---
title: Better Prompts in 5 Minutes
description: Learn the core principles behind effective AI prompts. No fluff, no hype - just practical techniques that actually improve your results.
date: January 20, 2026
author: Robert Soares
category: prompt-engineering
---

Most people treat AI prompts like Google searches. Type a few words, hit enter, hope for the best.

That works fine for simple questions. But the moment you need something more specific, like a certain tone or format or level of detail, vague prompts fall apart. The AI guesses. And guesses tend toward generic.

The good news: fixing this takes about five minutes of learning. Not five minutes of practice, five minutes of understanding. Once you see why prompts work the way they do, better results come naturally.

## Why Vague Prompts Fail

Here's what happens when you type "write me an email" into an AI.

The model has no idea who you're writing to. Is this a colleague? A customer? Your boss? It doesn't know the tone you want. Professional? Friendly? Apologetic? It can't tell what you're trying to accomplish. Are you asking for something? Following up? Sharing news?

So it picks the middle of everything. Generic professional tone. Medium length. Safe, forgettable content.

This isn't the AI being dumb. It's the AI doing exactly what you asked: filling in all the blanks with reasonable defaults. The problem is that reasonable defaults are almost never what you actually wanted.

Research backs this up. [Studies have found](https://clearimpact.com/effective-ai-prompts/) that prompts with clear, specific instructions lead to 25% higher completion rates compared to vague phrasing. That's a meaningful gap just from word choice.

## The Core Principle: Specificity

The single most effective change you can make is being specific about what you want.

Not "write me an email" but "write a friendly follow-up email to a potential client who hasn't responded in two weeks. Keep it under 100 words. The goal is to restart the conversation without being pushy."

Same task. Completely different result.

[Anthropic's prompting guide](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-4-best-practices) puts this at the top of their best practices for a reason. Their guidance emphasizes being explicit about your desired output, including the format, length, and what "done" looks like. The more the model understands about your goal, the less it has to guess.

This applies across every type of prompt:

**Vague:** "Summarize this document"
**Specific:** "Summarize this document in 3 bullet points, focusing on the budget implications"

**Vague:** "Help me with this code"
**Specific:** "This function is throwing an error on line 12. I think it's a type mismatch but I'm not sure. Can you identify the issue and suggest a fix?"

**Vague:** "Make this better"
**Specific:** "This paragraph is too formal for our company blog. Rewrite it in a conversational tone, like you're explaining it to a friend"

In each case, the specific version gives the AI a clear target. You're not asking it to read your mind.

## Show, Don't Just Tell

Sometimes explaining what you want isn't enough. Showing works better.

This is called few-shot prompting, a term that just means "give examples." If you want a certain format or style, include a sample of what that looks like.

Say you need product descriptions for an e-commerce site. Instead of explaining your preferred style, show it:

> "Write product descriptions like this example:
>
> **Ceramic Travel Mug**
> Keeps coffee hot for 4 hours. Fits in standard cup holders. Dishwasher safe. Available in 6 colors.
>
> Now write a similar description for a stainless steel water bottle."

The AI now has a template. It can match the length, the sentence structure, the level of detail. No ambiguity about what you're looking for.

[Research on prompt engineering](https://www.promptingguide.ai/) consistently shows that 3-5 diverse examples significantly improve output quality. The key word is diverse. Don't give five examples that all look the same. Show the range of what you want.

## Think Step by Step

Here's a technique that sounds almost too simple: ask the AI to think through the problem step by step.

It works because of how these models process information. When you ask for a direct answer to a complex question, the model sometimes jumps to a conclusion without working through the logic. But when you prompt it to reason out loud, each step informs the next.

[Google's research team first documented this in 2022](https://research.google/blog/language-models-perform-reasoning-via-chain-of-thought/). They found that adding "let's think step by step" to math problems improved GPT-3's accuracy from 18% to 58% on certain benchmarks. That's a huge jump from five extra words.

You don't need to use that exact phrase. What matters is encouraging the model to show its work:

- "Walk me through your reasoning before giving the final answer"
- "Break this problem into steps and solve each one"
- "Think about this carefully, then explain your conclusion"

This technique is especially useful for anything requiring logic, analysis, or multi-step thinking. Code debugging. Data analysis. Strategic decisions. Anywhere you'd want a human to explain their reasoning rather than just give you an answer.

## Give Context, Not Just Instructions

Most people focus on the instruction part of a prompt. But the context around that instruction matters just as much.

Consider the difference between:

> "Write a cover letter for a software engineering position"

And:

> "I'm applying to a startup that emphasizes creativity and shipping fast. I have 5 years of experience, mostly in Python. They specifically mentioned wanting someone who can wear multiple hats. Write a cover letter that highlights my adaptability and bias toward action."

Same basic task. But the second prompt gives the AI context to make smarter decisions. It knows what to emphasize, what tone to use, what kind of company you're writing to.

[Anthropic's documentation](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-4-best-practices) specifically recommends providing context or motivation behind your instructions. When the AI understands *why* you want something, it can generalize from that understanding. You don't have to specify every detail because the model can infer what fits the situation.

This is particularly useful when your requirements are nuanced. Instead of listing dozens of rules, explain the underlying goal. The AI can figure out the implications.

## The Iteration Mindset

Here's something that surprises a lot of people: your first prompt is rarely your best one.

Even experienced prompt engineers iterate. They try something, see what works and what doesn't, and refine. This isn't a sign you're doing it wrong. It's how the process works.

[MIT Sloan research](https://mitsloan.mit.edu/ideas-made-to-matter/study-generative-ai-results-depend-user-prompts-much-models) found that half of the performance gains people saw from upgrading to a more advanced AI model actually came from improving their prompts, not from the model itself. Users naturally got better at prompting through experimentation.

What makes iteration effective:

**Notice what's wrong.** Generic tone? Wrong format? Missing key details? Identify the specific gap.

**Fix that specific issue.** Don't rewrite everything. Add the instruction that addresses the problem.

**Keep what worked.** If you got the tone right but the length wrong, keep your tone instructions and adjust the length.

Think of prompts as living documents. The first version is a starting point, not a final answer.

## When to Give Examples vs. When to Explain

You have two main tools for guiding AI output: examples and explanations. Knowing when to use each makes a real difference.

**Use examples when:**
- Format matters (tables, bullet points, specific structures)
- Style is hard to describe but easy to show
- You want consistency across multiple outputs
- The task is creative and you have a target in mind

**Use explanations when:**
- The reasoning behind your request matters
- You need flexibility within constraints
- The task involves analysis or judgment
- You want the AI to generalize to new situations

Often the best prompts combine both. Show an example of what you want, then explain why that example works. The model gets both the pattern and the principle.

## Common Mistakes to Avoid

A few patterns consistently lead to worse results.

**Being polite to a fault.** "Could you perhaps maybe consider writing..." adds words without adding clarity. Just say what you need. AI models don't have feelings to hurt.

**Asking for too much at once.** "Write a blog post and create 10 social media captions and suggest three headlines" forces the model to split attention across tasks. Break complex requests into separate prompts.

**Contradicting yourself.** "Be comprehensive but keep it short" gives the AI conflicting goals. Decide which matters more and say that clearly.

**Assuming context.** The AI doesn't know about your previous conversation, your company, or your preferences unless you tell it. Include the context you'd give a new colleague.

**Over-specifying.** Yes, specificity helps. But you can also constrain so tightly that you leave no room for the AI to add value. Give direction, not a script.

## Model Differences Worth Knowing

Not all AI models respond the same way to prompts.

Newer reasoning models, like OpenAI's o1 series, actually work better with less hand-holding. [Research indicates](https://www.news.aakashg.com/p/prompt-engineering) that these models perform worse when you overload them with examples. They're designed to reason through problems, so giving them room to think outperforms forcing a specific approach.

Meanwhile, standard GPT models and Claude still benefit from explicit examples and step-by-step instructions. They follow directions more literally.

The practical takeaway: start with clear, specific prompts for any model. But if you're using a reasoning-focused model for complex tasks, try giving it the problem and letting it figure out the approach. You might get better results than micromanaging the process.

## Putting It Together

Here's a template you can adapt for most situations:

> **Context:** [Who you are, what you're working on, any relevant background]
>
> **Task:** [What you need the AI to do, specifically]
>
> **Format:** [How you want the output structured]
>
> **Constraints:** [Length limits, things to avoid, tone requirements]
>
> **Example (if helpful):** [A sample of what good output looks like]

You don't need all five elements every time. Simple tasks might just need a clear instruction. But when you're not getting the results you want, check which element is missing.

Let's see this in practice:

> "I'm a product manager writing release notes for our dev team. The update includes three bug fixes and one new feature (dark mode support).
>
> Write release notes that are technical enough for developers but also readable by non-technical stakeholders. Use bullet points for the changes. Keep the total under 150 words. Start with the new feature since that's what people are most excited about."

That prompt includes context (product manager, dev team audience), task (write release notes), format (bullet points), and constraints (under 150 words, feature first). The AI has everything it needs.

## The Five-Minute Version

If you remember nothing else, remember this:

1. Be specific about what you want, including format, length, and tone
2. Give context about why you need it
3. Show examples when format matters
4. Ask the model to think step by step for complex problems
5. Iterate when the first result isn't right

That's it. These aren't advanced techniques. They're the fundamentals that actually work.

The best prompt isn't long or complex. It's the one that gives the AI exactly enough information to understand what you need. Start there, and the rest follows.