AI in Fitness: What Personalised Coaching Can—and Cannot—Do

From generic plans to contextual feedback

Traditional digital fitness products often provide the same plan to every person in a broad category. AI systems can use session history, stated goals, and camera-derived movement landmarks to make guidance more contextual. This can help a user notice patterns and maintain a consistent routine.

Personalisation is only as useful as its inputs. Lighting, camera placement, model confidence, incomplete tracking, and inaccurate profile information can all affect the quality of a recommendation. Good products communicate uncertainty rather than treating every model output as fact.

How camera-based movement feedback works

Pose-estimation models infer keypoints such as shoulders, hips, knees, and ankles from video frames. A coaching layer can compare those landmarks over time, estimate movement phases, and offer simple cues. The output is an approximation of visible motion—not a clinical assessment or a guarantee that an exercise is safe for every individual.

• Place the camera far enough away to keep the full body in frame.
• Use steady lighting and avoid backgrounds that obscure limbs.
• Treat automated cues as prompts to review, not medical diagnoses.
• Stop if you experience pain, dizziness, or unusual symptoms.

Personalisation should preserve human agency

Useful AI supports a person’s own goals instead of pressuring them into unrealistic targets. Training suggestions should account for rest, accessibility, experience, and the fact that progress is rarely linear. A qualified trainer or clinician remains important when a person needs individual professional guidance.

A more responsible fitness technology stack

Responsible fitness AI makes its data practices understandable, asks permission before camera access, minimises unnecessary data collection, and offers a clear way to stop a session. It also avoids promising exact body measurements or guaranteed results from ordinary RGB cameras.