Can ChatGPT analyze running form?
Not really, because ChatGPT can describe a still frame you show it but does not track your joints from frame to frame, so it cannot find foot strike, measure landing reach or lean, or count cadence from a video.
What ChatGPT can do with a running photo
Give a chat model a photo of you mid-stride and it will say something. It can tell you the picture shows a runner, that the heel seems to be landing first, that the torso looks upright. It can explain what overstriding is and list drills. That is all real and some of it is useful.
The problem is the word "seems". The model is describing what it sees in one image, in words. It is not measuring. Ask it for your landing reach as a fraction of leg length and it will guess or refuse. Ask it twice and you may get two answers.
Why a video needs something different
Running form lives in the timing. The frame that matters is the one where your foot first touches the ground. A single step at easy pace lasts around a third of a second. To find foot strike you need every frame, and you need to know where the ankle was in the frames before and after it.
That is a job for a pose estimator. It marks your joints in each frame, then watches the ankle's height and speed to pick out foot strike, midstance and toe-off. Only then can it measure angles at the same moment every time. See how running form analysis works for the full chain.
A chat model does not do that. It does not run joint tracking across frames. It does not pick the foot strike frame for you. It cannot count foot strikes over 15 seconds to give you cadence. So the three numbers most worth having, landing reach, lean and cadence, are out of reach.
Where a chat model is useful
Be fair to it. After you have real numbers, a chat model is a decent place to ask questions. What does a 12 degree forward lean mean. Which drills raise cadence. Why does a straight front knee at landing matter. It can explain those in plain words.
It is also fine for a rough sanity check on a single paused frame. If you pause your own video at foot strike and the foot is obviously a long way in front of your hip, you do not need software to tell you. You need a plan. Read what overstriding is for what to look for.
Just do not treat the description as a measurement. And do not expect the same answer tomorrow.
What a purpose-built app does instead
An app like Caden runs a pose model on the uploaded clip, finds the three moments in every clip, and measures the same six things each time. Where your foot lands, knee bend on landing, knee bend at push-off, upper body lean, arm bend and head position, plus cadence. It gives a form score out of 100 and one cue per metric. Then you can start a practice plan built from the weakest scores.
Because every clip is measured at the same three moments, you can film again in two weeks and compare. That is the part a conversation cannot give you. For the wider list of options see what app can I use to check my running form.
How to check it with Caden
- Tap Run a new analysis and film 10 to 20 seconds from the side at easy pace, phone at hip height.
- Open Breakpoints and tap Foot strike. This is the frame a chat model cannot find for you, with the skeleton drawn on it.
- Read the foot landing dial. It gives landing reach as a number, not a description.
- Read the coach's tip for the plain-words version. If you still have questions, that is the moment to ask a chat model, with your real numbers in hand.
FAQ
Can ChatGPT watch a video of me running?
It can look at individual frames you give it and describe them. It does not track joints across the frames of a video or measure timing between them.
Can I use ChatGPT to make a running form plan?
It can suggest general drills. It will not know which of your metrics is weakest unless you measured them first. Measure, then ask.
Is Caden a chatbot?
No. Caden measures your video with pose estimation and reports numbers, freeze frames and a plan. The coach's tip is plain-language narration of those results, not a conversation.