Most AI coaching sounds like it was written for nobody.
It notices that a ride had effort. It recommends consistency. It says recovery matters. It gives you a paragraph that could fit any athlete, any route, any Tuesday.
That is not coaching. That is slop.
A useful coach should sound like someone who watched the ride. Not the whole life story. Just the facts that change the answer: recent work, selected route, workout shape, session goal, FTP, fatigue, and whether the road was rolling or one long climb.
The voice matters too. Good feedback can be warm without going soft. It can be positive without pretending the data says nothing. The next ride should feel clearer after you read it.
There is also a privacy line. Cloud AI coaching is optional and requires explicit in-app consent. Ride summaries and training context are not the same thing as raw FIT files, full GPS streams, HealthKit data, or wearable recovery snapshots. Those categories have their own controls.
The goal is not to make the app sound clever. It is to help you understand what happened and what to do next.
If the coach did not learn anything from the ride, it should not speak.