How to Write Prompts That Actually Work

Writing prompts isn’t magic. It’s communication — and communication has rules.

Every time you get a bad result from AI, it almost always comes down to one of three things: too little context, an unclear goal, or a poorly structured request. This guide walks you through how to write prompts that work — with concrete rules and examples.

What Makes a Good Prompt

A good prompt is one where, after reading it, the AI knows exactly:

• What to do
• Who it’s doing it for
• What format the output should be in
• What you don’t want

A bad prompt leaves too much to be inferred. The AI fills in the gaps on its own — and not always the way you’d like.

Rule 1: Start With a Role or Context

The first sentence of a prompt sets the entire tone. Instead of jumping straight to the task, tell the AI who it is or what context it’s working in.

❌ Bad: Write me a customer email.

✅ Better: You're an experienced customer success manager at a B2B SaaS company. Write a polite but direct email to a customer who...

The role gives AI context and shapes word choice, tone, and level of expertise.

Rule 2: Be Specific About Requirements

Vague request = vague output. The more specific, the better.

❌ Bad: Write me a short text about Make.com.

✅ Better: Write an opening paragraph (max 3 sentences) for a blog post about Make.com iterators. Target audience: intermediate Make users who know what an array is but have never used the Iterator module. Tone: direct, practical, no marketing jargon.

Rule 3: Specify the Output Format

AI doesn’t know how you want the result structured unless you say so. Always specify:

• Length (word count, paragraphs, items)
• Structure (list, table, prose, JSON)
• Language register (formal/informal)

Examples:

• Output as a JSON object with keys: title, description, tags
• Respond in 3 bullet points, max 2 sentences each
• No intro, get straight to the point

Rule 4: Examples Work Better Than Descriptions

Instead of a long explanation of what you want, show an example.

Write a headline in this style: "How to Iterate Arrays in Make.com Step by Step". That means: keyword first, action verb, concise value promise.

One good example says more than three paragraphs of instruction.

Rule 5: Say What You Don’t Want

Negative instructions are just as important as positive ones. AI tends to add things you might not want — intros, apologies, repeating back your request.

• Don't add an intro or conclusion
• Avoid words like: comprehensive, robust, synergy
• Don't tell me what you're going to do, just do it

Metaprompting: Let AI Write the Prompt

For complex tasks — like recurring workflows or specialized instructions — it’s worth using metaprompting.

Metaprompting means: ask AI to help you write a better prompt.

Request: I need a prompt for an AI that will write weekly summaries of my meetings from notes. Help me write detailed instructions. Ask me everything you need to know.

The AI will ask about key things: note format, summary length, audience, tone. The resulting prompt will be far better than anything you’d write from scratch.

Checklist Before Sending a Prompt

Before you hit Send, run through this quickly:

• Does the prompt include a role or context?
• Is the request specific (length, format, audience)?
• Did I include what I don’t want?
• Do I have an example or sample output?
• Is the prompt in natural language (not telegram style)?

One well-written prompt saves three rounds of revisions.

Summary

How to write prompts well isn’t about special techniques. It’s about clear communication. Context, specificity, format, examples, and negative instructions — these are the tools you always have available.

Start with the nearest task where AI gave you a bad result. Add context, specify the format, show an example. The output will be different.