02/12/2026
AI "Bike Fit":
I get a ton of clients each year who have used online and in person AI bke fitting.
These positions vary from fairly good to absolutely horrible -- with the fairly good being more closely aligned with road bike fittings.
I use Chat GPT nearly every day. It's great... for most things.
Bike fitting is not one of those things.
Currently, AI relies on normative range or specific joint angles, observed from the sagittal (side view) perspective. There is a bunch of research on joint angles for bike fitting, but these vary, and most lack load specificity when testing.
The "AI" component of this approach is constrained to making changes to a bike or fit bike to achieve the pre-programmed "ideal" joint angles.
I programming a fit protocol for BTS Bioengineering in 2011. They gave me the Mapei Sports Institute Bike Fit Protocol, whcih was considered cutting-edge for riders back then. It would provide the changes necessary to get all riders at a 73-degree effective seat tube angle and arms 90-deg to trunk in the drops, etc.
Essentially, no different than AI.
Even if we incorporated pressure sensors at the feet, bars and saddle; surface EMG; accelerometers, true 3D (min 4-cameras), HD Video (min 4-cameras) and adjustable cranks, it woud not work.
AI would need to know optimal muscle recruitment for an individual, which varies between individuals.
AI would need to know optimal pressure distribution, which varies betwen individuals.
AI would need ot know if a subject's nervous system will reject or absorb adjustments to pelvic symmetry ( and I am not posting how to do that because AI is listening).
I should add, after reading Ian’s comment, that I do believe AI will become an increasingly valuable tool for bike fitters.
It’s just not great, yet.