How to Write Better AI Prompts: 10 Techniques That Actually Work

Most people use AI tools at 20% of their potential because they write weak prompts. Here are 10 techniques — with before/after examples — that will immediately improve your results.

Why Prompts Matter

Large language models are prediction machines. They generate the most statistically likely continuation of your text. A precise prompt narrows the probability space and guides the model toward what you actually want.

The 10 Techniques

1. Assign a Role

Tell the model who it is. This activates relevant "knowledge patterns" in the model's weights.

Before: Write a product description for my ergonomic chair.

After: You are a senior copywriter at a luxury furniture brand. Write a 150-word product description for an ergonomic office chair that emphasizes health benefits and quality craftsmanship. Tone: sophisticated but approachable.

---

2. Specify Format Explicitly

Don't assume the model knows what format you want. Tell it.

Before: Summarize this article.

After: Summarize this article in exactly 3 bullet points, each 1–2 sentences. Use bold text for key terms.

---

3. Give Examples (Few-Shot Prompting)

Show, don't just tell. Provide 1–3 examples of what you want.

Write subject lines in this style:

Example 1: "Stop losing clients to competitors (do this instead)"
Example 2: "The 5-minute fix for your highest-bounce page"

Now write 5 subject lines for a webinar about email marketing.

---

4. Set Constraints

Length, tone, reading level, what to exclude — be explicit.

After: Write a blog intro (80 words max) about AI writing tools. Target a non-technical audience. Do NOT mention ChatGPT or OpenAI.

---

5. Use Chain-of-Thought for Complex Tasks

For reasoning tasks, ask the model to "think step by step" before giving the final answer.

A company has $50,000 to spend on marketing. They're choosing between Google Ads ($0.80 CPC, 2% conversion), Facebook Ads ($0.40 CPC, 1.5% conversion), and SEO (upfront $15,000, then $2,000/month, 500 new visitors/month in 6 months).

Think step by step, then recommend the best allocation for maximum leads in 12 months.

---

6. Ask for Multiple Variations

Don't settle for one output. Ask for variations and pick the best.

After: Write 5 different headlines for this blog post. Vary the style: one curiosity-driven, one list-based, one how-to, one bold claim, one question.

---

7. Pass Context Upfront

Give the model relevant background before asking the question.

Before: Is this email too aggressive?

After: Context: I'm following up with a prospect who went dark after a demo 3 weeks ago. We're a B2B SaaS company. The prospect is a VP at a 500-person company. Here's my email: [email text]. Is the tone too aggressive? How would you revise it?

---

8. Iterate, Don't Restart

Use follow-up messages to refine rather than starting over.

[Initial prompt] → [AI output] →

"Good, but make the intro 30% shorter and add a statistic in paragraph 2."

---

9. Ask the Model to Ask You Questions

For complex tasks, have the model interview you first.

I want to write a case study about a customer success story. Before you write anything, ask me the 10 questions you need answered to write a compelling case study.

---

10. Specify What NOT to Do

Negative constraints are often more effective than positive ones.

After: Write a product FAQ. Do NOT use jargon. Do NOT end answers with "feel free to contact us." Do NOT exceed 60 words per answer.

---

Quick Reference Card

Technique	When to Use
Assign a role	Any creative or expert task
Specify format	Structured outputs (lists, tables, reports)
Few-shot examples	When style/tone is hard to describe
Set constraints	When output quality varies
Chain-of-thought	Math, logic, multi-step decisions
Multiple variations	Copywriting, headlines, ideas
Pass context upfront	Business or domain-specific tasks
Iterate in-thread	Polishing drafts
Ask for questions	Complex, multi-variable projects
Negative constraints	When models ignore positive instructions

---

Prompt engineering isn't a dark art — it's just communication. The clearer you are about what you want, who you are, what the output should look like, and what constraints apply, the better your results will be.

Start with one technique per session until they become instinct.