How to use ChatGPT or Claude as your personal trainer

Yes, you can use ChatGPT or Claude as your personal trainer — they're genuinely good at programming, explaining, and adjusting training for both lifting and running. The catch is that out of the box they have no memory of your training and no access to your data, so the quality of their coaching depends entirely on what you feed them. This guide covers what AI coaching does well, where it falls apart, and how to set it up properly.

What AI assistants are actually good at

A modern AI assistant has read more training literature than any human coach. Ask it to:

• Build a 4-day lifting block around a stalled bench press

• Set interval paces from a recent 5K time

• Explain why your squat feels heavy the week after a PR

• Restructure your week when travel eats two sessions

...and it gives answers a decent coach would recognise. It's available at 6 am, doesn't charge £40 ($50) an hour, and never gets bored of your questions.

This isn't hype. It's also not the whole story.

Where AI coaching falls apart

Three failure modes show up fast:

1. It forgets everything. Every new chat starts blank. The program it wrote you last week, your knee complaint, your goal race — gone. You end up re-pasting your life story every session, and eventually you stop bothering.

2. It guesses when it lacks data. Ask "should I add weight this week?" without giving your recent numbers and it will answer anyway — confidently, and from nothing. An AI without your history isn't coaching; it's improvising.

3. Progress tracking dies in chat. A coach's real job is comparing what you did against what was planned, week after week. Chat transcripts are a terrible database. Scroll back through three weeks of messages to find your last deadlift top set and you'll feel it.

None of these are intelligence problems. They're memory and data problems.

Fix one: prompt it properly (the manual way)

If you want to stay manual, give the AI a standing context block at the start of each chat: your stats, goals, program, recent sessions, injuries, equipment. Keep it in a note and paste it in. Update it after every session.

This works. It's also exactly the kind of bookkeeping you were hoping to avoid, and the moment your note drifts out of date, the advice quietly degrades with it.

Fix two: give your AI actual memory (the better way)

The structural fix is connecting your AI to a persistent store of your training. That's what Swole Protocol is: a memory layer holding your goals, program, and full history — every set, run, PR, and bodyweight reading — that your AI reads and writes through the Model Context Protocol (MCP), an open standard that lets assistants use outside tools.

Once connected:

• "Log today: 5×5 squat at 100 kg (220 lb), last set RPE 8" → saved, mid-conversation.

• "What do I need to beat on pull day?" → answered from your actual history.

• "Build next week around a sore knee" → the AI already knows what this week contained.

Any new chat, any day, and — because the memory lives outside the assistant — any AI. Claude on Tuesday and ChatGPT on Thursday see the same history.

Setting it up

Claude: connectors are supported directly in the app. Add Swole Protocol from Settings → Connectors, authorise it, and it's available in your conversations.

ChatGPT: custom MCP connections run through what ChatGPT now calls apps. On a paid plan, enable Developer mode (Settings → Apps → Advanced), add the connection on the web, and select it in a conversation. The flow is in beta and changes occasionally, so expect OpenAI's UI to move things around.

Either way it's a one-time setup of a couple of minutes.

A realistic first week

1. Tell it who you are. One honest conversation: goals, schedule, history, injuries. Your AI saves it to your athlete profile.

2. Have it build a block. Lifting, running, or both — it writes the plan as planned workouts, not a chat message that scrolls away.

3. Train and log by talking. No forms, no dashboards.

4. Review on Sunday. "How did the week go against plan?" — and get an answer built from data, not memory.

One caveat that matters: an AI is not a medical professional. For pain, injury, or health conditions, see a human. This is general information, not medical advice.

FAQ

Is ChatGPT or Claude better for training advice? Both are strong, and with a shared memory layer the honest answer is: whichever you already use. The quality gap between "AI with your data" and "AI without it" is far bigger than the gap between the assistants.

Can an AI really program running as well as lifting? Yes. Paces, intervals, easy Zone 2 volume, and race builds are pattern-heavy and well-documented — exactly what these models are good at. It just needs your recent runs to work from.

Is this safe for beginners? AI coaching is guidance, not supervision. It can't watch your form. Beginners get the most from pairing it with a few in-person sessions or honest video self-checks — and anyone with pain or a health condition should see a professional first.

Do I need to pay for ChatGPT or Claude to do this? Custom connections in ChatGPT currently require a paid plan with Developer mode. Claude supports connectors in its apps; plan requirements vary, so check your settings.

What does Swole Protocol cost? Your first 50 logged workouts are free, then roughly $15 per year.

The point

The AI you already use is a capable coach with amnesia. Cure the amnesia and it stops giving generic advice and starts coaching you — the lifter running a PPL split, the runner following a free training plan, or both at once.

Connect your AI at swoleprotocol.com.