Prompt Best Practices

The Goal-Constraints-Format Framework

When you sit down to write a prompt, it helps to have a simple formula to follow. The Goal-Constraints-Format framework gives you a reliable structure for any prompt, no matter how simple or complex.

Goal: What Do You Want?

Start every prompt by clearly stating what you want the AI to do. This should be the very first sentence.

• “Write a product description for our new running shoes.”
• “Summarize this meeting transcript into action items.”
• “Create a weekly meal plan for a family of four.”

Constraints: What Are the Rules?

After stating your goal, add any limits, rules, or requirements. Constraints narrow down the infinite possibilities into something useful.

• “Keep it under 100 words.”
• “Use only vegetarian recipes.”
• “Do not mention competitor brands.”
• “The audience is teenagers aged 14-17.”
• “Write in Spanish.”

Format: How Should It Look?

Finally, tell the AI how to structure the output. Do you want a list? A table? A paragraph? An email template?

• “Format the result as a numbered list.”
• “Present the data in a table with three columns: Task, Owner, Deadline.”
• “Write the response as a professional email with a subject line.”

Putting It All Together

Here is a prompt that uses all three elements:

Goal: Write a social media caption promoting our new coffee blend.

Constraints: Keep it between 20-40 words. Do not use emojis. The tone should be sophisticated and warm. Target audience is adults aged 25-45 who appreciate artisanal products.

Format: Provide three caption options, each on its own line, numbered 1 through 3.

Compare that to simply writing “Write a social media caption for coffee.” The difference in output quality will be dramatic.

Reverse Prompting: Let the AI Ask YOU Questions

One of the most underused techniques in prompt engineering is reverse prompting. Instead of trying to write the perfect prompt yourself, you ask the AI to interview you first.

Here is how it works:

1

Give the AI a goal and ask it to ask you questions

Start with something like: “I want to create a business plan for a small bakery in Asuncion. Before you start writing, ask me 10 questions that will help you write a better plan.”

2

Answer the AI's questions

The AI will ask things like: “What is your target customer?”, “What is your monthly budget?”, “Do you plan to offer delivery?” Answer each question honestly.

3

Let the AI produce the result

After you answer all the questions, say: “Great. Now use my answers to write the business plan.” The result will be far more personalized and accurate than anything you could have prompted for in a single message.

Reverse prompting is powerful because it solves the biggest problem beginners face: you don’t know what you don’t know. You might not realize that a business plan needs a competitive analysis section or a break-even timeline. But the AI knows these elements are important and will ask about them.

Tip

You can use reverse prompting with any task. Try starting your next prompt with: “I want to [goal]. Before you begin, ask me the questions you need answered to do the best possible job.”

Being Specific vs. Being Vague

Specificity is the single biggest factor that determines whether your prompt produces something useful or something generic. Here are direct comparisons to train your instinct:

Vague vs. Specific: Email Writing

Vague: “Write an email to my boss.”

Specific: “Write a professional email to my manager requesting two days off next Friday and Monday for a family event. Keep the tone respectful but not overly formal. Mention that I will complete all pending tasks before leaving and that my colleague Maria will cover urgent items. Keep it under 150 words.”

Vague vs. Specific: Research

Vague: “Tell me about marketing.”

Specific: “Explain three low-budget marketing strategies that a small restaurant in a city of 500,000 people could use to attract more customers during weekday lunches. Focus on strategies that do not require technical skills or a large social media following.”

Vague vs. Specific: Creative Writing

Vague: “Write a story.”

Specific: “Write a 300-word short story about a grandmother teaching her granddaughter to make chipa (a traditional Paraguayan bread). Set the story in a small kitchen on a rainy afternoon. The tone should be warm and nostalgic. End the story with the granddaughter successfully making her first batch on her own.”

The pattern is clear: specific prompts produce specific, useful results. Vague prompts produce generic results that require multiple rounds of revision.

Few-Shot Prompting: Teaching by Example

Few-shot prompting means giving the AI one or more examples of what you want before asking it to produce something new. The AI learns the pattern from your examples and applies it to the new task.

Zero-shot (no example)

“Write a product tagline for a natural soap brand.”

The AI will produce something generic because it has no reference for your style.

One-shot (one example)

“Here is an example of a product tagline we like: ‘Pure ingredients. Real results. Your skin will thank you.’ Now write a similar tagline for our new shampoo line. Match the style: short sentences, natural-sounding, and focused on product benefits.”

The AI now has a concrete example to match.

Few-shot (multiple examples)

“Here are three taglines that match our brand voice:

1. ‘Pure ingredients. Real results. Your skin will thank you.’
2. ‘Nature’s best, bottled for your bathroom.’
3. ‘No chemicals. No compromises. Just clean.’

Now write three taglines for our new candle line in the same style.”

With multiple examples, the AI can identify the common patterns: short sentences, natural themes, simple language, benefit-focused. The results will be much more consistent with your brand.

Note

Few-shot prompting is especially useful when you need the AI to match a specific writing style, formatting convention, or brand voice. The more examples you provide, the more accurately the AI will match the pattern.

Testing and Iterating on Prompts

Writing a great prompt is rarely a one-shot effort. The best results come from iterating: writing a prompt, reviewing the output, identifying what could be better, and refining the prompt.

Here is a practical iteration process:

1

Write your first draft prompt

Start with the Goal-Constraints-Format framework. Do not try to make it perfect.

2

Run it and review the output

Read the AI’s response critically. Ask yourself: Is this what I wanted? What is missing? What is wrong?

3

Identify the gap

Find the specific difference between what you got and what you wanted. Was it too long? Wrong tone? Missing information? Wrong format?

4

Add the fix to your prompt

Do not start from scratch. Add a new constraint or clarification to your existing prompt to address the gap.

5

Run it again and compare

Run the updated prompt and compare the new output to the previous one. Did it improve? Is there still a gap?

Most prompts reach a “good enough” state after two or three iterations. Save your final version; it becomes a template you can reuse.

Common Mistakes Beginners Make

Here are the most frequent errors new prompt writers make, along with how to fix them.

Mistake 1: Being too polite at the expense of clarity

Problem: “Hi! I was wondering if you could maybe possibly help me write something? If it’s not too much trouble, could you perhaps write a little paragraph about dogs? Only if you want to, of course!”

Fix: “Write a 100-word paragraph about the health benefits of owning a dog. Target audience is families considering pet adoption.”

The AI does not have feelings to hurt. Politeness is fine, but not when it obscures what you actually want. Be direct.

Mistake 2: Asking for everything at once

Problem: “Write me a full business plan, marketing strategy, financial projections, logo design brief, website copy, and social media calendar for a new bakery.”

Fix: Break it into separate prompts. Start with the business plan. Then the marketing strategy. Then the financial projections. Each prompt can reference the previous output.

Mistake 3: Not specifying the audience

Problem: “Explain machine learning.”

The AI does not know if you are a PhD researcher or a 10-year-old. The explanation it gives could be at completely the wrong level.

Fix: “Explain machine learning to a high school student who has never taken a computer science class. Use everyday analogies and avoid technical jargon.”

Mistake 4: Accepting the first output without iterating

Many beginners type a prompt, get a response, and either accept it or give up. The real power comes from reading the output, giving feedback, and refining. Two or three rounds of iteration can transform a mediocre output into an excellent one.

Mistake 5: Forgetting to specify what NOT to do

Sometimes telling the AI what to avoid is just as important as telling it what to do.

Without exclusion: “Write a cover letter for a marketing position.”

With exclusion: “Write a cover letter for a marketing position. Do not use cliches like ‘passionate self-starter’ or ‘think outside the box.’ Do not exceed one page. Do not mention salary expectations.”

Quick Reference Checklist

Before sending a prompt, run through this checklist:

• Did I state my goal clearly in the first sentence?
• Did I specify who the audience is?
• Did I set constraints (length, tone, language, topics to avoid)?
• Did I specify the output format?
• Did I provide an example if the style matters?
• Did I mention what NOT to do, if relevant?

You do not need to check every box for every prompt. But the more boxes you check, the better your results will be.

Key Takeaways

• Use the Goal-Constraints-Format framework as your starting template for any prompt.
• Reverse prompting lets the AI ask you questions first, which produces better results with less effort.
• Few-shot prompting (providing examples) is the best way to get the AI to match a specific style or format.
• Always iterate: write, review, refine, repeat. Two or three rounds make a big difference.
• The most common beginner mistakes are being too vague, not specifying the audience, and accepting the first output without refinement.