Prompt Engineering:
Write Better AI Prompts
Most people treat AI like a search engine. This guide teaches you to treat it like a brilliant collaborator — with the right briefing, it will surprise you.
Why Prompts Matter More Than the Model
People obsess over which AI model is best. But research consistently shows that prompt quality has a bigger impact on output quality than model choice. A well-crafted prompt on a mid-range model will beat a lazy prompt on the best model — every time.
Here's the mental shift that changes everything: AI models don't read your prompt — they predict what should come next after it. Every word you include shifts the probability distribution of what gets generated.
How AI Actually Reads Your Prompt
When you send a prompt, the AI processes it token by token and predicts what should follow. It has learned patterns across billions of documents — your prompt is a context-setter that makes certain response patterns more probable.
Three practical implications:
Context is cumulative
Every sentence adds to the context window. Relevant background you provide is rarely wasted.
Order matters
Information at the start and end gets more attention. Put the most important constraint last.
Role shifts probability
"Senior doctor" vs. "medical student" actually changes vocabulary, depth, and caution level.
The Golden Rule
Be as specific as you'd be briefing a new employee on their first day.
Imagine hiring a brilliant freelancer who's never worked with you. They're smart and eager — but they don't know your business, audience, or standards. The brief you'd write for them is the prompt you should write for AI.
That brief would include:
- What they're making — the output format and length
- Who it's for — the audience's knowledge level and goals
- What success looks like — criteria for judging the output
- What to avoid — common mistakes or things you don't want
- Background context — why this matters, what it's for
The 6 Core Elements of a Great Prompt
Not every prompt needs all six — but knowing all six lets you diagnose why a prompt isn't working.
Role / Persona
Tell the AI who it is. This shapes vocabulary, depth, tone, and assumptions.
Task / Action
State exactly what you want. Use strong verbs: write, analyze, compare, summarize, rewrite, list.
Context / Background
The "why" and the "who." What situation are you in? What does the AI need to know?
Format / Output
How should the response look? Length, structure, tone, what to include or exclude.
Examples / Samples
Show the AI what "good" looks like. This is the most underused technique.
Constraints / Guardrails
What the AI should NOT do. Boundaries prevent the most common failure modes.
5 Frameworks Every Prompt Engineer Uses
Frameworks are checklists that ensure you haven't forgotten a key element. Pick the one that matches your use case.
Provide 2–5 examples of input → output pairs before your actual request. The AI learns the pattern from examples rather than instructions. The single most reliable technique for getting consistent output.
Input: [example 2] → Output: [desired output 2]
Now do the same for: [your actual input]
Advanced Techniques
1. Chain of Thought — Make the AI Show Its Work
Adding "think step by step" to a prompt increases accuracy on reasoning tasks by 20–40%. The AI is forced to reason before committing to an answer.
Think through this step by step. Show each calculation before giving the final answer.
2. Tree of Thoughts — Explore Multiple Paths
Instead of one answer, ask the AI to explore three approaches, evaluate each, and recommend the best.
1. Describe in 2–3 sentences
2. List 2 strengths
3. List 2 weaknesses
4. Rate feasibility (1–10)
After all three, recommend the best and explain why.
3. Chain of Density — Compress Without Losing Information
Ask the AI to write a summary, then rewrite it at the same length but with more density. Repeat 3–4 times. Each iteration replaces vague language with specific facts, numbers, and names.
4. Multi-Persona Debate
Ask the AI to evaluate an idea from multiple perspectives simultaneously — one of the most underrated techniques for decision-making.
Give me three perspectives:
- A skeptical VC who has seen this fail before
- An excited early adopter who loves this category
- A pragmatic operator who has to build and run this
After each, give an overall verdict: go, no-go, or iterate.
The 7 Most Common Prompting Mistakes
Being too vague about the output format
If you don't specify format, the AI decides — and chooses the most verbose option.
Fix: Always end with the output format: "Respond as 5 bullet points, max 20 words each."
Asking multiple questions in one prompt
AI answers all questions but gives each one less attention. The last often gets the worst.
Fix: One prompt = one focused task. Break multi-part requests into a sequence.
Using negative instructions only
"Don't be boring" is weaker than "Be direct and concrete." The AI knows what direct looks like.
Fix: Replace negatives with positives: "Don't be vague" → "Use numbers, names, and examples."
Not giving enough context
Short prompts force the AI to make assumptions. Those assumptions are often wrong.
Fix: Include relevant background. Paste the email you're replying to, the doc you're summarizing.
Accepting the first output
The first response is rarely the best. Iteration is where quality compounds.
Fix: Follow up: "This is good but too formal — rewrite it as if explaining to a friend."
Treating all models the same
ChatGPT, Claude, and Gemini have different strengths and respond differently to the same prompt.
Fix: Tune prompts per model. Claude is great for nuance; GPT-4 for structured tasks and code.
Skipping examples for consistent output
Describing a format takes 100 words. Showing it takes 20. Examples win every time.
Fix: Show 2–3 input/output pairs before the real request.
How Prompting Differs Across AI Models
The core principles are universal, but each model has nuances worth knowing.
ChatGPT / GPT-4
- Excellent at structured tasks and precise formatting
- Strong with code generation and debugging
- Rephrase sensitive questions in a professional context
Claude
- Excellent at nuanced writing, long documents, analysis
- Responds well to "be direct" and "skip the caveats"
- Very large context window — feed it the full document
Gemini
- Strong with recent information and search-augmented tasks
- Good multimodal prompting (text + images)
- Use when recency matters; training data is more current
Quick Prompt Checklist
Before sending any important prompt, run through this list:
Skip the Manual Work
Frompting applies RACE, CO-STAR, Chain of Thought, and 57 more frameworks to your topic automatically.