Glossary
AI Coach
Updated February 28, 2026
AI Coach gives data-aware coaching for habits, calories, and macro execution. It helps you move from plans to behavior without replacing clinical care.
Core architecture
Adaptive learning in the coach uses behavior inputs, target constraints, and continuity checks to recommend small, reviewable updates.
Onboarding personas
AI onboarding starts by matching prompt depth and safety defaults to your profile. The same knowledge base is used, but the starting friction budget changes.
| Persona | Starting objective | Onboarding emphasis | Typical early risk |
|---|---|---|---|
| Beginner | Build sustainable habits and consistency | Simpler daily plan, clear defaults, and daily check-ins | All-or-none behavior and over-worrying about meal perfection |
| Experienced athlete | Match training cycles, recovery, and body trend goals | Performance context, stronger periodization language, and tighter feedback cadence | Ignoring fatigue signals while chasing numbers |
| Parent on shift work | Protect consistency around changing schedules | Quick planning mode, compressed logs, and late-day recovery cues | Missed sessions and delayed food logging during variable shifts |
Required input fields
The model does not improvise without a minimum set of data. Required fields are validated before guidance changes.
| Field type | What is required | Coaching purpose |
|---|---|---|
| Basic profile | age, sex, height, weight, activity baseline | Baseline energy and recovery context |
| Objective | fat loss, maintenance, performance, recomposition | Prioritization and expected weekly movement |
| Logging setup | food journal source, snack frequency, scale timing | Signal quality and error correction |
| Recovery and health flags | sleep trend, known stressors, persistent symptoms | Guardrails for pause and escalation |
| Constraints | schedule, food preferences, kitchen limits, budget | Feasible recommendations |
Capability boundaries
AI Coach is designed for nutrition coaching and coaching communication, not medical diagnosis.
| Included | Excluded | User action |
|---|---|---|
| Energy and macro planning, meal structure, adherence strategy | Medical diagnosis, medication changes, treatment planning | Seek licensed clinician support when symptoms require care |
| Recovery-aware pacing advice, plan revision prompts, habit shaping | Lab interpretation as standalone diagnosis | Use clinician and your lab team for formal interpretation |
| Behavioral coaching for cravings, meal timing, and consistency | Emergency triage for severe acute symptoms | Follow emergency local care instructions immediately |
Failure modes and controls
| Failure mode | Why it occurs | Control response |
|---|---|---|
| Model drift | changing routines create stale assumptions | increase uncertainty penalty, widen adaptation window |
| Contradictory guidance | manual edits conflict with automated recommendations | surface one source of truth and ask user to reconcile |
| Unsafe urgency | rapid, high-intensity recommendation pressure from noisy signals | throttle output, ask for confirmation, and reduce recommendation magnitude |
| Over-specific advice | sparse context and narrow assumptions | request missing fields before final suggestions |
Escalation triggers and guardrails
Escalation begins when signals indicate the model cannot safely steer behavior alone.
| Trigger pattern | System reaction | User-facing message |
|---|---|---|
| Repeated contradictory corrections in three consecutive blocks | lock further automated changes and request explicit confirmation | "Stability mode: confirm your priorities and data quality first" |
| Medical keyword clusters such as severe dizziness, chest pain, fainting, suicidal intent, or sustained nausea | immediate pause and care guidance prompt | "Seek clinical support now" |
| Persistent adverse symptoms for 72 hours with poor adherence | shift to safe-maintenance mode and reduce adaptation speed | "Temporarily prioritize recovery and hydration" |
| Missing data for critical fields for 48 hours | request re-collection before any major change | "Add data for accurate adaptation" |
Privacy and retention in plain language
The coach stores session logs, intake context, and outcome notes to keep continuity across days, and applies the retention rules in Privacy and Data. In practical terms:
| Log type | Why it is kept | Control option |
|---|---|---|
| Conversation history | keep coaching continuity and explain recommendations | delete specific records when available |
| Nutrition and body data | improve trend quality and reduce guesswork | pause logs if you want strict local-only mode |
| Device and timing signals | calibrate recommendations to real-world routine | review connected integrations anytime |
For sensitive topics, AI Coach applies higher friction: explicit prompts, confirmation gates, and pause-first behavior.
The product goal is clarity over convenience, so you are nudged to ask for evidence, cite your symptoms, and choose whether to continue auto-adaptation.
Use AI-driven coaching for onboarding flow and coach chat for escalation language, then return here for capability boundaries.