Guides
How to Write Better AI Prompts: 9 Techniques That Actually Work
Nine practical techniques for writing better AI prompts, with before-and-after examples and a copy-paste template for ChatGPT, Gemini, and Claude.
Most bad AI answers are not caused by a weak model. They are caused by a vague prompt. When you tell a chatbot 'write about marketing', it has to guess what kind of writing you want, who it is for, and how long it should be. So it plays it safe and gives you something generic. Good prompting is simply the skill of removing that guesswork.
The nine techniques below work across today's major chatbots - ChatGPT, Claude, Gemini, and others - because all of them respond the same way to clear instructions, useful context, and concrete examples. None of these require technical knowledge or special syntax. The biggest gains come from a few simple habits: giving the model a role, saying exactly what output you want, and treating the first answer as a draft rather than a final product.
Each technique includes a short explanation and a before-and-after example. At the end you will find a reusable prompt template and the most common mistakes to avoid.
The 9 techniques that actually work
1. Give the AI a role
A role tells the model which kind of expertise to draw on. 'You are an accountant' produces a very different answer than 'you are a comedian', even for the same question. Be specific about the role rather than using generic words like 'expert'.
Weak prompt: Write a product description for a coffee maker.
Better prompt: You are an experienced e-commerce copywriter who writes punchy, benefit-led product descriptions in a warm, conversational tone. Write a 100-word product description for a pour-over coffee maker aimed at busy parents. Focus on speed and ease of cleaning, not brewing science.
The second version tells the model exactly what kind of writer to be, who is reading, and what to emphasize. The result will sound written for a purpose instead of written for everyone.
2. Be specific about output format
Models default to whatever format seems most likely, which is often a long essay with an introduction and conclusion. If you want bullet points, a table, a headline under 60 characters, or code with comments, say so directly. Formatting instructions are one of the highest-leverage things you can add to a prompt.
Weak prompt: Summarize this report.
Better prompt: Summarize the report below into exactly 5 bullet points. Each bullet must be one sentence, start with a verb, and the whole summary must be under 120 words. No introduction or conclusion.
Being strict about format saves you the editing pass you would otherwise have to do by hand.
3. Provide context and constraints
AI models do not know your situation unless you tell them. Background details - your goal, your audience, your budget, your deadlines, your skill level - dramatically improve relevance. Equally important are constraints: what to avoid, what tone to skip, what topics are off-limits.
Weak prompt: Help me plan a trip to Japan.
Better prompt: Help me plan a 7-day trip to Japan in mid-October. My budget is $3,000 excluding flights, I am vegetarian, I love hiking and dislike crowded tourist traps. I have never been to Asia before. Give me a day-by-day itinerary with approximate costs, and skip anything that involves raw fish.
Notice how every detail in the better prompt eliminates a wrong guess. Context is what turns a generic answer into a personal one.
4. Show examples (few-shot prompting)
If you can show the model what good output looks like, do it. Even one or two examples will steer results more reliably than a paragraph of description.
Weak prompt: Reply to these customer reviews in our brand voice.
Better prompt: Reply to customer reviews in our brand voice. Here are two examples of our replies: Example 1 - Review: 'Took forever to ship.' Reply: 'You're right, and we're sorry. Your order left our warehouse yesterday and you should have it by Friday. If it doesn't arrive, email us and we'll make it right.' Example 2 - Review: 'Love this product!' Reply: 'That just made our day. Thanks for the kind words - enjoy it!' Now reply to the reviews below in the same style: short, honest, no corporate jargon.
Examples teach tone and structure faster than any list of adjectives.
5. Ask it to ask you questions first
Sometimes you do not know which details matter, or the task is genuinely ambiguous. Instead of guessing, ask the model to interview you. This flips the usual dynamic: the AI figures out what it needs to know, and you just answer.
Prompt to use: Before you write anything, ask me up to 5 clarifying questions that would help you do this task well. Wait for my answers, then proceed.
This works especially well for resumes, business plans, and technical writing, where a missing detail can derail the whole answer.
6. Iterate - treat the first output as a draft
The most common beginner mistake is accepting the first answer and walking away. Professionals prompt in rounds. Get a rough draft, then refine it with short follow-up instructions: 'make it shorter', 'use simpler words', 'add a specific example', 'make the opening stronger'.
Weak approach: Generate a welcome email, then copy-paste it into your product.
Better approach: Generate the welcome email, then follow up: 'Cut this to half the length. Then rewrite the subject line to be under 40 characters and mention the free trial. Finally, make the tone friendlier and remove the exclamation marks.'
Three or four rounds of iteration usually beats any single-shot prompt.
7. Break big tasks into steps
Models perform better on small, focused tasks than on enormous ones. 'Write a 20-page business plan' produces something shallow and repetitive. Ask for an outline first, review it, then ask the model to expand one section at a time while referencing the outline.
Weak prompt: Write a complete business plan for my bakery.
Better prompt: Step 1: Create a detailed outline for a business plan for a sourdough bakery in Denver, including executive summary, market analysis, operations, and financials. I will review the outline before we continue. Do not write the full sections yet.
Then expand one section at a time: 'Now write the market analysis section, about 400 words, using the outline above.' Each section stays consistent because it references the same plan.
8. Set the audience and tone explicitly
The same facts need very different treatment for a 12-year-old, a busy executive, and a fellow specialist. If you do not name the audience, the model guesses - and it usually guesses 'general educated adult', which may not be what you need. Name the reader and the tone in every prompt that produces writing.
Weak prompt: Explain how machine learning works.
Better prompt: Explain how machine learning works to a 12-year-old who is good at math. Use a friendly tone, avoid jargon, and include one analogy involving baking or cooking. Keep it under 200 words.
Audience and tone are the fastest way to change the feel of an answer without rewriting it yourself.
9. Use follow-up prompts to refine
Follow-ups are where prompting gets powerful. Once you have output on screen, you can steer it with short commands instead of starting over. Keep a mental list of these handy:
- 'Make this more concise' - when the answer rambles
- 'Explain this step in more detail' - when something is glossed over
- 'Give me 3 alternative versions' - when you want options
- 'Rewrite this for a different audience' - when the tone is wrong
- 'What assumptions did you make?' - when the answer feels off, this surfaces the model's guesses
- 'Challenge this argument' - when you want weaknesses exposed
Asking the model to state its assumptions turns invisible guessing into something you can correct.
A prompt template you can copy
Combine the techniques above into one reusable structure. Fill in the brackets and paste it into any chatbot:
Role: You are a [specific role, e.g. senior financial analyst]. Context: [background the model needs: your goal, situation, relevant facts]. Task: [exactly what to produce]. Format: [bullets, table, word count, sections]. Constraints: [what to avoid, limits, tone to skip]. Audience and tone: [who this is for, how it should sound]. If anything is unclear, ask me up to 3 clarifying questions before starting.
You will not need every field for every prompt - for simple questions, two or three are enough. But when a task matters, this checklist takes thirty seconds and prevents most bad outputs.
Common mistakes to avoid
- Being polite but vague. 'Please write something nice about our company' gives the model nothing to work with. Politeness is fine; vagueness is the problem.
- Piling on contradictory instructions. 'Make it funny but professional, short but detailed, casual but formal' - the model cannot satisfy all of these. Pick one direction per axis.
- Asking for facts without checking them. Models can state false information confidently. Verify anything factual with a reliable source before publishing or acting on it.
- Starting over instead of refining. If 80 percent of the answer is good, keep it and fix the 20 percent with a follow-up prompt. Regenerating from scratch wastes the parts that worked.
- Using the same prompt for every model. Different models have different strengths and quirks. A prompt tuned for one chatbot may need small adjustments on another.
Frequently asked questions
Do longer prompts always work better?
No. More precise beats longer. A 40-word prompt with a clear role, format, and audience will outperform a 200-word prompt that rambles. The goal is to include every detail that changes the output and cut everything that does not.
What should I do if the AI still gives bad answers?
First, check whether your prompt actually contains the information the answer is missing - usually it does not. Add that detail and regenerate. If the output is still weak, break the task into smaller steps or try a different model.
Does prompt engineering still matter as models get smarter?
Yes, arguably more. Smarter models make better use of good instructions - they follow complex formats, catch subtle constraints, and maintain consistency over long tasks. Vague prompts just give a smarter model more room to guess wrong.
Do I need to learn special syntax or frameworks?
Plain language works. Most prompt 'frameworks' are just reminders to include context, instructions, and format, which is what this guide covers. Save the advanced techniques for later; the nine habits here will handle the vast majority of real-world tasks.