I Let AI Tell Me What My Body Needs.
Testing whether an AI-powered wearable can understand my biometric data well enough to give advice I’d actually follow.
The question
Can an AI wearable understand me well enough to influence what I do?
WHOOP collects a continuous stream of biometric data and turns it into measurements of things like sleep, recovery, strain, and stress. But the more interesting part isn’t the tracking, it’s what WHOOP does with that information.
WHOOP Coach uses large language models alongside a member’s biometric history to provide personalized recommendations about sleep, recovery, training, and health. It can tell me how hard I should exercise, suggest when I should recover, explain patterns in my data, and respond to questions about how I’m feeling.
For this experiment, I’m less interested in whether WHOOP can count my steps or measure my heart rate. I want to know what happens when an AI system spends weeks observing my body and starts telling me what I should do because of it.
Why this matters
Most of my interactions with AI happen when I intentionally ask for help. I open a chatbot, I type a prompt, and I decide what context it gets.
A wearable changes that relationship.
WHOOP sits on my wrist collecting information throughout the day and while I sleep. Its AI can combine that biometric history with information about my behaviors and goals to generate personalized recommendations. In other words, I don’t necessarily have to explain how I’m doing before asking for advice. The system already has its own interpretation.
That creates a question I’m increasingly interested in as AI moves beyond our computers and into physical devices: What happens when AI doesn’t just respond to what we tell it, but continuously observes us and forms its own conclusions?
So I’m going to wear WHOOP and keep my own record of how I think I’m doing before looking at what WHOOP says.
How well did I sleep? How much energy do I have? How stressed do I feel? Do I think I need rest or could I push myself harder today?
Then I’ll compare my assessment against WHOOP’s.
I’m not trying to determine whether WHOOP knows my body better than I do. I’m more interested in the moments when we disagree and what happens next.
Because if a device tells me I’m poorly recovered on a morning when I feel great, which one of us am I going to believe?
And after weeks of receiving recommendations from it, I want to know whether that answer changes.
Methodology
Before beginning the experiment, I’ll wear WHOOP long enough for the system to establish the baseline data required to generate personalized assessments and recommendations. WHOOP’s Healthspan features require at least 21 Recoveries within 31 days before they unlock.
Experiment
Once the experiment begins, I’ll complete a short self-assessment each morning before opening WHOOP. I’ll record:
- How well I think I slept
- My perceived energy level
- My perceived stress level
- How recovered I feel
- Whether I think I should rest, maintain my normal activity, or push myself harder
Only after recording my own assessment will I look at WHOOP.
I’ll then document WHOOP’s assessment of my sleep, recovery, strain, and stress, along with any recommendation provided by WHOOP Coach.
For each observation, I’ll compare three things:
- Agreement: Did WHOOP’s assessment align with how I thought I was doing?
- Recommendation: What did WHOOP suggest I do based on its interpretation of my data?
- Influence: Did seeing that recommendation change what I actually did?
Disagreements will be especially important. If I feel well-rested but WHOOP reports poor recovery, or I feel exhausted while WHOOP suggests I’m ready for strain, I’ll document which assessment I trusted and why.
I’ll also track whether that behavior changes over the course of the experiment. The goal isn’t to determine whether WHOOP knows my body better than I do. It’s to observe whether repeated exposure to an AI system’s interpretation of my body causes me to increasingly trust, question, or act on its recommendations.
Measurement
Agreement rate How often my assessment and WHOOP’s assessment point in the same general direction.
Disagreement rate How often WHOOP and I reach meaningfully different conclusions about my condition or readiness.
Recommendation adherence How often I follow, partially follow, or ignore WHOOP’s recommendation.
Behavior change How often I would have made a different decision if I had relied only on my own assessment.
Influence over time Whether my likelihood of following WHOOP’s advice increases, decreases, or stays roughly the same as the system accumulates more data about me.
Timeline
Experiment opened. Hypothesis: I expect WHOOP’s AI to make generally reasonable inferences about how my body is doing, but I expect disagreements when those inferences translate into recommendations about when I should take it easy or reduce my activity.
First night of sleep with Whoop.
How to follow this experiment
This experiment publishes findings as they emerge, not on a schedule. Come back and check out the timeline above for updates!