# 28 Ways I Use AI for Health, Fitness, Weight Loss, and Training - Canonical URL: https://mattfarmer.ai/ai-health-fitness-prompts - Author: Matt Farmer - Published: 2026-07-20 - Last verified: 2026-07-20 - Primary phrase: AI prompts for health and fitness - Prompt count: 28 This is the complete machine-readable companion to Matt Farmer's practical health and fitness prompt library. It explains what to provide, what AI can do well, the main failure mode, and how to verify each result before giving the full prompt. Matt is working toward 170 lb. If achieved, that would represent an 80 lb loss in roughly eight months. This page does not claim the goal has been reached and does not claim AI caused a transformation. ## The boundary AI is useful for organizing records, showing arithmetic, comparing trends, formatting plans, surfacing questions, and designing low-risk experiments. A general AI assistant should not diagnose, prescribe, clear medication or supplement combinations, estimate exact body fat from an ordinary photo, turn correlation into causation, or replace urgent or professional care. Risk levels: - Level 1 — Draft and organize: Planning, formatting, sequencing, and other low-stakes work. - Level 2 — Analyze, then verify: Estimates and trends where the inputs, math, and conclusion need checking. - Level 3 — Prepare a professional question: Labs, interactions, symptoms, or decisions that may change care. ## Quick start For most people, start with four bounded workflows: 1. Macro-Based Meal Planning 2. Workout Split Around Exact Equipment 3. Accountability and Weekly Check-Ins 4. Doctor-Appointment Preparation ## Universal accuracy block Append this when the consequences matter: ```text Use only the information I provide and clearly label any assumption. Preserve all units and reference ranges. Show arithmetic where relevant. Give ranges instead of false precision. Separate observations from interpretations and recommendations. Cite the sources used for factual claims and tell me when the evidence is weak, mixed, old, or not directly applicable. State what information is missing. Give the strongest counterargument to your recommendation. End with what I can verify myself and what requires a qualified professional. Do not diagnose, prescribe, or tell me a medication or supplement combination is safe. ``` ## Nine accuracy habits 1. Provide exact inputs instead of asking the model to guess. 2. Keep units, dates, timezones, and lab-specific reference ranges attached to the data. 3. Ask the model to show arithmetic and reconcile every total. 4. Request a range and confidence level instead of false precision. 5. Make it list missing data and label every assumption. 6. Ask which sources support factual claims and how strong the evidence is. 7. Request contradictory examples and the strongest counterargument. 8. Turn interesting patterns into a prospective two- or four-week test. 9. End with what you can verify yourself and what belongs with a qualified professional. ## Privacy before upload A consumer health app is not automatically covered by HIPAA. Temporary or no-training modes do not remove every retention, access, deletion, or security question. Minimize what you upload and review current product controls. 1. Remove names, birth dates, record numbers, addresses, and other direct identifiers. 2. Crop screenshots to the fields the task actually needs. 3. Strip photo metadata and avoid recognizable backgrounds when identity adds no value. 4. Generalize locations, employers, providers, and exact dates when precision is unnecessary. 5. Check the product's current retention, training, deletion, and sharing controls before uploading. # Complete prompt library ## 01. Supplement Timing and Scheduling - Category: Supplements - Risk: Level 2 — Analyze, then verify - Scope: Level 2, with Level 3 escalation for medical conditions or prescription medications. - You provide: Exact ingredient, chemical form, dose, product label, reason for use, meals, sleep schedule, caffeine, medications, and relevant conditions. - AI does well: Turns a long list into a practical schedule and distinguishes supported instructions from flexible timing. - Main failure mode: Inventing timing rules or treating weak mechanistic evidence as proven benefit. - Verify by: Check the product label and cited human evidence; take medication or condition questions to a pharmacist or clinician. ### Copy-ready prompt ```text Build a practical daily schedule for this exact supplement list: [LIST EACH PRODUCT, FORM, AND DOSE]. My meal times are [TIMES], training time is [TIME], and bedtime is [TIME]. For each item, check whether the label or strong human evidence supports taking it with food, with fat, fasted, in the morning, at night, or separated from another item. Show the source and evidence strength. If timing is flexible or evidence is weak, say that instead of optimizing imaginary precision. Flag anything that needs a pharmacist or clinician review. ``` --- ## 02. Supplement and Medication Interaction Questions - Category: Supplements - Risk: Level 3 — Prepare a professional question - Scope: Level 3. - You provide: Generic and brand names, exact dose, route, schedule, reason, start date, age, major conditions, and planned supplement. - AI does well: Organizes exact pairs, mechanisms, source links, missing context, and questions for a pharmacist. - Main failure mode: False reassurance, false positives, missed dose or route context, or fabricated mechanisms. - Verify by: Use the result only as a pharmacist-review checklist; never treat it as clearance that a combination is safe. ### Copy-ready prompt ```text Create a pharmacist-review checklist from these medications and supplements: [EXACT LIST WITH DOSE, FORM, ROUTE, AND TIMING]. Surface possible interaction flags using authoritative sources such as official labels, NIH fact sheets, and established interaction references. For every flag, give the exact pair, mechanism, evidence/source, severity stated by the source, and the question I should ask my pharmacist. Do not clear any combination as safe. If the answer depends on dose, route, timing, diagnosis, or lab results, say exactly what is missing. ``` --- ## 03. Macro-Based Meal Planning - Category: Nutrition - Risk: Level 1 — Draft and organize - Scope: Level 1 for healthy adults; higher with clinical dietary needs. - You provide: Calorie and macro targets, preferred foods, allergies, budget, cooking time, meals per day, culture, and relevant medical constraints. - AI does well: Drafts repeatable meals around real targets, preferences, time, and budget while showing the totals. - Main failure mode: Totals that do not add up, unrealistic portions, low fiber or micronutrient variety, or an unsustainable menu. - Verify by: Recalculate a sample day from food labels or a named nutrition database and review clinical needs with a qualified professional. ### Copy-ready prompt ```text Draft a 7-day meal plan for [CUT/BULK/RECOMP] using [CALORIES] kcal, [PROTEIN] g protein, [CARBS] g carbs, and [FAT] g fat per day. I like [FOODS], avoid [FOODS], have [ALLERGIES/CONSTRAINTS], spend about [BUDGET], and can cook [TIME]. Prioritize protein, fiber, high-volume foods, and meals I will realistically repeat. Give gram weights, per-meal macros, daily totals, and a grocery list. Show the math and keep daily totals within 5% of target. Flag any likely micronutrient or fiber gap. Do not invent nutrition values; name the database or label assumption used. ``` --- ## 04. Workout Split Around Exact Equipment - Category: Training - Risk: Level 1 — Draft and organize - Scope: Level 1 for healthy training; Level 2 if pain or limitations are present. - You provide: Full equipment inventory, load ranges, training age, schedule, session length, goals, current numbers, joint limitations, and preferred movements. - AI does well: Builds a usable program from the equipment and time you actually have instead of an imaginary gym. - Main failure mode: Generic programming, impossible loading, excessive volume, or weak progression rules. - Verify by: Check every exercise against the equipment list, total the weekly sets, and stop if a movement produces concerning pain. ### Copy-ready prompt ```text Program a [DAYS]-day [STRENGTH/HYPERTROPHY/BOTH] split using only this equipment: [FULL INVENTORY WITH WEIGHT RANGES AND ATTACHMENTS]. I have [EXPERIENCE], [MINUTES] per session, and these limitations/preferences: [LIST]. Give exercises, warm-up sets, working sets, rep ranges, rest times, weekly set totals by muscle group, and a progression rule. For each exercise give one substitution from my equipment. Check that the weekly volume and schedule are internally consistent. Do not diagnose pain or prescribe rehab. ``` --- ## 05. Recovery and Fatigue Review - Category: Recovery - Risk: Level 2 — Analyze, then verify - Scope: Level 2. - You provide: At least 2–4 weeks of sleep, resting heart rate, HRV, training volume and intensity, performance, soreness, steps, calories, body weight, stress, illness, and alcohol. - AI does well: Combines several recovery signals and forces the recommendation to name both supporting and contradictory evidence. - Main failure mode: Making a confident decision from one noisy score or mistaking correlation for cause. - Verify by: Compare the call with your normal baseline, symptoms, and actual training performance; use a conservative adjustment when confidence is low. ### Copy-ready prompt ```text Review my recovery data for [DATE RANGE]: [PASTE TABLE/EXPORT]. Classify today as train as planned, reduce volume/intensity, or rest. Base the call on trends, not one reading. Name the specific signals that support and contradict the recommendation, distinguish device data from self-reported symptoms, and state confidence. If there is not enough baseline data, say so. Give one conservative adjustment and the trigger for returning to normal training. ``` --- ## 06. Blood-Work Trend Organization - Category: Labs and medical prep - Risk: Level 3 — Prepare a professional question - Scope: Level 3 for interpretation and action. - You provide: Complete values, units, lab-specific ranges, collection date and time, fasting status, same-lab indicator, medications and supplements, recent training or illness, and prior panels. - AI does well: Normalizes panels over time, explains terms, and prepares a compact factual agenda for a clinician. - Main failure mode: Diagnosing from isolated values, using the wrong units or ranges, or ignoring context. - Verify by: Compare against the original report line by line and bring the organized table and questions to the clinician who ordered or interprets the tests. ### Copy-ready prompt ```text Organize these blood panels for a clinician discussion: [PASTE COMPLETE RESULTS WITH UNITS, LAB RANGES, DATES, AND FASTING STATUS]. Relevant medications, supplements, symptoms, recent illness, and hard training are: [LIST]. Create a table showing each marker over time, percent and absolute change, whether the lab flagged it, and plain-language definitions. Separate what the numbers literally show from possible explanations. Do not diagnose or recommend treatment. Identify missing context and write the 7 highest-value questions for my clinician. ``` --- ## 07. Stretching, Warm-Up, and Mobility Routine - Category: Training - Risk: Level 1 — Draft and organize - Scope: Level 1 for general mobility; Level 3 when symptoms suggest injury or neurologic involvement. - You provide: Exact movement restriction, location, onset, aggravating and easing movements, training context, pain scale, and red-flag screening. - AI does well: Creates a short routine with a purpose, dose, stop rule, and simple way to track whether it helps. - Main failure mode: Guessing a diagnosis or assigning corrective drills that worsen an injury. - Verify by: Screen the listed red flags first, stop on worsening symptoms, and seek a physio assessment if the planned tracking test does not improve. ### Copy-ready prompt ```text Build a 10-minute warm-up and mobility routine for this movement goal: [GOAL/RESTRICTION]. It appears during [MOVEMENT], began [WHEN], feels like [DESCRIPTION], and changes when [AGGRAVATING/EASING FACTORS]. First list the symptoms that make self-directed mobility inappropriate. If none are present, give low-risk drills, dose, what each drill is intended to change, and a stop rule. Do not diagnose the cause. Give a two-week tracking test and tell me what lack of improvement should trigger a physio assessment. ``` --- ## 08. Standardized Progress-Photo Comparison - Category: Photos and video - Risk: Level 2 — Analyze, then verify - Scope: Level 2, with additional privacy sensitivity. - You provide: Same pose, distance, camera, lighting, clothing, time of day, hydration context, and date interval. - AI does well: Annotates visible changes while explicitly separating stronger observations from camera, lighting, pose, hydration, and pump effects. - Main failure mode: Reading lighting, posture, pump, or lens distortion as body-composition change. - Verify by: Repeat the photos under the same controls and compare them with measurements or other independent trend data. ### Copy-ready prompt ```text Compare these two standardized photo sets from [DATES]. The camera, distance, lighting, pose, clothing, and time of day were [SAME/DIFFERENT IN THESE WAYS]. Describe only visible differences by region. Label each as high, medium, or low confidence. Explicitly identify changes that may come from pose, lighting, camera angle, hydration, or muscle pump. Do not estimate body-fat percentage or diagnose health. End with the measurements or repeat-photo controls that would verify the trend. ``` --- ## 09. Personal Health-Data Correlation Analysis - Category: Personal-data platform - Risk: Level 2 — Analyze, then verify - Scope: Level 2; Level 3 when the outcome informs treatment. - You provide: Tidy timestamped data dictionary, units, missingness, baseline period, known interventions, and enough observations. - AI does well: Checks coverage and missingness, ranks plausible relationships, and turns an exploratory pattern into a prospective test. - Main failure mode: Spurious correlations, autocorrelation, time trends, multiple testing, and causal claims. - Verify by: Reject tiny samples, inspect contradictory observations, and test the strongest plausible pattern prospectively before acting on it. ### Copy-ready prompt ```text Analyze this timestamped health dataset: [CSV/JSON + DATA DICTIONARY]. Outcomes of interest are [OUTCOMES]. Candidate drivers are [VARIABLES]. Report sample size, missingness, time coverage, lag assumptions, effect size, uncertainty, and sensitivity to outliers. Correct for testing many relationships or clearly label the exploratory nature. Do not call correlation causation. Reject results below [MINIMUM SAMPLE] observations and propose a prospective two-week or four-week test for the strongest plausible pattern. ``` --- ## 10. Wearable Trend Interpretation - Category: Wearables and glucose - Risk: Level 2 — Analyze, then verify - Scope: Level 2. - You provide: Raw or daily values for at least 30 days, device and model, baseline, travel or illness, training, alcohol, caffeine, late meals, and stress. - AI does well: Starts with raw measures, looks for repeatable behavior associations, and surfaces contradictory days and confounders. - Main failure mode: Treating proprietary readiness or sleep-stage scores as ground truth. - Verify by: Compare the model’s claim with raw trends and run one low-risk experiment; concerning symptoms go to a clinician regardless of the score. ### Copy-ready prompt ```text Review [30/60/90] days from my [DEVICE]: [EXPORT]. My behavior log is [TRAINING, ALCOHOL, CAFFEINE, LATE MEALS, TRAVEL, ILLNESS, STRESS]. Analyze trends in raw measures first, then the device's composite scores. Rank three candidate behavior associations by effect size and consistency. Show contradictory days and confounders. Do not diagnose a sleep or heart condition. Recommend one low-risk experiment and the symptoms that should go to a clinician regardless of the wearable score. ``` --- ## 11. Food-Photo Calorie and Macro Range - Category: Nutrition - Risk: Level 2 — Analyze, then verify - Scope: Level 2. - You provide: Photo plus plate or container size, ingredient list, preparation method, restaurant or recipe name, sauces, oils, and portion clues. - AI does well: Breaks a meal into components and makes uncertainty visible as low, midpoint, and high estimates. - Main failure mode: Missing invisible fats and portion depth or presenting false single-number precision. - Verify by: Check the ingredient, oil, sauce, and portion assumption that contributes the most uncertainty against a label or measurement. ### Copy-ready prompt ```text Estimate this meal as a range, not a single number. I will provide the photo plus: plate/container size [SIZE], ingredients I know [LIST], cooking method [METHOD], sauces/oils [LIST/UNKNOWN], and restaurant or recipe [IF KNOWN]. Break the meal into components with portion ranges, calorie ranges, protein/carbs/fat ranges, confidence, and the hidden variable that matters most. Give a low, midpoint, and high total. Tell me what one measurement or label would reduce uncertainty most. ``` --- ## 12. Recipe Conversion to Fit a Target - Category: Nutrition - Risk: Level 1 — Draft and organize - Scope: Level 1. - You provide: Original recipe with weights and servings, target per serving, non-negotiable flavors and textures, allergies, and available substitutes. - AI does well: Recalculates a recipe against a target and explains the flavor, texture, and yield tradeoffs of each swap. - Main failure mode: Invented nutrition values or substitutions that ruin yield and texture. - Verify by: Recalculate from the cited database or product labels and confirm the final yield before trusting per-serving numbers. ### Copy-ready prompt ```text Modify this recipe to reach about [CALORIES] kcal and [PROTEIN] g protein per serving: [RECIPE WITH WEIGHTS AND SERVINGS]. Preserve [FLAVOR/TEXTURE NON-NEGOTIABLES] and avoid [INGREDIENTS]. Show original versus modified ingredient weights, yield, estimated macros, and the source/label used for each nutrition value. Explain each swap and its likely taste/texture effect. Recalculate the totals visibly and give one less-aggressive alternative. ``` --- ## 13. Grocery List and Meal-Prep Sequence - Category: Nutrition - Risk: Level 1 — Draft and organize - Scope: Level 1. - You provide: Final meal plan, household servings, pantry inventory, store preference, budget, equipment, storage space, and prep window. - AI does well: Consolidates quantities and turns a final menu into a practical, parallel prep sequence. - Main failure mode: Duplicate quantities, unsafe storage assumptions, or impossible parallel timing. - Verify by: Reconcile every quantity back to the final meal plan and check storage guidance against authoritative food-safety advice. ### Copy-ready prompt ```text Turn this final meal plan into: (1) one consolidated grocery list by store section with purchase quantities, (2) a [TIME]-minute prep sequence showing tasks that run in parallel, and (3) a storage/reheat plan: [MEAL PLAN]. Subtract this pantry inventory: [LIST]. State food-safety storage assumptions, mark items better cooked later, and reconcile every grocery quantity back to the plan. ``` --- ## 14. Weight-Loss Plateau Troubleshooting - Category: Nutrition - Risk: Level 2 — Analyze, then verify - Scope: Level 2. - You provide: Daily weigh-ins, rolling average, waist, intake, adherence, steps, training, cycle if relevant, sodium and carbs, medication changes, sleep, and duration. - AI does well: Separates scale noise from a sustained trend and forces a diagnostic period before another calorie cut. - Main failure mode: Declaring a plateau too early or cutting calories before checking noise and adherence. - Verify by: Use rolling averages, fill missing adherence data, and run the two-week diagnostic plan before changing intake. ### Copy-ready prompt ```text My weight trend has been [DESCRIBE] for [WEEKS]. Here are daily weights, calories/macros, adherence notes, steps, training, waist, sleep, sodium/carbohydrate changes, and medication changes: [DATA]. Use rolling averages and separate scale noise from a meaningful stall. Rank plausible explanations, cite the evidence for and against each, and identify missing data. Give a two-week diagnostic plan before suggesting a calorie change. Do not recommend extreme restriction. ``` --- ## 15. Lift-Video Technique Review - Category: Photos and video - Risk: Level 2 — Analyze, then verify - Scope: Level 2. - You provide: Correct camera angle, full body and implement visible, load, reps, set number, goal, and pain status. - AI does well: Reviews visible setup, path, tempo, range, balance, and repeatability while naming what the camera cannot show. - Main failure mode: Judging joints or bar path that the camera cannot see, or implying an injury diagnosis. - Verify by: Film the recommended second angle and compare only the stated visible cues; pain or injury questions need a qualified professional. ### Copy-ready prompt ```text Review this [LIFT] video filmed from [ANGLE] at [LOAD] for [REPS]. My goal is [TECHNIQUE/STRENGTH/HYPERTROPHY] and I have [NO PAIN / DESCRIBE SYMPTOMS]. Describe only what is visible: setup, bar or implement path, range of motion, tempo, balance, and repeatability. List the two highest-priority changes with one cue each. State every important thing the camera angle prevents you from judging. Do not diagnose injury. Recommend the next camera angle for verification. ``` --- ## 16. Training Modification Around a Tweak - Category: Training - Risk: Level 3 — Prepare a professional question - Scope: Level 2 to Level 3. - You provide: Location, onset, mechanism, severity, symptoms at rest, swelling or bruising, neurologic symptoms, aggravating movements, and current program. - AI does well: Creates a conservative modification that protects unaffected training and starts with reasons to stop and seek assessment. - Main failure mode: Treating a serious injury as a normal tweak or producing overconfident rehab programming. - Verify by: Use the red-flag screen first, stop on worsening symptoms, and have a qualified professional assess anything concerning or persistent. ### Copy-ready prompt ```text Help me modify, not diagnose, this training week. The issue is [LOCATION/DESCRIPTION], began [WHEN/HOW], pain is [0-10], appears during [MOVEMENTS], and at rest [DETAIL]. Swelling, bruising, weakness, numbness, tingling, fever, or deformity: [YES/NO FOR EACH]. Current program: [PASTE]. Start with reasons I should stop and seek urgent or professional assessment. If self-management is reasonable, keep unaffected training, remove or reduce aggravating loads, and give conservative criteria for reintroduction. Do not name a diagnosis. ``` --- ## 17. Periodization and Deload Planning - Category: Training - Risk: Level 2 — Analyze, then verify - Scope: Level 1 to Level 2. - You provide: Goal, competition or date, training history, recent performance, current weekly volume, available days, and recovery constraints. - AI does well: Turns a current program into a time-bounded block with explicit progression, deload, and autoregulation rules. - Main failure mode: Arbitrary phases, excessive fatigue, or vague autoregulation. - Verify by: Audit weekly volume, confirm the schedule fits real recovery constraints, and use performance to adjust rather than following the block blindly. ### Copy-ready prompt ```text Build a [WEEKS]-week block for [GOAL/EVENT] from this current program and performance history: [PASTE]. I can train [DAYS/MINUTES], and my recovery constraints are [LIST]. Give phase goals, weekly set/rep/load targets, progression, planned deload logic, and exact autoregulation rules for missed reps, poor sleep, illness, or unusually high fatigue. Show weekly volume by main movement or muscle group and explain the tradeoffs. ``` --- ## 18. Restaurant-Menu Navigation - Category: Nutrition - Risk: Level 2 — Analyze, then verify - Scope: Level 1 to Level 2. - You provide: Menu link or photo, remaining calorie and macro target, allergies, dietary rules, hunger level, and willingness to modify. - AI does well: Ranks realistic choices as ranges and identifies the oil, sauce, portion, and allergy questions that matter most. - Main failure mode: False precision where preparation and portions are unknown. - Verify by: Confirm allergy details with the restaurant and treat calorie and macro values as ranges unless official nutrition facts exist. ### Copy-ready prompt ```text Review this restaurant menu: [LINK/PHOTO/TEXT]. My remaining target is roughly [CALORIES AND MACROS], allergies are [LIST], and I prefer [FOODS]. Rank five options using calorie and macro ranges. Show the assumptions, exact modifications to request, and the uncertainty from oil, butter, sauce, and portion size. Do not invent official nutrition facts. Mark any option that requires confirmation from the restaurant for an allergy. ``` --- ## 19. Supplement or Fitness Claim Audit - Category: Supplements - Risk: Level 2 — Analyze, then verify - Scope: Level 2. - You provide: Exact claim, product or ingredient and dose, target population, cited study, and outcome. - AI does well: Defines the claim precisely and compares the marketing language with the best relevant human evidence. - Main failure mode: Counting studies instead of assessing quality, substituting animal or mechanistic evidence for human outcomes, or ignoring funding. - Verify by: Open the strongest sources, check population and dose, and see whether the measured outcome actually matches the advertised outcome. ### Copy-ready prompt ```text Audit this exact health or fitness claim: [CLAIM + SOURCE/LINK]. Define the population, intervention, comparator, outcome, dose, and time frame implied by the claim. Find the best human evidence and prioritize systematic reviews, meta-analyses, trials, and official guidance. Report effect sizes and uncertainty where available, funding/conflicts, and whether the marketing outcome matches the measured outcome. Rate the claim strong, promising, weak, unsupported, or contradicted, and explain what evidence would change the rating. ``` --- ## 20. Doctor-Appointment Preparation - Category: Labs and medical prep - Risk: Level 3 — Prepare a professional question - Scope: Level 2 to Level 3. - You provide: Concise history, symptoms and dates, medications and supplements, measurements, questions, prior tests, and goal for the visit. - AI does well: Turns scattered notes into a factual one-page timeline, medication list, missing-record checklist, and focused questions. - Main failure mode: Producing a bloated speculative summary or steering the clinician toward an AI-generated diagnosis. - Verify by: Compare the summary with the original records, correct every date and dose, and bring it as preparation—not a diagnosis. ### Copy-ready prompt ```text Prepare me for an appointment about [TOPIC]. Here is my timeline, symptoms, medications/supplements, measurements, prior tests, and concerns: [PASTE]. Create a one-page factual summary with dates, a medication/supplement list, the five highest-value questions, and missing records to bring. Separate facts from my interpretations. Do not propose a diagnosis or tell the clinician what treatment to prescribe. ``` --- ## 21. CGM and Meal-Response Analysis - Category: Wearables and glucose - Risk: Level 3 — Prepare a professional question - Scope: Level 2 to Level 3. - You provide: Timestamped CGM export, meals and portions, exercise, sleep, stress, medication, sensor changes, symptoms, and timezone. - AI does well: Looks for repeatable meal and behavior patterns while counting observations and naming confounders. - Main failure mode: Overinterpreting normal variability, ignoring sensor lag or error, or recommending treatment changes. - Verify by: Repeat low-risk experiments and take patterns that could change treatment or explain symptoms to a clinician. ### Copy-ready prompt ```text Analyze this CGM export with my meal, activity, sleep, stress, and medication log: [DATA]. Use the correct timezone and flag missing or implausible readings. Identify repeatable associations, not isolated spikes. For each pattern show how many observations support it, timing, magnitude, confounders, and confidence. Suggest low-risk experiments such as meal order, portion, pairing, timing, or a walk. Do not recommend medication changes or diagnose glucose problems; list patterns to discuss with a clinician. ``` --- ## 22. Accountability and Weekly Check-Ins - Category: Habits and planning - Risk: Level 1 — Draft and organize - Scope: Level 1. - You provide: A small set of controllable behaviors, target ranges, daily logging format, and weekly review schedule. - AI does well: Keeps check-ins short, tracks rolling behavior patterns, and turns a missed week into one small adjustment. - Main failure mode: Cheerleading, shame, overreacting to single days, or changing targets constantly. - Verify by: Review the weekly summary against the raw check-ins and change only one controllable bottleneck at a time. ### Copy-ready prompt ```text Act as a concise accountability coach for these behaviors: [3-5 BEHAVIORS AND TARGETS]. At each check-in ask only for [FIELDS]. Track completion, rolling averages, and the reason for misses. Keep daily replies under 100 words. Every seven entries, summarize the pattern, identify one bottleneck, challenge one excuse, and propose one small change for the next week. Do not change the plan from a single bad day or use shame-based language. ``` --- ## 23. Sleep and Recovery Pattern Review - Category: Recovery - Risk: Level 2 — Analyze, then verify - Scope: Level 2. - You provide: Sleep and wake times, time in bed, device estimates, subjective quality, caffeine and alcohol, exercise, light exposure, travel, and symptoms. - AI does well: Prioritizes schedule, duration, awakenings, and daytime function before noisy proprietary sleep-stage estimates. - Main failure mode: Treating consumer sleep stages as a clinical sleep study. - Verify by: Run one two-week behavior experiment and seek medical evaluation for persistent or concerning symptoms. ### Copy-ready prompt ```text Review [WEEKS] of sleep timing, device estimates, subjective sleep quality, caffeine, alcohol, training, travel, and morning energy: [DATA]. Prioritize stable observations such as schedule, duration, awakenings, and daytime function before proprietary sleep stages. Rank three behavior associations with contradictory examples and confidence. Propose one two-week sleep experiment. Do not diagnose a sleep disorder; list symptoms that warrant medical evaluation. ``` --- ## 24. Progressive-Overload Autoregulation - Category: Training - Risk: Level 2 — Analyze, then verify - Scope: Level 1 to Level 2. - You provide: Exercise history, load and reps, RPE or RIR, technique notes, target rep range, equipment increments, and recovery notes. - AI does well: Makes one bounded next-session decision per lift and cites the exact log entries behind it. - Main failure mode: Increasing load from noisy effort data or without a stable progression rule. - Verify by: Keep every change inside the stated progression rule and repeat the session when technique or effort data are inconsistent. ### Copy-ready prompt ```text Use this training log to recommend only the next session for each lift: [LOG WITH LOAD, REPS, SETS, RIR/RPE, AND TECHNIQUE NOTES]. My progression rule is [RULE] and available load increments are [INCREMENTS]. For each lift choose increase load, add reps, repeat, or reduce. Cite the log entries that drive the decision, show the exact target, and keep changes inside the progression rule. If technique or effort data are inconsistent, repeat rather than invent certainty. ``` --- ## 25. Micronutrient Gap Screen From Diet Logs - Category: Nutrition - Risk: Level 2 — Analyze, then verify - Scope: Level 2. - You provide: At least 7–14 days of weighed food data, brands and fortification, supplements, age and sex, and relevant medical context. - AI does well: Separates food and supplement contributions and identifies possible intake gaps without calling them deficiencies. - Main failure mode: Incomplete food database entries, supplement double-counting, and jumping from an intake estimate to a deficiency diagnosis. - Verify by: Check database completeness and supplement totals; diagnosis or treatment of a suspected deficiency requires a clinician and appropriate testing. ### Copy-ready prompt ```text Screen this [7-14]-day weighed food log plus supplement list for possible micronutrient intake gaps: [DATA]. My age/sex and relevant context are [DETAILS]. Use a named authoritative reference for intake targets. Report average intake, data completeness, food-versus-supplement contribution, and uncertainty from missing brand/fortification data. Say "possible intake gap," not "deficiency." Suggest food-first options and list anything that requires a clinician or lab assessment. ``` --- ## 26. Habit Friction and Adherence Design - Category: Habits and planning - Risk: Level 1 — Draft and organize - Scope: Level 1. - You provide: Target behavior, current routine, failure moments, environment, schedule, motivation, and past attempts. - AI does well: Separates motivation from time, cue, environment, skill, and plan problems, then designs a small prospective test. - Main failure mode: Generic motivation advice instead of changing cues and friction. - Verify by: Use a simple pass or fail measure for two weeks and judge the environmental change from the result, not the pep talk. ### Copy-ready prompt ```text Analyze why this health habit keeps failing: [HABIT + 14 DAYS OF CHECK-INS]. My schedule, environment, and common failure moments are [DETAILS]. Separate motivation problems from time, cue, environment, skill, and plan problems. Find the smallest reliable version of the habit, one environmental change, one implementation intention, and one fallback for bad days. Design a two-week test with a simple pass/fail metric. ``` --- ## 27. Travel, Shift-Work, or High-Stress Adaptation - Category: Habits and planning - Risk: Level 2 — Analyze, then verify - Scope: Level 1 to Level 2. - You provide: Dates and time zones, work shifts, sleep opportunities, equipment, meal access, current plan, and non-negotiable priorities. - AI does well: Protects the highest-value behaviors and creates minimum, normal, and recovery-day versions for an abnormal week. - Main failure mode: Pretending the normal plan fits an abnormal week or recommending extreme sleep or caffeine tactics. - Verify by: Check the schedule against real sleep, meal, and equipment access and keep caffeine or supplement changes conservative. ### Copy-ready prompt ```text Adapt my normal training, meals, sleep, supplements, and step targets for this [TRAVEL/SHIFT-WORK/HIGH-STRESS] period: [SCHEDULE, TIME ZONES, ACCESS, AND CONSTRAINTS]. Protect the highest-value behaviors and deliberately reduce lower-priority volume. Give a minimum plan, a normal plan, and a recovery-day plan. Keep caffeine and supplement advice conservative and evidence-linked. State what should wait until my schedule normalizes. ``` --- ## 28. Personal Health-Platform Data Architecture - Category: Personal-data platform - Risk: Level 2 — Analyze, then verify - Scope: Level 2, with privacy and security implications. - You provide: Sources, fields, cadence, goals, threat model, local or cloud preference, retention policy, and intended AI tasks. - AI does well: Designs a minimal, reversible, auditable data system and limits the AI layer to the access it actually needs. - Main failure mode: Collecting everything, weak identity separation, accidental public logs, or giving an AI agent unnecessary write access. - Verify by: Threat-model the design, test restores and deletions, review access logs, and keep the AI interface read-only by default. ### Copy-ready prompt ```text Design a privacy-minimized personal health-data system for these sources and goals: [SOURCES, FIELDS, CADENCE, AND QUESTIONS]. My preferred storage is [LOCAL/CLOUD/HYBRID] and threat model is [DETAILS]. Create a data dictionary, canonical timestamps/units, import flow, validation checks, retention schedule, identifier-removal step, role-based access, audit log, backup plan, and read-only AI interface. Collect only fields that answer a defined question. Keep medical records and public/creator data separated. Propose a reversible export format and a plan for deleting source and derived data. ``` # Questions people ask ### What is the best AI model for health and fitness prompts? Use a capable model that can follow long instructions, work with tables or images when needed, and cite sources. The model name matters less than giving it clean inputs, requiring visible math and uncertainty, and verifying anything consequential. ### Can AI diagnose a health problem from my data? No. A general AI assistant can organize information, explain terms, compare trends, and prepare questions. It should not diagnose, prescribe, clear medication or supplement combinations, or replace a qualified professional. ### What should I try first? Start with a macro meal draft, an equipment-constrained workout split, a weekly accountability review, or a factual doctor-appointment summary. These are useful, bounded tasks with clear inputs and verification steps. ### Is it safe to upload health data to a chatbot? Treat health data as sensitive. Minimize what you upload, remove identifiers and photo metadata, and review the product's current retention, training, deletion, and sharing controls. A consumer app is not automatically covered by HIPAA. ### Can I use Matt's personal health platform? Not today. It is Matt's personal system for bringing training, nutrition, sleep, weight, labs, glucose, experiments, and an AI coach into one place. It may be shared more broadly in the future, but there is no public release date or repository. # Primary and official sources Sources were checked on July 20, 2026. Peer-reviewed evidence is labeled separately from official guidance. 1. [World Health Organization: Ethics and governance guidance for large multimodal models in health](https://www.who.int/news/item/18-01-2024-who-releases-ai-ethics-and-governance-guidance-for-large-multi-modal-models) — Health AI should be designed and used with explicit attention to safety, transparency, accountability, and human oversight. 2. [U.S. Food and Drug Administration: Artificial intelligence-enabled medical devices](https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices) — Regulated medical-device evidence is not interchangeable with a general-purpose chatbot answer. 3. [U.S. Department of Health and Human Services: Access rights, health apps, and APIs](https://www.hhs.gov/hipaa/for-professionals/privacy/guidance/access-right-health-apps-apis/index.html) — Data sent to a consumer-selected app may no longer be protected by the HIPAA Rules when the app is not a covered entity or business associate. 4. [Federal Trade Commission: Complying with the Health Breach Notification Rule](https://www.ftc.gov/business-guidance/resources/complying-ftcs-health-breach-notification-rule-0) — Some health apps and connected services outside HIPAA still have breach-notification responsibilities. 5. [NIH Office of Dietary Supplements: Dietary supplement fact sheets](https://ods.od.nih.gov/factsheets/list-all/) — Ingredient, chemical form, dose, and population matter; supplement timing and interaction claims are not universal. 6. [National Center for Complementary and Integrative Health: How medications and supplements can interact](https://www.nccih.nih.gov/health/know-science/how-medications-and-supplements-can-interact/interactions-with-over-the-counter-medications) — Medication and supplement combinations need authoritative checking and professional review. 7. [Centers for Disease Control and Prevention: Steps for losing weight](https://www.cdc.gov/healthy-weight-growth/losing-weight/index.html) — Sustainable weight management depends on a realistic plan, habits, monitoring, and appropriate professional support. 8. [American College of Sports Medicine: Resistance training guidelines update](https://acsm.org/resistance-training-guidelines-update-2026/) — Training plans need sensible frequency, volume, progression, and fit with the person's experience and goals. 9. [American Academy of Sleep Medicine: Consumer sleep technology position statement](https://aasm.org/consumer-sleep-technology-position-statement/) — Consumer sleep technology can support discussion and tracking but should not diagnose or treat sleep disorders. 10. [MedlinePlus: How to understand your lab results](https://medlineplus.gov/lab-tests/how-to-understand-your-lab-results/) — Units, lab-specific ranges, history, symptoms, and other tests all affect interpretation. 11. [U.S. Food and Drug Administration: FDA clears first over-the-counter continuous glucose monitor](https://www.fda.gov/news-events/press-announcements/fda-clears-first-over-counter-continuous-glucose-monitor) — Users should not make medical decisions from OTC CGM output without talking to a healthcare provider. 12. [Peer-reviewed systematic review: Image-based dietary assessment: a systematic review](https://pmc.ncbi.nlm.nih.gov/articles/PMC10836267/) — Food-image estimates vary widely and should be handled as ranges with visible uncertainty. # Important caveat This guide is for planning, organization, trend review, and better questions. It is not medical advice, diagnosis, treatment, or emergency guidance. If symptoms are urgent, severe, new, or worsening, use appropriate medical care instead of waiting for an AI answer.