Blog

How to Evaluate AI Supplement Recommendations

Unfair Team • January 13, 2026

An AI recommendation is not a prescription. It is a suggestion generated by matching your stated inputs (goals, history, constraints) against a model trained on clinical evidence and user patterns. Some suggestions will be excellent. Others will be generic, poorly timed, or outright inappropriate for your situation. The skill is knowing how to tell the difference before you open the bottle.

The pre-trial checklist

Before you try any AI-recommended supplement, run it through these five checks. If it fails on any of them, do not proceed until the issue is resolved.

1. Does the rationale make sense to you?

A good recommendation comes with a stated reason. "Magnesium glycinate 300mg before bed, because you reported disrupted sleep and your current stack has no sleep-support component" is a rationale. "Try magnesium for better wellness" is not.

What to look for:

• The recommendation names a specific supplement, dose, and timing window.
• It connects to your stated goal (not a different goal you did not ask about).
• It explains why this supplement was chosen over alternatives.

Red flag: The recommendation does not explain its reasoning, or the reasoning sounds like a marketing claim rather than an evidence-based argument.

2. Is the dose within studied ranges?

AI systems can sometimes recommend doses that are either too low to produce a meaningful effect or too high relative to the evidence base. Check the recommended dose against a credible reference.

Quick reference for commonly recommended supplements:

If the AI recommends a dose outside these ranges, ask why. There may be a valid reason (your body weight, your tolerance history), but it should be stated, not assumed.

3. Does it account for your medications and health history?

This is the most important safety check. An AI recommendation that ignores your medication list is not personalized. It is dangerous.

Specific interactions to watch for:

• SSRIs or SNRIs + 5-HTP or St. John's wort. Serotonin syndrome risk. Any AI system that recommends these combinations without flagging the risk is not safe to use. ⁸
• Anticoagulants (warfarin) + high-dose omega-3, curcumin, or glucosamine. Bleeding risk changes. ^{7 9}
• Thyroid medication + calcium, iron, or magnesium taken at the same time. Absorption interference. ⁵
• Stimulant medications + caffeine-containing supplements. Additive cardiovascular load. ²

If the AI did not ask about your medications, or if it recommended a supplement in one of these interaction categories without mentioning the interaction, treat that as a system failure, not a minor oversight.

4. Does it fit your actual schedule?

A recommendation that requires four precisely timed doses across the day will fail if you have a hectic schedule. Adherence is not a willpower problem. It is a design problem. ¹⁰

Ask:

• Can I realistically take this at the recommended time every day?
• Does this conflict with other supplements I am already timing (e.g., thyroid medication that needs a fasting window)?
• Is this adding complexity that will reduce my overall compliance?

If the protocol is too complex, ask the AI to simplify. A simpler protocol you follow consistently beats an optimized protocol you follow half the time.

5. Is the expected timeline realistic?

Different supplements work on different timescales. The AI should tell you when to expect results and when to evaluate.

If the AI implies you will "feel the difference" from creatine in two days, or that ashwagandha will transform your stress response in a week, the timeline is wrong and the recommendation loses credibility.

How to run the trial

If the recommendation passes all five checks, test it with a structured trial:

Baseline (7-14 days). Before adding the supplement, log your target metrics (the outcome the supplement is supposed to improve) daily. This gives you a comparison point that is based on data, not memory.

Intervention (7-42 days, depending on the supplement type). Add the recommended supplement. Change nothing else. Log the same metrics daily using structured response labels.

Review. At the end of the trial period, compare your intervention averages to your baseline averages.

What to do when you disagree with a recommendation

Disagreement is not defiance. It is judgment.

If the AI recommends something and your instinct says "that doesn't seem right," investigate before dismissing. Check the rationale. Look up the referenced evidence. Ask whether the AI had accurate information about your situation.

But if, after investigating, you still disagree, do not take the supplement. You are the decision-maker. The AI is a tool. Tools that override human judgment in health contexts are poorly designed.

Evaluating recommendations in Unfair

Unfair's recommendation engine shows the rationale behind every suggestion, including the evidence it drew from and the personal data points it used. When you log a response (positive, negative, or neutral), that data feeds back into the model so future recommendations improve. Recommendations you rejected or that produced side effects are deprioritized automatically. The goal is a system that gets better at predicting what works for you specifically, not one that pushes the same generic advice regardless of your feedback.

Continue with The Role of AI in Supplement Recommendations, How AI Personalizes Supplement Recommendations, and Evidence-First Supplement Prioritization.

References