Form Feedback and Video Analysis Notes
Watching a client's lift video, taking notes, and writing a coherent reply is one of the most valuable things you do — and one of the slowest. AI can't watch the video for you (yet, reliably), but it can absolutely turn your raw notes into a polished, kind, and actionable reply in seconds.
What You'll Learn
- A repeatable workflow for video form review with AI assistance
- Prompts that turn rough form notes into clear client feedback
- Capturing and reusing your own coaching cue library
- Where AI vision tools currently help — and where they don't
The AI-Assisted Form Review Workflow
You still watch the video. You still spot the issues. Your eyes are the part of this that AI cannot replace. AI's job is the writing layer.
The workflow:
- Watch the video at normal speed once, then in slow-motion / frame-by-frame
- Type rough notes as you watch (3-7 bullet points is plenty)
- Hand the notes to AI with a prompt that includes the client's context
- Edit the draft, send
A 3-rep squat video that used to take 12 minutes (watch + note + write) now takes 4-5 minutes.
Step 1: How to Take Useful Form Notes
Notes don't need to be polished prose. They need to capture what you saw at what moment.
Example raw notes (squat, 3 reps, side angle):
rep 1: knees track ok, slight valgus on right knee at depth, butt wink last 5cm, hip drive ok, neutral spine top to mid-descent then loses a bit rep 2: better knee track, less butt wink, stayed braced longer rep 3: similar to rep 2, depth a touch above parallel general: bar position high-bar, looks ~RPE 7-8, breathing pattern fine priority fix: depth control + brace through full ROM
These rough notes are what you feed to AI.
Step 2: The Form Feedback Prompt
Below are my raw notes from watching a client's [exercise] video. Turn them into a reply to the client.
Structure:
- 1-line acknowledgment (something specific they did well)
- 1 priority fix only (most important issue)
- 1-2 specific cues they can try next session
- 1-line note on what I want to see in the next video
Tone: encouraging, technical but plain language, no jargon they wouldn't know. No medical advice.
Client snapshot: [paste] Notes: [paste raw notes]
Step 3: Real Example
Using the squat notes above, AI returns:
Solid set overall — your brace by rep 2-3 was tighter and the knee track on the right side cleaned up.
The priority for next session is depth control. You hit good depth in reps 1-2 but lost a few cm by rep 3, and there's a small tuck (the "butt wink") at the bottom of rep 1 that I want to keep an eye on.
Two cues to try:
- "Sit between your hips, not into them" — keep the brace going as the hips drop, don't relax at the bottom
- Pause for 1 count at depth on your first warm-up set — gives you a chance to feel the position before going heavy
For the next video: same angle, 3 reps at the same load, with a 1-count pause at depth on each rep.
You skim, lightly edit for voice, send. 4 minutes for a thoughtful, specific form review.
Why This Beats Writing From Scratch
Most coaches write form feedback that's either too short ("looks ok, fix your knees") or too long (a 600-word essay nobody reads). AI nudges you toward the middle: structured, focused, one priority at a time.
It also keeps your tone consistent across clients and across days when you're tired.
Building Your Cue Library
Cues are the most reusable asset you have as a coach. Build a personal cue library and AI can pull from it instead of inventing generic ones.
Format
Save it as a simple table per movement pattern. Example:
Squat cues
- Depth lost at the bottom: "Sit between your hips, not into them"
- Knee valgus: "Push the floor apart with your feet"
- Butt wink: "Pause one count at depth — feel the brace before standing"
- Forward fall: "Stack the bar over mid-foot — chest tall, not chest up"
- Heels lifting: "Drive heels through the floor like you're standing up out of mud"
How to Use It With AI
Paste the cue library at the top of your form-feedback prompt:
Here is my personal cue library. When writing client feedback, pull from these cues whenever the issue matches. Add new cues only if none of mine fit.
[paste cue library]
Now generate a feedback reply for: [notes]
This is the difference between a draft that sounds like AI and one that sounds like you.
Handling Common Lifts
Squat Form Review
Notes from squat video. Form review reply, max 200 words. Priority fix only — list secondary issues at the end as "watch list, not priority yet" so the client knows but isn't overwhelmed.
Deadlift Form Review
Notes from deadlift video. Form review reply, max 200 words. Always check for: bar path, hip vs back rise rate, lockout posture. Flag if my notes suggest the client is rounding the lumbar spine — that's an immediate stop-and-deload conversation, not a cue.
Bench Form Review
Notes from bench video. Form review reply, max 200 words. Address bar path, leg drive, and elbow flare order of priority. If notes mention shoulder pain, recommend the client deload and consult their PT — do not coach through pain.
Olympic Lift Review
Notes from [snatch/clean] video. Form review reply, max 250 words. Olympic lifts have many points of failure — pick one to address, not three. Always reference position-by-position (setup, first pull, transition, second pull, catch) so the client knows where in the lift the issue lives.
AI Vision Tools — What's Real Today
There are AI tools that claim to analyze lift videos directly. As of 2026, the landscape:
- Coach's Eye, Hudl Technique — drawing/annotation tools, not analysis
- OpenPose, MediaPipe — pose estimation libraries, useful for engineers, not turnkey for coaches
- Apple Visual Intelligence, Google Lens — can describe what's in a frame, not lifting biomechanics
- Specialized lift-tracking apps — some claim AI bar-path analysis; most are still keypoint-based with mixed accuracy
Honest take: rely on your own eyes. Use AI for the writing. The tooling will improve, but right now if you let an app tell you a lift is fine, you're outsourcing a judgment call you shouldn't.
When AI Form Feedback Is Wrong
- AI invents details from your notes ("good knee track" when you didn't say that)
- AI piles on too many cues — counter with "only one priority cue"
- AI reaches for medical-sounding advice — strip it out
- AI uses cue language that's not yours — your library fixes this
- AI normalizes pain ("a bit of soreness is fine") — sub the language to "stop and contact me if it hurts again"
Quick Wins to Implement Today
- Build your cue library, even just for the big 4 (squat, bench, deadlift, OHP)
- Save 3 form-feedback prompt templates in a notes app — squat, deadlift, bench
- Pre-write a "what to film next time" addendum so clients send better videos
- Use AI to write a one-time client guide on "how to film your lift videos for review"
Key Takeaways
- AI cannot watch your client's lift — your eyes still do the work
- Rough notes + a structured prompt = polished feedback in minutes
- Build a personal cue library and feed it to AI for consistent voice
- Always limit to one priority fix per video; list others as "watch list"
- AI vision tools aren't reliable yet for serious form analysis