Prompt Engineering for Fitness Coaches

Prompt engineering sounds technical. For a fitness coach, it's a way to think about communication: the clearer you brief the AI, the more useful the output. The same skill that makes you a good coach — explaining what you want with the right context — makes you good at this.

What You'll Learn

• The five-element prompt that solves 90% of coaching tasks
• Few-shot prompting using your own samples
• Chain-of-thought prompts for reasoning-heavy outputs
• Iteration patterns that beat starting over

The Five-Element Prompt

Almost every effective coaching prompt has five elements. Memorize these.

1. Role — who the AI is acting as
2. Context — what the situation is, who the client is
3. Task — what to produce
4. Format — how to present it
5. Constraints — what to avoid, what limits to respect

Side-by-Side Example

Without the five elements:

Write a meal plan for fat loss.

You'll get something generic — 1,800 calories, three meals, made-up portions, possibly with hype.

With the five elements:

Role: You are a fitness coach providing general healthy-eating guidance, not a registered dietitian.

Context: My client is a 38-year-old woman, 5 years training experience, 2 kids, 5'6" 75kg, fat-loss goal. She cooks at intermediate level, omnivore, no allergies. Lives in the UK.

Task: Generate 3 sample days of meals at 1,800 kcal and 140g protein per day.

Format: Per day, 4 meals with food in grams + macros per meal + 1-line prep note. End with a single grouped grocery list.

Constraints: No medical claims. No supplements. Use UK supermarket ingredients. No "magic" foods.

The output is usable in 2 minutes of editing.

Few-Shot Prompting With Your Own Samples

Few-shot prompting means giving AI examples of the output you want, then asking it to produce more.

It's the single biggest lever for matching your voice.

Example: Caption Voice Cloning

Below are 3 captions I've written for my Instagram. I want you to write 3 more in exactly this voice and style.

Notice: short paragraphs, dry humor, plain English, no emojis, ends with a CTA that's specific not generic.

--- Sample 1 --- [paste your caption]

--- Sample 2 --- [paste your caption]

--- Sample 3 --- [paste your caption]

Now write 3 new captions on the topic: [topic]

The output blends in with your real captions. Try this once and you'll never go back to single-shot prompting for content.

Example: Check-in Reply Voice

Same pattern. Paste 3 of your past check-in replies. Tell AI: "Match this voice." Provide a new client message. The reply lands closer to your style than any single instruction can produce.

Chain-of-Thought Prompting for Reasoning Tasks

When the task involves judgment ("which exercise should we substitute and why?"), explicitly asking AI to reason step-by-step improves answers significantly.

Prompt Template

Think through this step by step before answering.

First, list the constraints in the situation. Second, identify which constraint is most limiting. Third, list 3 candidate solutions and the trade-offs of each. Fourth, recommend the best option and explain the reasoning in 2 sentences.

Question: [your question]

Begin.

Real Example

Think through this step by step before answering.

A client of mine has 3 weeks until a wedding. Goal: look as lean as possible without crashing energy. Current stats: 71kg, 14% bf estimate, 3 weeks of intermediate-level training under her belt.

Step 1: List the constraints. Step 2: Identify the most limiting constraint. Step 3: List 3 candidate approaches and their trade-offs. Step 4: Recommend and reason.

You'll get a much more thoughtful answer than asking "what should she do for the wedding."

Iteration Patterns

First drafts are first drafts. Pros iterate.

Pattern 1: Edit, Don't Restart

When AI output is 80% right, don't rewrite the prompt. Reply with the edit:

Replace day 1's bench with a chest-supported variation. Drop the leg extension finisher and add 3 sets of seated calf raises instead.

AI applies the diff. Faster than re-prompting.

Pattern 2: "Make This More X" / "Less Y"

Make the tone less hype-y, more grounded.

Make the answer 30% shorter.

Make the language more plain-English. No words like "transformative" or "elevated."

These directional prompts shape output without re-specifying the whole task.

Pattern 3: Critique Then Revise

Before revising, critique the draft above as if you were a senior coach reviewing my work. List 3 weaknesses. Then rewrite, addressing them.

Gets you a better second draft than just asking for one.

Pattern 4: Generate Variants

Generate 5 variations of the caption above. Each one with a different opening hook style: question, statistic, contrarian claim, story moment, command.

You pick the strongest. Or combine elements from two.

Negative Prompting — What to Exclude

AI tools sometimes default to language you don't want. Tell them what to avoid.

Example: Coaching Voice Constraints

Do not use any of these phrases: "transform your body", "unlock your potential", "level up", "game-changer", "here's the thing", "spoiler alert", "real talk".

Do not promise specific outcomes (e.g., "you'll lose 5kg in 4 weeks").

Do not use exclamation points. Do not use emojis.

Avoid clinical or medical-sounding advice. Stay within general fitness coaching scope.

This banlist alone improves output noticeably.

Structured Output Prompts

When you need a specific format (table, JSON, CSV), say so explicitly.

Example: Markdown Table

Output as a markdown table with columns: Exercise | Sets | Reps | Load | Rest | Cue. No extra prose around it. No introduction or conclusion. Just the table.

Example: JSON for Automation

When feeding AI output into a Zapier or Make automation:

Output JSON only, no prose. Schema:

{
  "client_summary": "string, 50 words max",
  "primary_goal": "string",
  "constraints": ["array of strings"],
  "week_1_priorities": ["array of strings, max 4 items"]
}

Now your automation can parse the response cleanly.

Example: CSV for Spreadsheet Import

Output as CSV with headers: name, day, exercise, sets, reps, load_RPE, rest_sec, cue. No quotes around fields unless the field contains a comma. No introduction or conclusion text.

Paste straight into Google Sheets.

Role-Play Prompts for Hard Conversations

A useful, often-overlooked use: rehearsing difficult coach-client conversations.

Example: Pricing Conversation

Role-play with me. You're a prospective coaching client. I quoted you £350/month and you said "that's too expensive." Push back on me realistically — be skeptical, ask for justification, mention you saw a coach for £150 elsewhere. I'll respond, you continue. Don't break character.

You practice. AI plays the foil. By the third round, you've drafted lines you'll use in real calls.

Meta-Prompts: Asking AI to Improve Your Prompts

When a prompt isn't working, ask the AI itself to fix it.

The prompt below isn't producing the output I need. Tell me what's missing or unclear, and rewrite it to produce a better result. Don't write the output itself yet — just rewrite the prompt.

[paste your prompt]

You'll often get a sharper, better-structured version than your own.

A Coach's Prompt Checklist

Before sending any non-trivial prompt, scan for:

• Role specified
• Client context (or "use this saved snapshot: [paste]")
• Task in one clear sentence
• Format spec (length, structure)
• Constraints (banlist, scope, tone)
• Few-shot examples if voice matters
• Permission to ask clarifying questions

Most prompts that fail are missing 2-3 of these.

Building a Personal Prompt Library

Save your best prompts. Use a folder in Notion, Apple Notes, Raycast snippets, or TextExpander.

Suggested structure:

• Programming: program drafter, deload, sub builder, beginner block, hotel-week conversion
• Nutrition: sample days, halal/vegetarian/budget variants, habit-focus ladder
• Communication: standard check-in, rough-week check-in, big-win check-in, escalation flag
• Marketing: hook sprint, caption batch, carousel, YouTube description, sales page
• Operations: intake-to-brief, welcome line, day-7 pulse summarizer

Within 2-3 months, you'll have a library that handles 90% of repeated work in seconds.

Final Principles

• Specificity is leverage; vague prompts produce vague output
• Voice samples beat voice descriptions every time
• Iteration loops are faster than restarts
• A banlist is just as important as a goal
• The coach's judgment is the part AI can never automate

Key Takeaways

• The five-element prompt (Role, Context, Task, Format, Constraints) handles 90% of cases
• Few-shot prompting with your own samples is the single biggest voice lever
• Chain-of-thought prompts produce better output for judgment tasks
• Edit, don't restart; iteration patterns save time
• Build a personal prompt library and reuse the same patterns across clients