Can I upload a golf swing video to Claude?

Claude works from images, not video. To analyze a swing with Claude directly you need to pull still frames out of your video first (setup, top of backswing, impact, finish) and upload those as images.

Is Claude accurate at analyzing a golf swing?

On clear, well-framed stills, Claude gives coherent feedback on posture, alignment, club position, and sequencing that is consistent with standard instruction. It is not a substitute for a coach who can watch you hit balls, and its accuracy drops sharply on blurry frames, bad angles, or frames that are not at the true key positions.

How is BirdieSwing different from just asking Claude?

BirdieSwing does the parts Claude cannot do alone: it finds impact in your video using audio and motion, extracts the eight key positions automatically, applies a fixed scoring rubric per swing phase and per club so results are comparable over time, and stores every analysis so you can track progress. The technique analysis itself uses Claude's vision model.

Can Claude analyze my golf swing?

Yes, Claude can analyze your golf swing, with two caveats. Claude reads images, not video, so you have to pull the key frames out of your clip yourself. And because it has no fixed scoring rubric, two analyses of the same swing can read quite differently. Below is how to do it by hand in about five minutes, what to expect, and how BirdieSwing automates the whole pipeline using Claude's vision model.

What Claude is good at

Give Claude a clear, well-lit still of a golfer at address, and it will comment sensibly on spine angle, knee flex, grip position, ball position relative to the stance, and alignment. Give it a frame at the top of the backswing and it will talk about clubface orientation, lead-arm structure, and how much the hips and shoulders have turned. This is real, useful feedback, phrased the way a decent instructor would phrase it.

Where it shines is explanation. If you say "my miss is a big slice with the driver, what in these frames could be causing it?", you'll get a ranked list of plausible causes tied to what it can actually see in the image, not a generic listicle.

What Claude can't do on its own

It can't watch video. You need still images. That means you're the one deciding which frames matter, and if you pick a frame a few hundredths of a second off impact, the feedback is about the wrong position.
It doesn't know where impact is. Finding the true impact frame is the hard part of swing analysis. Humans are bad at it, and scrubbing a phone video by hand is worse.
No consistent scoring. Ask twice and you may get a "7/10" and then an "8/10" for the same swing. Without a rubric, numbers are vibes.
No memory of your last session. Unless you build your own spreadsheet, there's no way to see whether your transition is actually improving.
No club context unless you supply it. A wedge swing and a driver swing are judged on different criteria. Claude will assume unless you tell it.

How to do it by hand (5 minutes)

Film your swing face-on or down-the-line, phone at hip height, roughly 10 to 12 feet away, in decent light. Lock focus if your camera app allows it.
In your phone's video player, scrub to these positions and take a screenshot of each: address, top of backswing, impact, and finish. If you're patient, add takeaway (shaft parallel to the ground) and halfway down.
Upload the images to Claude in that order and use a prompt like this:

I'm a right-handed golfer. These are stills from a down-the-line video of my 7-iron swing: address, top, impact, finish. My typical miss is a pull. For each frame, tell me what looks solid and what looks off relative to sound fundamentals. Then give me the single most likely cause of my miss and one drill to fix it. Be specific about what you can and can't see.

Ask follow-ups. "What would this look like if it were correct?" and "What feel should I try on the next swing?" both work well.

That last line of the prompt matters. Asking the model to be explicit about what it can't see is the best defence against confident feedback on a blurry frame.

How BirdieSwing automates this

BirdieSwing is that workflow with the manual steps removed, running on Claude's vision model:

Impact detection. Audio, motion, and pose signals locate impact in your clip, then the app selects the eight critical positions automatically. No scrubbing, no screenshots.
A fixed rubric. Each of five phases (setup, backswing, transition, impact, follow-through) is scored against the same criteria every time, adjusted for the club you selected. That makes a 72 today and a 78 next week mean something.
One drill, not ten. The result is an overall score, your strongest and weakest phase, a "feel and fix" cue, and a single recommended drill.
Tempo. The app reports your backswing-to-downswing tempo ratio; roughly 3:1 is the range most good players fall into.
History. Every analysis is saved, so progress is visible rather than remembered.
Privacy. Videos are processed for your analysis only and never used to train models.

Results come back in under 30 seconds. The free tier includes a starter set of analyses; Premium unlocks unlimited analyses and progress tracking. See how it works for the full walkthrough.

Honest limits, either way

AI swing analysis, ours included, works from what the camera sees. A poorly framed video produces poor feedback whether Claude is reading it directly or through BirdieSwing. It also can't watch your ball flight or feel your pressure shift the way a coach standing next to you can. Treat it as the best between-lesson feedback loop you can get for the price of propping up a phone, not as a replacement for a good instructor.