any unusual events in a single sentence, and move on. Precision matters less than consistency.
When to Conclude
The experiment ends when your active phase is complete — not when you feel confident enough to stop early (which is usually when things are going well and you'd rather not risk a negative data point).
Stopping early when the effect looks promising is how confirmation bias sneaks in. Commit to the timeline before you start and hold to it.
If you get sick or face a major life disruption during the active phase, you have two choices: log it as a confounded experiment and note the limitation in your conclusions, or restart the active phase entirely once things normalize.
What You're Actually Building
One well-designed experiment is worth more than a year of casual supplementation. It gives you a personal data point that population studies can't provide. Aggregated research tells you what worked on average for a group of strangers. Your experiment tells you what happened in your body, at your dose, in your life.
That's the difference between taking supplements based on podcast advice and taking them based on evidence.
Be the first to try Prova
We're building an app to track whether supplement experiment design actually works. Join the waitlist.