Samsung Patents an AI Health Coach That Promotes Its Best-Performing Model
Samsung's patent describes a health coaching system that runs two AI models in competition, watches which one you respond to better, and then gives that model progressively harder coaching tasks while the other stays at the basic level.
How Samsung's competing AI coaches learn what you like
Every time you ask your phone or smartwatch for health advice, you're probably getting one answer from one AI. Samsung's patent imagines something different: two coaching AIs answering the same question at the same time, like two trainers giving you their take on the same problem.
You see both responses, and based on how you react (whether you follow one, rate it, or engage with it more), the system decides which trainer you trust. From that point on, your preferred AI gets assigned the more complex, in-depth questions, while the other stays at a simpler level. The idea is that your feedback becomes the training signal, gradually giving you more personalized and detailed guidance from the coach that fits your style.
This plays out across multiple rounds of questions. A "manager" AI coordinates the whole process, sending out requests, reading your feedback, and adjusting which model earns the harder assignments next time. It's a tiered system that gets more tailored the more you use it.
… based on a feedback about the first health guide, select at least one of the first sub-coaching model and the second sub-coaching model as a preferred coaching model …
Translation: The system learns which AI model gives better advice by tracking how the user reacts to its health suggestions.
How the manager model picks a winner and reassigns difficulty
The patent describes a three-model architecture for health coaching on an electronic device (a phone, tablet, or wearable). There are two "sub-coaching models" (think of them as two AI assistants with different coaching styles) and one manager model that orchestrates them.
Here's how a session works:
- The manager model collects your health status data (activity, sleep, vitals, or whatever the device tracks) and sends the same prompt to both sub-coaching models simultaneously.
- Both models return a response at what the patent calls "first level" detail, meaning a baseline depth of advice. You see both answers bundled into one health guide on your screen.
- Your feedback on that guide (implicit or explicit) tells the manager which model you preferred. That model becomes the "preferred coaching model."
- On the next round, the preferred model is asked a "second level" question, meaning a more complex or detailed coaching task, while the non-preferred model stays at the first level.
The "levels" here work like difficulty tiers. The model that earns your trust gets progressively more demanding prompts, pushing it to deliver richer guidance tailored to you. The system is designed to adapt over time rather than deliver a one-size-fits-all answer from a single static model.
… acquiresthird coaching information of a second level, higher than the first level and acquires, from the other sub-coaching model, fourth coaching information of the first level.
Translation: The device prioritizes the better AI model to provide more advanced health tips while keeping the other model as a backup.
What Samsung's coaching patent means for wearable health advice
For anyone using a Samsung Galaxy Watch or phone for health tracking, this architecture could mean advice that actually adapts to how you respond to coaching, not just to your raw health numbers. Most current health apps push the same generic tip formats to everyone. A system that watches which style of advice you actually follow and promotes that model could, in theory, feel meaningfully more personal over time.
The filing is specific enough to a Samsung device context that it reads like infrastructure for a future Galaxy AI health feature rather than a broad platform play. Samsung's Galaxy ecosystem, which already ties together phones, watches, and earbuds under a health umbrella, is a natural home for a coaching system that learns user preferences across sessions. Samsung's health-AI filings are part of a broader wave of new Big Tech patents pushing AI coaching beyond simple step counts and into adaptive, preference-driven guidance.
Samsung's fourth filing since July in our AI models working in teams watchlist adds to earlier work on switching assistants mid-talk and AI grading its own answers, showing how the company keeps refining how AI models collaborate.
Claim 1 locks in a specific tournament structure: a manager model that solicits outputs from two sub-coaching models at a defined difficulty level, compares them via user feedback, designates a preferred model, and then routes a follow-up query accordingly. That sequence is narrow enough that a product using three sub-models, or one that skips the explicit feedback-driven selection step, would likely fall outside the claim's reach. What it does protect is meaningful if Samsung ships this architecture, because any health coaching product that follows this exact manager-plus-two-competitors-plus-feedback loop would be squarely within its scope.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
18 drawing sheets from US 2026/0245698 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →