Fair question, and it deserves a straight answer rather than a sales page: yes, AI can genuinely check your serve — some of it, with real numbers — and any tool claiming to check all of it is padding the truth. Here's exactly where the line sits, using CoachVision's serve analysis as the worked example, limits included on purpose.

What body tracking actually measures

Film a serve and the system tracks 17 body points frame by frame — 2-D keypoints from a single camera — then computes measurements from how those points move. For the serve, split by what each camera angle can honestly see:

From the side: how high your tossing arm reaches and whether it stays straight, the bend in your hitting arm at the trophy and where your elbow sits against your shoulder line, whether your wrist actually folds below your elbow behind you (the racquet drop), how high above your head you meet the ball, how much your knees load and your hips rise, your up-slow/down-fast rhythm, how much you lean into the court, and how balanced you land.

From behind: how steeply your shoulders stack at contact — hitting shoulder over toss shoulder — and where your hitting arm finishes across your body.

From either, with two or more serves: how alike your serves are, hit to hit.

These aren't estimates dressed as data — they're computed from tracked motion, in body-relative units (torso-lengths) so a tall server and a short one compare fairly.

What it honestly can't see — and won't pretend to

  • The ball and the racquet. There are no keypoints in your hand and none on the ball. Every toss read comes from your tossing arm, not the ball's flight — so it never claims a ball height, a clock position, or "your toss went behind you." The racquet drop is read from your wrist folding below your elbow; the racquet head falls deeper than that, unseen.
  • Pronation. The forearm snap that whips the racquet face through the ball has no keypoints, so it is never scored. A coach's-eye read of the contact frame can comment on it; a number can't, and we don't fake one.
  • Speed. No mph, no kph, ever. Hitting-wrist speed is used only to find the contact moment; it's never converted to a serve speed.
  • Serve type. Flat, slice, kick, second — the tracker can't tell them apart. You tell it, because a kick serve legitimately tosses more overhead and finishes differently, and grading it like a flat serve would be noise.
  • The court. No baseline, no lines, no foot faults, no in-or-out. Landing balance is body-relative only.
  • The exact strike instant. At 30fps the true contact lasts a frame or two; the system estimates it. Contact shapes (reach, lean, tilt) are trustworthy; contact timing to the millisecond is not, and it never pretends otherwise.

Why the limits are the feature

A measured read on the checkpoints that have keypoints, plus a coach's-eye read of the frames for the things that don't, turns out to be exactly what a serve check needs: numbers where numbers honestly exist, plain hedging where they don't, and a trend line — same angle, week over week — that no mirror or memory can give you.

So: can AI check your serve? The measurable half, better than your hitting partner's eyeball — and it will tell you plainly which half is which. Film one serve, side-on at chest height. See what the honest version looks like.