I Tested ChatGPT, Gemini and Mistral for a Fitness Program: Unexpected Results

Have you ever wondered which AI generates the best fitness programs? I decided to find out by sending exactly the same prompt to three of the most advanced AI models on the market: ChatGPT (OpenAI), Gemini (Google) and Mistral (French). The result is instructive, sometimes surprising, and revealing of the current limitations of generative AI in fitness coaching.

The test: a simple prompt, three different AIs

Here is the message I sent to all three AIs, just as an average user might write it:

No sophisticated request. No mention of advanced techniques. Just a basic profile like thousands of users formulate every day.

ChatGPT: solid structure, but generic

What works

ChatGPT provided a well-organized response with 4 sessions respected, an emphasis on legs with 2 dedicated sessions as requested, and shoulder protection avoiding military press and dips. The time structure is clear: 5 min warm-up + 20 min workout + 5 min core.

Observed limitations

ChatGPT offers a functional and safe program for a beginner. It is a good starting point, but lacks the depth you would expect from an experienced coach.

Gemini: relevant precautions, conversational format

What works

Gemini (in thinking mode) adopted a conversational approach with an intelligent Half-Body structure (2 lower body days, 2 upper body days). Shoulder precautions are detailed in a dedicated section with clear rules. The pedagogical tone explains the why behind choices.

Face pulls are recommended as an exercise for shoulder health. The definition advice (70% nutrition, 30% training) is relevant.

Observed limitations

Gemini excels in pedagogical approach and safety precautions. However, like ChatGPT, it remains at a surface level without scientific progression.

Mistral: balanced but classic

What works

Mistral proposed a balanced approach with 4 clear sessions and precise objectives. The Full body session at the end of the week promotes active recovery. Shoulder protection avoids risky movements. Core is integrated in 3 of the 4 sessions with a cardio suggestion (jump rope) in session 3.

Observed limitations

Mistral offers a solid and safe program, but remains in conventional standards without bringing methodological innovation.

Smart Rabbit: when the user no longer needs to know how to prompt

Now, here is the fundamental difference.

With ChatGPT, Gemini and Mistral, I had to write a detailed prompt. A user who would just write "fitness program" would get an even more basic result.

With Smart Rabbit, I did not write a prompt. I simply filled out a web form with checkboxes: age 31-40 years, male gender, beginner level, very active condition, 4 sessions per week, 30 minutes per session, gym equipment, muscle definition goal, complementary activity boxing 2 times per week, left shoulder limitation, preference lots of leg work.

Smart Rabbit automatically transformed this basic profile into a 3000-line meta-prompt including all the fitness expert logic, then sent it to Claude 4.5.

Automatically generated program

4 sessions of exactly 30 minutes:

6-week progression integrated automatically

5 automatically generated justification axes

The system provided detailed explanations on:

• Muscle group distribution: Why 2 leg sessions, push/pull alternation
• Exercise selection: Functional movements, shoulder protection with compensatory unilateral rowing
• Progressive intensification: Rest-pause weeks 3-4, drop sets weeks 5-6
• Neuromuscular justifications: Motor unit recruitment, mixed fibers 8-15 reps
• Psychological justifications: Motivation, variety, boxing synergy

Automatically generated interactive interface

The program was generated in a React interface with 3 tabs (Program, Justifications, Progression), navigation between the 4 days, technical details per exercise, and tempo guide.

What makes the difference: the architecture that builds the prompt for you

The fundamental difference between Smart Rabbit and generic AIs does not lie in the language model. Smart Rabbit uses Claude 4.5, which everyone can query.

The difference lies in the fact that the user no longer needs to know how to prompt.

How it works

The 3000 lines of invisible fitness logic

When you fill out the Smart Rabbit form, the system automatically builds:

• Prioritization system: 3 exercise levels with automatic timing calculation to respect requested duration
• Precise tempo management: 5 different types (3-0-1-1, 2-0-2-0, 1-0-X-0, isometric, explosive) with slow eccentric phase for definition
• Intensification techniques: Rest-pause, drop sets, pyramids, supersets according to profile
• Intelligent adaptations: Shoulder protection with compensatory unilateral rowing, reduced volume for boxing, explosive exercises for power
• 12+ personalization parameters: Age, level, frequency, equipment, limitations, preferences

Why AI alone is not enough yet

The three tested AIs are excellent for understanding natural language, providing consistent responses, avoiding dangerous movements, and giving general advice.

But they entirely depend on the quality of the user prompt. If you do not know how to ask for precise tempos, a 6-week progression, intensification techniques, timing calculation per exercise, or injury compensations, you will not get them.

Smart Rabbit eliminates this barrier. You do not need to know these concepts. The system builds the expert meta-prompt for you.

Visual comparison of results

Here is what each approach produces:

Conclusion: the future is not "which AI" but "who builds the prompt"

ChatGPT, Gemini and Mistral all provide correct and safe programs. A beginner can train with these programs without risk.

However, the quality of the result entirely depends on your ability to write a good prompt.

Smart Rabbit revolution is not using "a better AI". It is eliminating the need to know how to prompt.

You fill out a simple form. The expert architecture transforms your profile into a 3000-line meta-prompt. Claude 4.5 generates an optimized program.

You get the results of a prompt engineering expert without having to become one.

This is precisely the architecture you can test for free on Smart Rabbit Fitness — a simple form that hides 3000 lines of fitness logic designed by Jacques Chauvin, certified coach and 4th at the WNBF World Championship. 100% free with Claude free plan. For users who want automatic generation without copy-paste, Lazy Rabbit is available at €9.90.

Test conducted on December 19, 2024. Identical prompt sent to ChatGPT (free), Gemini (thinking mode free), and Mistral (free). Smart Rabbit program generated for free via web form with Claude Sonnet 4.5. Paid version Lazy Rabbit (€9.90) available for automatic generation without copy-paste.