Blog

How AI Delivers Truly Personalized Supplement Recommendations

Unfair Team • January 23, 2026

Generic supplement advice fails because it treats everyone as a population average. The person who gets anxious from 100mg of caffeine and the person who drinks 400mg daily without noticing are not the same user, and they should not receive the same recommendations. AI-driven personalization works by building an evolving model of you from your goals, your logged responses, your schedule, and your history of what worked and what did not.

This is not magic. It is structured data collection feeding a decision engine that narrows options based on evidence and your individual patterns.

What "personalized" actually means

A recommendation is personalized when it accounts for at least three of the following inputs:

If a recommendation engine does not use at least three of these inputs, it is just a filtered search, not personalization.

How the feedback loop works

AI personalization is not a one-time output. It is a cycle that improves with each logged data point.

Cycle 1: Initial recommendation based on goals and constraints

You tell the system your primary goal (e.g., "improve afternoon focus without disrupting sleep") and any relevant constraints (medications, stimulant sensitivity, schedule). The AI generates an initial protocol drawn from evidence-based candidates that match your profile.

Example initial output:

> Goal: Afternoon focus, sleep-safe

>

> Recommended protocol:

>

> - L-theanine 200mg + caffeine 50mg, taken before 1 PM

> - No additional stimulants after that window

>

> Rationale: L-theanine and low-dose caffeine have evidence for improved attention-related outcomes in combination. ^{2 3} The low caffeine dose and early cutoff reduce sleep disruption risk based on a typical 4-5 hour half-life. ⁴

Cycle 2: Response data refines the model

You log structured responses for 7-14 days. The data might look like this:

The AI observes: focus labels are "Sharp" on 5 of 6 dosing days versus "Scattered" on the skip day. Sleep is unaffected. One mild headache on day 6 is noted but not a pattern. The protocol is working.

Cycle 3: Adjusted recommendation

Based on the data, the AI might suggest:

• Keep the current protocol. Focus improvement is consistent and sleep is not disrupted.
• Optional next experiment. If you want to test whether L-theanine alone (without caffeine) produces similar focus results, run a 7-day comparison. This would tell you whether caffeine is doing the work or if L-theanine is sufficient on its own.

Notice what the AI does not do: it does not suggest adding three more supplements because "more is better." Each recommendation is a single testable change.

Cycle 4 and beyond: The model gets sharper

Over months of logged data, the system accumulates enough pattern information to make increasingly specific suggestions:

• "You consistently report 'Disrupted' sleep when you take caffeine after 1:30 PM. Your effective cutoff appears to be 1 PM, not 2 PM."
• "Ashwagandha produced 'Stable' mood labels in 80% of your logs over 6 weeks. Consider keeping it in your foundation stack."
• "You tried rhodiola twice and logged 'No change' both times. It is deprioritized for your profile."

This kind of personalization is only possible with consistent logging over time.

Where AI personalization breaks down

Responsible AI systems are transparent about their limits:

• Garbage in, garbage out. If your logs are inconsistent, backdated, or use vague labels ("felt okay"), the AI has nothing meaningful to learn from. Personalization quality tracks directly with logging quality.
• Correlation is not causation. The AI can identify that your best focus days correlate with your morning stack. It cannot prove the stack caused the focus. Confounders (sleep quality, workload, stress) are always present in real life. ⁵
• New supplements have no personal data. When the AI recommends something you have never tried, it relies on population-level evidence and your profile constraints, not personal response data. The first trial of any new supplement is always the least personalized.
• AI cannot replace clinical judgment. For supplement-medication interactions, contraindications, and medical conditions, AI screening is a safety net, not a substitute for a conversation with your doctor or pharmacist. ¹

Practical guidelines for getting the most from AI personalization

• One goal at a time. "Improve sleep" and "boost focus" are separate experiments. Running both simultaneously muddies the data.
• One change at a time. If the AI suggests adding a supplement, do not also change your dose of something else that week. Attribution requires isolation.
• Log consistently, even when nothing changes. "Normal" days are data. They establish your baseline. Without them, the AI cannot tell whether a "Sharp" day is an improvement or just your average.
• Review the rationale. Before accepting any recommendation, read the AI's stated reasoning. If the rationale does not make sense to you, do not take the supplement.

Privacy and control

Your supplement history, response labels, and health notes are personal data. A trustworthy personalization system should:

• Store data on your device by default, with cloud sync as an opt-in choice you control.
• Make all recommendations editable and overridable. The AI suggests. You decide.
• Show why each recommendation was made. "Because you told us your goal is sleep and magnesium glycinate showed positive results in your logs" is transparent. A recommendation with no explanation is not trustworthy.

Personalized recommendations in Unfair

Unfair's recommendation engine uses the feedback loop described above. Each suggestion is linked to your stated goals, your logged response history, and your current stack. When you log a dose and a response label, that data feeds back into the next recommendation cycle. Over time, the recommendations narrow toward what has actually worked for you, not what works for an average person in a clinical trial. Every recommendation includes a rationale you can read before deciding.

Continue with The Role of AI in Supplement Recommendations, Evaluating AI Supplement Recommendations, and AI-Assisted Dose Logging.

References