Create an AI Training Plan Yourself: Using ChatGPT, Claude and Gemini Properly

Want to create your own training plan? Good call. The problem: if you just type "make me a training plan", you get a generic plan for an imaginary average person. No regard for your level, your goals, your limitations.

The solution: a good prompt and your context. The difference isn't in the tool, it's in what you bring to the table: your training knowledge, your documents, your numbers. This guide walks you through it step by step – with ChatGPT, Claude or Gemini, straight from the chat, no detours. Including copy-paste prompts for three training goals and the most common mistakes I see over and over again.

Step 1: More context, better plan

The more relevant information you give the AI chatbot (large language model, or LLM for short), the better the plan. That sounds trivial – but it's the core of the prompt paradox: you need to know enough about training to supply the right information. Anyone who gets this already has a decisive edge over 90% of users.

Upload existing plans

The strongest context isn't text – just upload your old plan. ChatGPT, Claude and Gemini can read files. If you already train with a plan or have one you like, upload it. That gives the tool more information than any description could.

Using your current plan as a starting point makes sense when you've been training to a specific programme for months and want the next step. Instead of describing everything from scratch, you let the tool analyse your existing plan and build on it.

Role: You are an experienced sport scientist and strength coach.

Attached you'll find my current training plan, which I've been following for 4 months.

MY SITUATION:
- The plan has worked well, but I've been stalling on bench press and overhead press for 3 weeks
- Leg exercises are still progressing
- I now have 5 days per week available instead of 4

TASK:
1. Analyse the existing plan: what works, where are the weaknesses?
2. Create a follow-up plan for the next 8 weeks that builds on it
3. Fix the upper-body stall – explain what you're changing and why

NOT WANTED:
- No completely new plan that ignores what I've been doing
- No changes to leg exercises without justification

Providing a reference format is worth it when you've seen a plan whose structure you like – the table layout, the type of progression, the level of detail. Upload it and tell the tool: "Create my plan in the same format." That steers the result better than any format description. Uploads work with PDFs, Excel files, screenshots and photos of handwritten plans.

Step 2: With the right prompt structure to a plan that works

The most common mistake: asking for a plan directly, without context. Here's the difference.

Prompt – how NOT to do it

Make me a training plan for muscle building, 4 days a week.

The result: a generic plan for a made-up average person. No progression, no regard for your level, no safety limits. The tool has no choice – you gave it nothing to work with.

Your code or prompt here...

A role definition forces the tool into a professional framing. Concrete numbers stop it from guessing. Exclusions filter out common weaknesses in AI-generated plans. And the format field makes sure you get what you need – not what the tool thinks looks nice.

Step 3: Test, adjust, improve

The first plan is rarely perfect – and that's a good thing. This is exactly where the real advantage of prompting yourself shows up: you can intervene in real time, shape the plan and have the tool explain why it made certain decisions. That turns a generated draft into your plan.

Adjust and deepen:

• "Replace barbell rows with an alternative – I have lower back issues with them"

• "Explain the logic behind the exercise order on push day"

• "Make Friday shorter, max 45 minutes"

Expand:

• "Add warm-up routines for every day"

• "Create a deload week for weeks 4 and 8"

• "What do I do when I can't hit a weight anymore?"

Think of the tool as an informed sparring partner, not a vending machine. You give feedback, it produces a new version – until the plan really fits you. That's exactly the difference from a ready-made plan off the internet.

Step 4: Platform features that take your workflow to the next level

Once you've built a good plan – you don't want to start from zero next time. ChatGPT, Claude and Gemini have features that cut the effort to a minimum. Worth knowing.

ChatGPT: Custom GPTs & Memory

Custom GPTs are specialised ChatGPT versions with their own persona and knowledge base. You describe your fitness coach, upload reference material (training programmes, exercise databases) and save the whole thing. From then on, every training request automatically has that context.

Memory remembers information across conversations. Enter your stats once, available permanently. Convenient – but with pitfalls.

Projects (Team/Enterprise) allow shared workspaces with persistent files and instructions – relevant for coaches working with several athletes.

Claude: Projects & Skills

Projects are persistent workspaces with their own knowledge base. You create a project, upload training data and reference material, define fixed instructions (e.g. "Always answer with concrete numbers, no vague recommendations") – and every conversation in the project has access to that knowledge.

Skills are uploadable instruction sets that define Claude's behaviour for a specific task. File outputs (Excel, PDF, PPTX) are one use case — but assessment before planning can also be structured this way. The Training Plan Skill for Claude does exactly that.

Gemini: Gems & Google integration

Gems are Gemini's counterpart to Custom GPTs – with one decisive advantage: integration with Google Drive. You create a Gem with fitness coach instructions and link a Google Doc with your training log. When you update the doc, your Gem sees the changes automatically. No manual uploading needed.

Deep Research automatically searches scientific sources for complex questions and produces a report – useful for questions like "Create an evidence-based plan for muscle building in advanced athletes."

Features compared

Prompt templates for your AI training plan

You replace the placeholders in square brackets with your own data. All three templates follow the same structure: role, context, goal, exclusions, format. The more you get out of the brackets, the better the plan.

Strength training (muscle building)

For hypertrophy, strength gains or both. Enter your current maxes – without numbers the tool guesses, and you'll notice in the result.

Your code or prompt here...

Running (race preparation)

For a half marathon, marathon or any other distance. Enter your last race time or your current training times – only then can the tool set realistic paces.

Role: You are an experienced running coach with a sport science background.

Create a training plan for [distance] in [target time].

MY DATA:
- Current level: [last race time or current training times]
- Running experience: [X years, km/week currently]
- Weeks until the race: [X]
- Available training days: [X]
- Other activities: [strength training, other sports]
- Limitations: [injury history, time constraints]

NOT WANTED:
- No sessions longer than [X] km without justification
- No pace targets in min/km without reference to my current performance

FORMAT: Weekly structure with session type (interval, tempo, long run, easy run), distance, pace target, heart rate zone. Tapering phase (reduced training volume before the race) in the last [X] weeks.

If you really want to go deep, here's how to build your own running training plan with AI – for 5k, 10k, half marathon and marathon!

General fitness

For beginners or anyone coming back after a break. Describe your current state honestly – "inactive for 6 months" isn't a problem, but the tool needs to know.

Your code or prompt here...

All templates follow the same principle: define the role, provide context, concretise the task, set the format, explicitly exclude weaknesses. You replace the placeholders in square brackets with your own data – and that's exactly why you need the baseline knowledge that no prompt can replace.

What AI tools can do – and where you're on your own

Even with clean prompts, real limits remain – and you should know them before blindly relying on the output.

• No real-time state. The tool can't see how you feel today – whether you slept badly, have DOMS from yesterday or are stressed. It plans based on what you enter, not on what's actually going on right now.

• No technique correction. A prompt can give you execution cues for squats – but whether your knees cave in is something the tool can't see. For technique you need eyes, not text.

• Quality rises with knowledge input. The prompt paradox, empirically confirmed: beginners with little context get weak plans, advanced users with precise data get usable ones. That's shown by a study on AI training plans for the marathon.

• No substitute for pain or injuries. For acute complaints, chronic conditions or post-injury, you go to a doctor or physiotherapist – not to a chatbot. The tool can factor in limitations, but it can't diagnose.

• No motivation. Informing yes, driving you no. An LLM can explain to you why progressive overload (systematically increasing load) works – but it can't make sure you still head to the gym on Tuesday evening.

AI tools are planning instruments, not coaches. Execution is on you. Where specialised fitness apps have a structural edge – real-time sensor data and adaptive adjustment – is laid out in the direct comparison of LLMs vs. fitness apps.

Your plan, your control

Context is everything – the more relevant information, the better the plan. The first draft is the beginning, not the end: iterate until the logic holds. Use platform features for recurring planning. And above all: keep asking until you understand the plan. Then a generated text turns into your training plan – one you built and thought through yourself. If you apply the same approach to nutrition, the create your own AI nutrition plan guide is the next logical step.

What AI can actually do in training – and where it's used – is shown in the overview: where AI is used in fitness.