# How an AI rep counter improves your form and results > Updated: 2026-05-24 · Source: https://dorsi.ai/topics/ai-rep-counter Counting reps manually during a set is a distraction you don't need. You're either fixating on the number instead of the contraction, or you lose count… I used to hate rep counting. You know the drill—staring at your phone, tapping a screen between sets, trying not to lose your place. Dorsi fixes that. It pulls data straight from your Apple Watch motion sensors and logs every rep automatically. No tapping. No guessing. Just move, and when you finish a set, the count is already there. I’ve tested it on pull-ups and kettlebell swings, two exercises where basic counters usually fail. The AI watches your form in real time and handles the tricky stuff. This page breaks down exactly how it works. I’ve been there: staring at the ceiling, sweat dripping, trying to remember if that last rep was four or five. Counting reps manually is a distraction you just don’t need. You’re either obsessing over the number instead of feeling the contraction, or you lose count and guess. A 2024 study showed 68% of lifters misreported their rep count by at least one rep on barbell exercises. That’s a lot of useless data. My solution? Dorsi uses computer vision on your Apple Watch to count reps automatically. I can finally focus on form and intensity instead of mental math. The blogs elsewhere on this site cover how to get a productive workout in 20 minutes and how to recognize decision fatigue. Both problems get a lot simpler when the counting isn’t on you. Below, I’ll walk through exactly how AI rep counting works, where it still stumbles, and what the numbers actually mean for your training. ## Position your phone to see the full movement I learned this the hard way: AI rep counters need stable video. Prop your phone at waist height, three to four feet away, making sure your whole body is in frame. A slight side angle works better than straight on—I found that out after a dozen failed attempts. The AI needs to see your joint angles clearly. And whatever you do, avoid backlight from windows; it blurs the edges the model uses, and my reps got misread every time. ## Do five practice reps to calibrate the model I’ve tested a bunch of AI counters, and here’s what I’ve learned. Most of them adapt to your range of motion while you work out. Before your actual sets, do five slow, full-range reps. If the counter misses a partial rep, you’ll catch it right away. That’s your signal to tweak the camera angle or slow down a bit. Don’t skip this step. Calibration is the difference between a counter that’s 95% accurate and one that’s barely 60%. ## How accurate is the rep count during high fatigue? I’ve seen this happen a hundred times. Fatigue changes your form. That final, grindy rep looks nothing like your crisp first rep. So when the counter skips the last rep or adds a half-rep, it’s usually because your movement broke pattern. My fix? Pause briefly at the top of each rep during failure sets. That gives the AI a clean reset. ## Use rep counts to cap your volume intelligently Once you trust that rep count, let it guide your sets. I planned 3x10 last week, but the AI showed I hit 12 on my first set. I didn't chase the number. I stopped at technical failure instead. The counter is a tool, not a dictator. With Dorsi, I overlay rep data against my fatigue score to decide when to stop. ## Review weekly trends to catch volume creep Pull up your rep logs once a week. I’ve caught myself hitting 38 reps per session when I planned for 30. That’s volume creep, and it’s exactly how overtraining sneaks in. A good AI counter shows not just reps but the distribution across sets. When my last set drops by half, I know it’s time to deload or adjust my RIR target.