Prompt EngineeringJanuary 26, 2026

How to Write Better ChatGPT Prompts (With Examples)

Practical techniques for improving ChatGPT prompts with before/after examples — showing exactly what to change and why for dramatically better output.

Most people write prompts the way they'd text a friend — short, vague, and full of assumed context. That works for casual conversation. It doesn't work for getting useful output from AI.

This guide shows you exactly how to improve your prompts, with before/after examples for every technique.

The Core Problem: Vague In, Vague Out

AI models are instruction followers. They give you exactly what you ask for — which means if you ask vaguely, you get vague output.

Before: "Write me a marketing email."

After: "Write a product launch email for our project management tool targeting engineering managers at companies with 50-200 employees. Tone: professional but not stiff. Length: under 200 words. Include a specific pain point (context switching between too many tools), the key benefit (all-in-one workspace), and a CTA to start a free trial. Subject line should create curiosity, not urgency."

The second prompt takes 30 seconds longer to write and saves 20 minutes of back-and-forth revision.

Technique 1: Specify the Role

Telling ChatGPT who it's acting as frames every aspect of the response — vocabulary, depth, perspective, and recommendations.

Before: "How should I price my SaaS product?"

After: "You're a SaaS pricing consultant who has helped 50+ B2B companies optimize their pricing. I have a project management tool for engineering teams. Current pricing: $15/user/month. 200 paying customers, 40% annual churn. Competitors charge $10-25/user/month. How should I evaluate whether to raise prices, change my pricing model, or add tiers?"

Why it works: The role tells the AI what expertise to draw on. The context gives it the specific situation to apply that expertise to. Together they produce a tailored recommendation instead of a textbook answer.

Technique 2: Define the Output Format

The biggest quick win in prompt writing is telling ChatGPT exactly what format you want.

Before: "Give me ideas for blog posts about AI."

After: "Give me 10 blog post ideas about AI for marketing professionals. Format as a table with columns: Title, Target Keyword, Search Intent (informational/commercial/transactional), and Estimated Word Count. Prioritize topics with commercial intent."

More format examples:

"Format as a numbered list with one sentence per item"
"Create a markdown table with columns for..."
"Structure your response with H2 headers for each section"
"Respond in bullet points, no paragraphs"
"Give me a JSON object with these fields..."

Technique 3: Provide Examples (Few-Shot)

When you want output that matches a specific style or structure, show ChatGPT what good looks like.

Before: "Write product descriptions for my store."

After: "Write product descriptions for my online store. Here are two examples that match our brand voice:

Example 1: 'The Everyday Tote — Built for the person who carries their whole life in one bag. Waxed canvas exterior handles whatever you throw at it (literally). Interior organizer keeps your laptop, notebook, and coffee money where you can find them. 14" x 18" x 6".'

Example 2: 'The Weekend Duffle — For trips that don't need a checked bag. Water-resistant nylon, shoe compartment on the bottom, and a luggage sleeve for when you're being responsible. Fits overhead on every airline we've tested.'

Now write descriptions in this same style for these 3 products: [list products with key specs]."

Why it works: Examples communicate tone, length, structure, and style more effectively than describing them. Two examples is usually enough.

Technique 4: Set Constraints

Constraints force better output. Without them, ChatGPT defaults to verbose, generic responses.

Useful constraints to add:

Length: "Keep each section under 100 words" or "Total response under 500 words"
Audience level: "Explain for someone with no technical background" or "Assume the reader is a senior developer"
Exclusions: "Don't use jargon" or "Don't include generic advice like 'know your audience'"
Priorities: "Focus on actionable steps, not theory"
Tone: "Direct and confident, not hedging with 'it depends' or 'consider maybe'"

Before: "Explain Kubernetes to me."

After: "Explain Kubernetes in 3 paragraphs. Paragraph 1: what it is and what problem it solves (for someone who understands Docker but hasn't used orchestration). Paragraph 2: the 3 most important concepts to understand first. Paragraph 3: when you need it vs. when it's overkill. No jargon beyond Docker terminology."

Technique 5: Chain Your Prompts

Complex tasks produce better results when broken into sequential prompts.

Instead of one mega-prompt: "Write me a complete content strategy with personas, topics, calendar, and KPIs."

Chain it:

Prompt 1: "Define 3 buyer personas for [business]. Include role, pain points, content preferences, and buying triggers."

Prompt 2: "Based on these personas [paste output], generate 20 content topics. For each, note which persona it serves and where it fits in the buying journey."

Prompt 3: "Organize these topics into a 3-month content calendar. Include publishing cadence, content format, and which team member should own each piece."

Prompt 4: "Create a KPI framework for this content strategy. Include metrics for each funnel stage and monthly targets."

Each prompt builds on the previous output, producing more coherent and detailed results than a single prompt ever could.

Technique 6: Ask for Reasoning

When you need recommendations or analysis, ask ChatGPT to explain its reasoning. This produces better answers and lets you evaluate the logic.

Before: "Which marketing channel should I focus on?"

After: "Given my business context [describe], recommend which 2 marketing channels to prioritize this quarter. For each recommendation: explain your reasoning, identify the key assumption behind the recommendation, describe what would change your recommendation, and estimate the timeline to see results."

Why it works: Asking for reasoning activates chain-of-thought processing. The model produces more nuanced output when it has to justify its answer.

Technique 7: Use Anti-Patterns

Tell ChatGPT what NOT to do. This is surprisingly effective at eliminating the generic filler that makes AI content feel like AI content.

High-impact anti-patterns:

"Don't start with 'In today's digital landscape' or any similar throat-clearing"
"Don't use the phrase 'it's important to note that'"
"Don't hedge every statement with 'it depends' — give a direct answer, then note exceptions"
"Don't summarize what you just said at the end of each section"
"Don't use exclamation points"
"Don't include a generic introduction that could apply to any article on this topic"

Before/after impact:

Before: "In today's rapidly evolving digital landscape, content marketing has become more important than ever. It's crucial for businesses to understand..."

After (with anti-patterns): "Content marketing drives 3x more leads per dollar than paid advertising for B2B companies. Here's how to build a strategy that actually produces those results."

Technique 8: Iterate, Don't Regenerate

When the output is 70-80% right, don't start over. Refine it.

Effective follow-up prompts:

"Make the tone more conversational — like you're explaining this to a smart friend over coffee"
"The second section is too generic. Add specific examples using [industry/tool/scenario]"
"Shorten every paragraph to 2-3 sentences max"
"The recommendations are good but too safe. Make them more specific and opinionated"
"Reorganize this so the most actionable advice comes first"

This approach is faster and produces better results than regenerating from scratch, because you keep what's working and fix what isn't.

Quick Reference: The Prompt Improvement Checklist

Before sending a prompt, check:

Role: Have I told the AI who/what to be?
Context: Have I provided my specific situation?
Task: Is the task specific enough that there's only one way to interpret it?
Format: Have I specified the output structure?
Constraints: Have I set boundaries (length, tone, audience, exclusions)?
Examples: Would an example make the expected output clearer?

You don't need all six for every prompt. But any prompt that produces disappointing output is probably missing at least two of these elements.