Skip to content
How-To Guide

How to Write Better AI Prompts: A Beginner’s Guide to Prompt Engineering

Difficulty: Beginner
Read time: 10 min
Updated:
Best for: Anyone who uses AI assistants and wants to get consistently better results from their prompts

Prompt engineering is the skill of writing clear instructions that help AI tools like ChatGPT, Claude and Gemini give you better results. Whether you use AI for writing, coding, research or daily tasks, the way you phrase your request directly determines the quality of the response. This guide teaches you 7 practical techniques, backed by official documentation from OpenAI, Anthropic and Google, that work with any major AI assistant.

Key Takeaways

01

Be specific: vague prompts get vague answers. Tell the AI exactly what format, length and tone you want before it starts writing

02

Give context: explain who you are, what the output is for and any constraints. The more relevant background you provide, the more targeted the response

03

Use examples (few-shot prompting): showing the AI one or two examples of your desired output is more effective than describing it in words

04

Break complex tasks into steps: instead of asking for everything at once, chain multiple prompts together. Each step builds on the previous one

What Is Prompt Engineering and Why Does It Matter?

Prompt engineering is the process of writing instructions (called “prompts”) that guide AI models to produce the output you actually want. Think of it as giving directions to a talented assistant who has never worked with you before. The clearer your instructions, the better the result.

This matters because the same AI model can produce drastically different outputs depending on how you phrase your request. A vague prompt like “write something about marketing” might return a generic 200-word overview. A well-engineered prompt like “write a 500-word blog introduction about email marketing for small e-commerce businesses, using a conversational tone and including one real-world statistic” will return something you can actually use.

The techniques below are drawn directly from the official prompt engineering documentation published by OpenAI (ChatGPT), Anthropic (Claude) and Google (Gemini). They work across all three platforms.

Technique 1: Be Specific About What You Want

The single most effective improvement you can make to any prompt is being more specific. Vague instructions force the AI to guess what you mean. Specific instructions eliminate guesswork.

According to Anthropic’s official documentation: “Think of Claude as a brilliant but new employee who lacks context on your norms and workflows. The more precisely you explain what you want, the better the result.”

This principle applies equally to ChatGPT and Gemini. All three platforms perform better when you specify:

  • Format: paragraph, bullet list, table, JSON or numbered steps
  • Length: 100 words, 3 paragraphs, single sentence
  • Tone: professional, conversational, technical or casual
  • Audience: who will read this output
  • Constraints: what to include or exclude

Before and after example

Vague prompt Specific prompt
Write about SEO Write a 300-word explanation of on-page SEO for small business owners who have never done SEO before. Use a friendly tone. Include 3 actionable tips they can implement today without any technical knowledge
Help me with this code Debug this Python function that should return the sum of even numbers in a list but returns 0 for every input. Explain what the bug is and show the corrected code with comments
Summarize this article Summarize the key findings from this research paper in 5 bullet points. Each bullet should be one sentence. Focus on practical implications rather than methodology

Pro tip from Google’s documentation: You can also specify constraints: what the AI should NOT do. For example: “Do not use jargon” or “Keep each bullet under 20 words.” Constraints narrow the output and reduce the chance of getting something unusable.

Technique 2: Provide Context and Background

AI models perform significantly better when they understand the context behind your request. Without context, the AI has to make assumptions about your situation, audience and goals. Those assumptions are often wrong.

Context includes:

  • Who you are: your role, industry or expertise level
  • What the output is for: a blog post, email, presentation or internal document
  • Who the audience is: executives, beginners, developers or customers
  • Any constraints: word limits, style guides, brand tone or required facts

Example: context transforms the output

Without context:

“Write an email about a product delay.”

The AI will write a generic apology email with no personality or strategy.

With context:

“I run a small DTC skincare brand. Our best-selling moisturizer is delayed by 2 weeks because of a supplier issue. Write an email to customers who pre-ordered. Keep the tone warm and transparent. Mention that we are adding a free sample to every delayed order as a thank-you. Keep it under 150 words.”

The second prompt gives the AI everything it needs to write a response you can actually send, often without editing.

Technique 3: Use Examples (Few-Shot Prompting)

One of the most powerful techniques in prompt engineering is showing the AI what you want through examples. This approach is called few-shot prompting and all three major platforms (OpenAI, Anthropic and Google) recommend it in their official documentation.

Google’s Gemini documentation states: “We recommend always including few-shot examples in your prompts. Prompts without few-shot examples are likely to be less effective.”

Here is how it works: instead of explaining the format you want, you show it.

Example: categorizing customer feedback

Without examples (zero-shot):

“Categorize the following customer reviews as positive, negative or neutral.”

This might work, but the AI may categorize ambiguous reviews inconsistently.

With examples (few-shot):

“Categorize each customer review as positive, negative or neutral. Here are some examples:

Review: ‘The product arrived on time and works great.’ → Positive
Review: ‘It broke after two days. Waste of money.’ → Negative
Review: ‘It is okay. Nothing special.’ → Neutral

Now categorize these reviews: [paste your reviews]”

The AI now has a clear pattern to follow, which produces more consistent and accurate results.

How many examples? Start with 2-3 examples. Google’s documentation notes that models can often pick up patterns from just a few examples. Too many examples can cause the model to overfit (copying your examples too literally) instead of generalizing the pattern.

Technique 4: Assign a Role (Persona Prompting)

Telling the AI to “act as” a specific role changes how it approaches your task. This technique is called role prompting or persona prompting and it is recommended by all three platforms.

Anthropic’s documentation describes it this way: “Setting a role in the system prompt focuses Claude’s behavior and tone for your use case. Even a single sentence makes a difference.”

When you assign a role, the AI adjusts its vocabulary, depth of explanation, assumptions about your knowledge level and communication style.

Effective role prompts

Task Role prompt
Writing marketing copy “You are a senior copywriter at a digital marketing agency with 10 years of experience writing conversion-focused landing pages.”
Debugging code “You are a senior Python developer. When explaining bugs, provide the root cause, the fix and a brief explanation of why the fix works.”
Learning a new topic “You are a patient tutor explaining machine learning to a college freshman with no coding experience. Use analogies and avoid technical jargon.”
Legal review “You are a contract lawyer reviewing a freelance agreement. Flag any clauses that are unusual or potentially unfavorable to the freelancer.”

Important note: role prompting is a starting point, not a replacement for expertise. The AI is simulating the role, not actually a lawyer, doctor or financial advisor. Always verify critical outputs with a qualified professional.

Technique 5: Break Complex Tasks into Steps (Prompt Chaining)

When you have a complex task, do not try to get the AI to do everything in one prompt. Instead, break it into smaller steps where each prompt builds on the output of the previous one. This technique is called prompt chaining.

AI models handle focused tasks much better than multi-part requests. A single prompt asking the AI to “research competitors, write a market analysis, create a strategy document and draft an executive summary” will produce shallow results for each section. Chaining gives you deeper and more accurate output at every stage.

Example: creating a blog post using prompt chaining

Step 1 — Research and outline:

“Create a detailed outline for a 1,500-word blog post about the benefits of remote work for small businesses. Include 5 main sections with 2-3 subtopics each.”

Step 2 — Draft the introduction:

“Using the outline above, write a compelling 200-word introduction that hooks the reader with a surprising statistic about remote work adoption.”

Step 3 — Write each section:

“Now write section 2 from the outline: ‘Cost Savings.’ Include at least one real-world example and keep it between 250-300 words.”

Step 4 — Edit and polish:

“Review the complete draft. Fix any inconsistencies between sections. Make sure the tone is consistent and the transitions between sections flow naturally.”

Each step produces focused output at a higher quality than trying to generate the entire post in one prompt.

Technique 6: Use Structured Formatting in Your Prompts

How you format your prompt matters, especially for complex requests. Using clear structure (with headings, numbered lists, bullet points or delimiters) helps the AI parse your instructions without confusion.

Anthropic recommends using XML tags to separate different parts of your prompt. OpenAI recommends delimiters like triple quotes (“””) or hash marks (###). Google recommends consistent formatting with clear section labels.

The principle is the same across all platforms: visually separate your instructions from your content so the AI knows which part is the task and which part is the data.

Example: structured prompt for data extraction

Instead of writing everything in one paragraph, structure it like this:

Task: Extract contact information from the following email and return it as JSON.

Required fields: name, email, phone, company

Rules:

  • If a field is missing, use null
  • Format phone numbers as +1-XXX-XXX-XXXX
  • Return only the JSON. No extra text

Email content: [paste email here]

This structured approach produces dramatically more reliable outputs than dumping everything into a single unformatted paragraph.

Technique 7: Ask the AI to Think Step by Step (Chain-of-Thought)

For tasks that require reasoning, analysis or math, you can ask the AI to show its thinking process before giving a final answer. This is called chain-of-thought prompting and it significantly reduces errors on complex tasks.

All three major AI assistants support this:

  • ChatGPT: has a built-in “Thinking mode” (GPT-5.3) that shows step-by-step reasoning
  • Claude: has “Extended Thinking” with visible step-by-step reasoning available on all plans including the free tier
  • Gemini: supports chain-of-thought when prompted explicitly

Even without these built-in features, you can trigger step-by-step reasoning by simply adding “Think step by step” or “Explain your reasoning before giving a final answer” to any prompt.

When to use chain-of-thought

Use it for Skip it for
Math and logic problems Simple factual questions
Debugging code Creative writing
Comparing multiple options Translation tasks
Strategic analysis Formatting or reformatting text
Complex research questions Short summaries

Example

Without chain-of-thought:

“Which is a better investment for a small bakery: a $3,000 commercial mixer or a $1,500 stand mixer?”

The AI might give a quick surface-level recommendation.

With chain-of-thought:

“Which is a better investment for a small bakery that makes 200 loaves per day: a $3,000 commercial mixer or a $1,500 stand mixer? Think step by step. Consider capacity, durability, daily output requirements, maintenance costs and break-even timeline before giving your recommendation.”

This prompt forces the AI to analyze each factor systematically before reaching a conclusion, producing a much more useful and nuanced answer.

Common Mistakes to Avoid

Knowing what to do is important, but knowing what NOT to do prevents frustration:

Mistake Why it fails What to do instead
Asking everything in one prompt The AI tries to satisfy every requirement at once, producing shallow output for each Break it into steps using prompt chaining
Using vague instructions “Make it better” or “improve this” gives the AI no direction Specify exactly what to improve: tone, structure, detail level or accuracy
Not reviewing the output AI models can hallucinate (generating confident-sounding but incorrect information) Fact-check any claims, statistics or dates the AI produces
Giving up after one try Prompt engineering is iterative. The first attempt rarely produces the perfect result Refine your prompt based on what was wrong with the first output
Using negation only “Do not use technical terms” is less effective than telling the AI what to do Say what you want: “Use everyday language that a 10-year-old would understand”

Quick Reference: The Prompt Engineering Checklist

Use this checklist every time you write a prompt to make sure you are covering the basics:

  • Task: Did I clearly state what I want the AI to do?
  • Format: Did I specify the output format (list, table, paragraph, JSON)?
  • Length: Did I set expectations for how long or short the response should be?
  • Tone: Did I specify the voice or style?
  • Audience: Did I tell the AI who will read this?
  • Context: Did I provide enough background information?
  • Examples: Would adding 1-2 examples make my instructions clearer?
  • Constraints: Did I mention any rules or restrictions?

You do not need to check every box for every prompt. A quick question to ChatGPT or Claude does not need a detailed prompt. But for anything important (a work deliverable, a research analysis or a coding project), running through this checklist will consistently improve your results.

Share this review

Related Guides

How to Automate Tasks with AI: A Step-by-Step Guide for Beginners

AI automation is no longer reserved for large enterprises with dedicated engineering teams. In 2026, anyone can automate repetitive tasks using no-code platforms that connect apps through visual interfaces and plain language instructions. Whether you want to auto-sort your emails, generate reports from spreadsheets or sync data between dozens of tools, there is a practical way to set it up today. This guide walks you through the entire process from identifying what to automate to building your first workflow and scaling it across your daily operations.

Beginner 16 min
Read

What Are AI Agents and How to Use Them: A Beginner’s Guide

AI agents are the next step beyond chatbots. Instead of just answering questions, an AI agent can plan tasks, use tools and take action on your behalf. It can research a topic, write a report, send an email and update a spreadsheet all from a single instruction. This guide explains what AI agents are in plain language, shows you the most useful agents available right now and walks you through how to start using them today.

Beginner 12 min
Read

How to Choose the Right AI Assistant in 2026: ChatGPT vs Claude vs Gemini

I pay for all three AI assistants — ChatGPT Plus, Claude Pro, and Gemini Advanced — totaling $60/month. After six months of daily use, I've learned exactly which one excels at what. This guide breaks down pricing, writing quality, coding ability, and unique features so you can pick the right one for how you actually work.

Beginner 8 min
Read

Top Picks Roundups

Ready to put these techniques into practice?

All three major AI assistants offer free plans. Pick one, apply these prompting techniques and see the difference in output quality for yourself.

All recommendations are editorially independent. We do not accept paid placements.