Ask AI to improve your training program — safely.

The promise — and the real risk

The appeal is obvious: you describe your goals, an AI spits out a program, and you train. The problem is equally obvious: a generic plan has no idea that your left shoulder stalls every push session, that you have been skipping leg days, or that your chest volume is already twice what your back receives. Without data, AI improvises. With data, it reasons.

There is a second risk beyond bad programming: an AI that changes your program silently, without your review, can introduce errors — a movement you cannot perform, a volume jump your joints are not ready for, a session structure that conflicts with your recovery. The safety mechanism that matters is not smarter AI; it is a supervised approval model where every proposed change is visible before it takes effect.

Why data-grounded AI beats a generic plan

When you connect SammaGym to ChatGPT or Claude via the MCP connector at https://app.sammagym.com/api/mcp, the AI gains read access to the information that actually matters for programming decisions.

• Volume per muscle group across your recent weeks — not what the program prescribes, but what you actually completed.
• Your active program in full detail: exercises, sets, rep targets, and structure.
• Strength progress with estimated 1RM calculated via the Epley formula, so the AI can see where you are stalling and where you are progressing.
• Bodyweight trend over time.
• Cardio metrics — sessions, distances in km, pace, and calories — without GPS coordinates.
• Timed workout results from AMRAP, EMOM, Tabata, and For Time sessions.

Research on hypertrophy consistently points to weekly volume per muscle group as a key driver of growth. Schoenfeld documents the dose-response relationship: too little volume leaves adaptation on the table; too much accumulates fatigue without additional gain. The NSCA frames progressive overload as the foundational principle — systematic increases in load, volume, or density over time. When AI can see that your quads have received eight sets per week for the past six weeks while your hamstrings have averaged three, it is not guessing at an imbalance. It is reading one.

Generic AI programs are written for an imaginary average athlete. Data-grounded AI reviews are written for the athlete you actually are, based on what you have actually done.

What you can ask AI to do with your program

Once connected, you can ask your AI to propose a range of meaningful changes. The AI submits proposals; you apply them.

• Identify and fix muscle group imbalances. The AI reads your volume distribution and can propose adding sets to undertrained groups or redistributing sessions.
• Fix exercise order. Compounds should precede isolation work in most sessions. If your program has this inverted somewhere, the AI can spot and correct it.
• Add a timed finisher. An AMRAP, EMOM, Tabata, or For Time block can be added to any session as a conditioning element.
• Build a program from scratch. You describe your goals, available equipment, and schedule; the AI proposes a complete program with warm-up sets, working sets, and rep targets.
• Write a deload week. When fatigue accumulates — measurable through stalled lifts, dropped consistency, or your own reported feel — the AI can propose a reduced-volume deload.
• Add intensity techniques. Rest-pause sets, cluster sets, drop sets, and pyramid rep schemes (different reps per set) can be inserted into existing exercises.
• Create custom exercises. If you need a movement not in the library, the AI can propose a new exercise with the correct muscle group tag, which is added to your library on approval.

How the supervised approval model works

The supervised approval model is the reason this workflow is safe rather than reckless. Here is the full sequence.

• 1Connect once. In ChatGPT (developer mode / custom connectors) or Claude (Pro, Max, or Team plan — Settings → Connectors → add custom connector), add the SammaGym connector at https://app.sammagym.com/api/mcp. One-click OAuth; you authorize only your own data; disconnect anytime.
• 2Ask your question. Type your request in natural language (see the prompts section below). The AI reads your data, reasons over it, and formulates a proposal.
• 3Proposal is submitted — not applied. The AI sends the proposed change to your SammaGym account. Your active program is completely unchanged at this point.
• 4Review in the Proposals screen. Open SammaGym and go to the Proposals screen. You see the full detail of what the AI wants to change — every exercise, set, rep target, technique, and note.
• 5Approve with one tap — or discard. If the proposal looks right, one tap applies it. If it looks wrong, discard it and ask the AI to revise. Your program changes only when you explicitly approve.

This is not a convenience feature. It is the core safety mechanism. No proposal bypasses your review. No change is silent. The AI is a collaborator that drafts; you are the decision-maker who approves.

Good prompts to use

These prompts work because they are specific, ask for a proposal rather than general advice, and reference your data explicitly. Copy and adapt them.

• "Look at my last 8 weeks. Which muscle group is undertrained compared to the others, and propose a fix I can approve."
• "Review my active program for exercise order. If any session has isolation work before compound lifts, propose a corrected order."
• "My squat estimated 1RM has not moved in 4 weeks. Propose one change to my lower body session that could break the stall — a set/rep adjustment, a technique, or an exercise swap."
• "Based on my volume and consistency over the last 6 weeks, propose a full deload week."
• "Build me a 4-day Upper/Lower program for hypertrophy from scratch. I have a barbell, dumbbells to 50 kg, a cable stack, and a lat pulldown. No leg curl machine."
• "Add a 10-minute AMRAP finisher to my next upper body session. Propose it so I can approve it."
• "My hamstring volume is low relative to my quads. Propose a rebalance across the week without increasing total session time."

Notice the pattern: each prompt ends with an instruction that makes the AI submit a proposal rather than give conversational advice. This routes the response through the approval workflow.

What to double-check before approving

Even well-reasoned proposals deserve a ten-second review before you approve. Run through this checklist.

• Equipment match. Does every exercise in the proposal require equipment you actually have? AI can propose movements that assume machines or barbells you do not own.
• Volume jump. If the proposal adds sets, check the week-over-week increase. The NSCA recommends conservative, gradual increases for most trainees to manage injury risk.
• Exercise order. Confirm compounds still precede isolation work in each session, and that higher-skill movements are early when you are fresh.
• Intensity technique placement. Drop sets, rest-pause, and clusters significantly extend effective time under tension. They belong at the end of an exercise's sets, not at the start, and not on every exercise in a session.
• Recovery balance. Check that the same muscle group is not trained on back-to-back days unless that is deliberate (e.g., a specialization block you requested).
• Custom exercises. If the proposal creates a new exercise, confirm the muscle group tag is correct — this affects how that movement appears in your volume tracking going forward.

If anything looks off, do not approve. Tell the AI what you noticed and ask for a revised proposal. The loop is fast, and your program stays intact until you are satisfied.