How to Write Better AI Prompts (With Examples)

Prompt Engineering Is Just Clear Communication

There's an entire industry around "prompt engineering," full of courses and certifications that make simple concepts sound complicated. Here's the truth: writing good AI prompts is just clear communication. If you can write a clear email to a coworker, you can write good prompts.

That said, there are specific techniques that consistently produce better results. Here they are, with real examples you can adapt.

The Basic Framework

Every good prompt answers four questions:

1. Who should the AI act as? (Role)
2. What do you need? (Task)
3. How should the output look? (Format)
4. Why does context matter? (Background)

You don't always need all four, but including more of them produces more targeted responses.

Technique 1: Role Setting

Bad prompt: "How do I improve my website's SEO?"

Good prompt: "You are an SEO consultant who specializes in small B2B SaaS companies. My website sells project management software to teams of 10-50 people. We currently get 2,000 monthly organic visitors. Give me five specific, actionable SEO improvements ranked by expected impact."

The role focuses the AI's knowledge. Instead of generic SEO advice, you get recommendations calibrated to your situation.

Technique 2: Few-Shot Examples

Show the AI what you want by providing examples of good output.

Prompt: "I need product descriptions for an e-commerce store. Here are two examples of the style I want:

Example 1: 'The Apex Trail Runner grips loose gravel like it's pavement. We tested it on three different mountain trails and it held up for 400 miles before the tread showed wear. At $89, it's not the cheapest option, but your ankles will thank you.'

Example 2: 'This thermos keeps coffee hot for 14 hours. We know because we measured it at 6am and checked again at 8pm. Still steaming. The lid seals tight enough to throw in a backpack without worry.'

Now write a description for a waterproof Bluetooth speaker that retails for $45."

The AI mirrors the tone, length, and structure of your examples. This technique is more effective than trying to describe the style you want in words.

Technique 3: Chain of Thought

For complex reasoning tasks, ask the AI to show its work.

Prompt: "Our company is deciding whether to build a feature in-house or buy a third-party solution. Here are the details: [details]. Walk through the decision step by step. Consider cost, timeline, maintenance burden, and team capacity. Then give your recommendation with reasoning."

Asking for step-by-step reasoning reduces errors and produces more thoughtful analysis than asking for a direct answer.

Technique 4: Constraints and Guardrails

Tell the AI what boundaries to stay within.

Prompt: "Write a blog post about cloud security for a non-technical audience. Constraints: no jargon (explain any technical terms in parentheses), no longer than 800 words, use concrete examples instead of abstract concepts, do not use the words 'leverage,' 'synergy,' or 'ecosystem.'"

Constraints prevent the AI from reverting to default patterns and force more specific output.

Technique 5: Iterative Refinement

Use follow-up prompts to shape the output:

• "Good, but make it more conversational."
• "This is too long. Cut it to the five most important points."
• "The third paragraph is vague. Replace it with a specific example from e-commerce."
• "Rewrite this assuming the reader already knows Python basics."

Think of it as editing with the AI. Your first prompt is a draft. Subsequent prompts are revision notes.

Technique 6: Output Templates

Give the AI a template to fill in:

Prompt: "Analyze this competitor using the following template:

Company: [name]
Product: [one-sentence description]
Target Market: [who they sell to]
Pricing: [their pricing model]
Strengths: [3 bullet points]
Weaknesses: [3 bullet points]
Threat Level to Us: [low/medium/high with one-sentence justification]"

Templates produce structured, consistent output that's immediately useful. They also make it easy to process multiple items using the same format.

Common Prompt Mistakes

Too vague: "Help me with marketing." What kind of marketing? For what product? To what audience? Through what channel?

Too specific too soon: Don't micro-manage the AI's approach in your first prompt. Give the task and constraints, let it respond, then refine.

Multiple unrelated tasks: One prompt, one task. If you need the AI to do five different things, send five separate prompts. Cramming everything together produces mediocre results on all of them.

No success criteria: How will you know if the output is good? Define what "done" looks like before you prompt.

Model-Specific Tips

GPT-5: Responds well to systematic, structured prompts. Works best when you number your requirements.

Claude: Excels with conversational prompts. You can be more natural and less formulaic. Handles nuance and ambiguity better than most models.

Gemini: Strongest when you include images or visual context. Text-only prompts work fine but you're not using its best feature.

The differences are subtle, and a well-structured prompt works well on any model. But if you're optimizing, these tendencies can help.

Start Simple, Add Complexity

Don't try to write the perfect prompt on your first attempt. Start with a basic version, see what comes back, and add constraints and context in follow-ups. This iterative approach is faster and more effective than trying to anticipate everything upfront.