AI Coach Accuracy by Metric: Heart Rate, Power, Pace, and RPE

Your AI coach reads the same metrics your Garmin, Apple Health, and power meter send. But it doesn't trust them all equally. Power data gets weighted differently than pace, heart rate gets checked against RPE, and GPS gets cross-validated against your actual training response. Understanding which signals your coach relies on most tells you where to invest in better equipment and which metric gaps matter most.

Heart rate is reliable for zone placement, not absolute effort

Heart rate is the workhorse metric for AI coaching because it's stable, real-time, and broadly available. Your Garmin or Apple Health-connected device collects it every second, so the coach sees the full picture of your cardiovascular response. The accuracy is reasonable. Most optical wearables read heart rate within roughly 5-10 bpm at rest and in steady-state, with error widening during high-intensity work when motion artifact increases noise. That margin is small enough to matter for zone placement, though it is based on controlled testing conditions and real-world results vary.

What makes heart rate valuable for an AI coach is consistency. Your resting heart rate follows patterns. Your heart rate in zone 2 (the aerobic base zone where most low-intensity endurance work happens) sits in a predictable range for your fitness level. Your max heart rate changes little with training. Because of that stability, the coach can use your heart rate to infer your current aerobic capacity and stress state. When you show up for a workout with an elevated resting heart rate or a lower-than-usual max, the coach knows something is off. It could be fatigue, poor sleep, or illness.

But heart rate has a ceiling for absolute effort measurement. Two athletes can have identical heart rates at the same pace and be working at different physiological intensities. One might be untrained, hitting 160 bpm at a conversational jog. The other might be a trained endurance athlete, sitting at 140 bpm at the same speed with their aerobic system barely working. The raw number is almost useless without context. This is why the coach doesn't use heart rate alone.

Power is the gold standard because it's force, not response

Power (measured in watts) is the single most trustworthy metric an AI coach can read because it measures actual work done, not how your body reacted to it. A power meter on your bike or a running power estimate tells the coach the exact energy you expended. 150 watts is 150 watts. No variables. No fitness level context needed.

Cycling power meters are the most reliable. Quality cycling power meters from manufacturers like SRM and Quarq publish precision specs in the 1-2% range per their own testing documentation, meaning a reading of 300 watts is accurate to within 3-6 watts. The sources of error are small: temperature drift, cadence-related noise, and transducer calibration. A coach reading power from your power meter can trust it as the actual demand your legs are producing.

This is why power-based training is the backbone of cycling coaching. The coach can say: spend 30 minutes at 85% functional threshold power, and it knows exactly what intensity you're doing. It doesn't guess based on your heart rate or how you feel. It measures the work.

Running power estimates are less accurate. Most running watches estimate power from acceleration, cadence, and estimated ground impact. The margin of error is wider, typically in the range of 5-15% depending on the algorithm and terrain, though exact figures vary by device and conditions. But even with that slop, power gives the coach a better read on effort than pace alone does. A runner doing 8 min/km on flat asphalt and a runner doing 8 min/km on a steep trail are doing different amounts of work. Power captures that difference.

GPS pace is noisy but tells the story of your trend

GPS pace is everywhere, easy to collect, and notoriously inconsistent. Urban canyons, tree cover, bridges, and tunnels all scatter the signal. Horizontal GPS error in running watches varies with conditions, generally ranging from a few meters in open terrain to notably higher in obstructed environments, which translates to meaningful pace error on shorter efforts. In open air, accuracy is better. On a trail surrounded by forest, it degrades.

Absolute pace numbers are unreliable for adjusting workouts in real time. A coach can't say "run at 4:30/km" based on GPS and expect consistency between runs. But pace over long periods is reliable for detecting fitness progression. If your easy runs have drifted from 6:00/km to 5:45/km over 12 weeks while your heart rate in that zone dropped 8 beats, the coach sees the trend and knows your aerobic fitness improved. That trend is what matters.

GPS is also your coach's only window into terrain and elevation. Flat road, rolling hills, and vertical climbs all change the effort required at a given pace. The coach uses elevation data from GPS to understand whether you went easy or hard on a certain route. A 10 km run at 5:30/km on a flat course looks fast. The same pace over a hilly route looks harder. The elevation data helps the coach separate pace from actual work.

RPE (perceived effort) is a signal, not the signal

Rate of Perceived Exertion, usually reported on a 6-20 scale (the original Borg Scale) or the modified 1-10 scale depending on the coach or lab protocol, is how hard you feel a workout is. It's subjective. It varies wildly depending on your mood, sleep, caffeine, and how much pain tolerance you're in that day. An RPE 7 run for you might be a completely different intensity than an RPE 7 run for someone else.

But RPE is your coach's reality check. When you report a workout as an RPE 9 (all-out effort) and your power meter shows you were at 65% functional threshold power, the coach knows something is off. Either your power meter is reading wrong, you have a fitness question in need of investigation, or you're bad at self-assessment. More likely, the coach uses that discrepancy to learn your personal RPE scaling. It notes that you always perceive high-intensity work as harder than the numbers suggest.

RPE matters most as a cross-check against the data. If you finish a power-based interval session and report it felt easier than last week, but the power numbers look identical, the coach might infer that your fatigue has cleared or your fitness has improved. If you report it felt much harder while the metrics stayed the same, the coach might dial back the next session to prevent overreach.

Pure RPE coaching without metrics is unreliable. Pure metrics without RPE ignores the human side of training. The best coaches use both. Your AI coach does the same.

How AI coaches weight conflicting signals

When your metrics disagree, your coach has to pick a side. Heart rate is elevated but pace looks easy. Power is high but you reported low effort. Pace is off the chart but heart rate barely moved. Which signal wins?

The hierarchy usually favors power, then heart rate, then pace, with RPE as a modifier. Here's why:

Power measures work. If power is solid, the coach trusts it most because it's measuring the actual demand, not the response or the estimate. A high power output means you did the work, period.

Heart rate is next because it's responsive and real-time. It tells the coach how hard your cardiovascular system is working. It's less precise than power, but much more consistent than GPS pace.

GPS pace is useful for trend-spotting and terrain context, but least reliable for intensity. A drift in pace can mean a million things: hills, wind, GPS noise, fatigue, or fitness change. The coach weighs it last.

RPE sits outside the hierarchy. It's a modifier that either confirms the other signals or flags a discrepancy for investigation. If RPE aligns with power and heart rate, the coach has high confidence. If it disagrees, the coach slows decisions about workout adjustments until the picture clears.

Gaps that matter most

The biggest blind spot for an AI coach is the lack of a power meter for runners. A power meter eliminates the ambiguity between pace and effort. Running power is estimated on most watches, not measured, so the coach has to infer your actual work from slower, noisier signals. If you ride cyclocross or run trails, where terrain changes constantly, your coach works with one hand tied behind its back.

Strength training is another gap. Most AI coaches see your strength sessions as "logged manually" or not at all because wearables don't measure barbell lifts well. Your Garmin sees elevated heart rate and motion, but it can't tell if you did 5 sets of 3 reps at 80% one-rep max (heavy) or 3 sets of 10 reps at 50% (light). The load, the mechanical tension, the bar speed. All invisible to the watch. This is why AI coaches tend to adapt your cardio tightly and your lifting loosely. Strength requires more human input.

Apple Health integration and Garmin Connect close some gaps by pooling data across devices. If you log your strength session manually and sync your wearable metrics, the coach can align the training stress and adjust the next cardio session. But it's still reactive, not prescriptive.

What you can do to help your AI coach

1. Pair a power meter with cardio if you cycle. Power transforms the coach from guessing to knowing. It cuts iteration cycles in half.
2. Keep your wearable up to date. Your watch's HR algorithm improves with firmware updates. Older watches drift.
3. Log RPE at the end of sessions. One number per workout, takes 5 seconds. It helps the coach learn your personal scaling and flags anomalies.
4. Log your strength work. Manual entry is tedious but powerful. The coach sees that you deadlifted heavy yesterday and pulls back the run load today.
5. Use the same wearable for consistency. Switching between watches or relying on different platforms fragments the data your coach sees.

How Movement Rebels fits

Movement Rebels reads your Garmin power data (if you have a power meter) and your Apple Health heart rate, validates them against each other, and weights them precisely like a coach would. When your power drifts but heart rate stays steady, the app flags that your fitness might have improved. When you report high RPE on a low-power session, it logs that discrepancy for next time. The AI coach uses all four signals: power, heart rate, pace, and RPE, and the weekly plan adapts to what you did, how hard it was, and how your body is responding. If your power and heart rate diverge for two sessions in a row, the coach schedules an easier day before you feel the fatigue.