How to Write Better AI Prompts: 10 Techniques with Examples
Most disappointing AI answers come from short, vague prompts. The model is not reading your mind; it is filling gaps with the most generic answer that fits. Better prompts close those gaps. You do not need special syntax or "magic words". You need to tell the AI what a smart new colleague would need to know to do the job well.
The ten techniques below work across ChatGPT, Claude, Gemini, Microsoft Copilot and most other chat assistants. Each comes with a before and after example you can adapt.
The 10 techniques at a glance
| # | Technique | Fixes this problem | Use it when |
|---|---|---|---|
| 1 | Give context | Generic, one-size-fits-all answers | Almost always |
| 2 | Assign a role | Wrong level of expertise or tone | You want a specific perspective |
| 3 | State the goal and audience | Answers that miss the point | Writing anything for readers |
| 4 | Show examples | Wrong style or format | You have a clear model to copy |
| 5 | Set constraints | Too long, too fancy, off-limits content | Output has rules |
| 6 | Specify the format | Walls of text | You will reuse or paste the output |
| 7 | Ask for step-by-step reasoning | Sloppy logic, wrong maths | Analysis, planning, calculations |
| 8 | Break big tasks into steps | Shallow results on complex work | Long documents, projects |
| 9 | Let it ask you questions | You are not sure what to include | Unclear or high-stakes tasks |
| 10 | Iterate and give feedback | Stopping at a mediocre first draft | Every important task |
1. Give context
The model knows nothing about your situation unless you say it. Context is the single biggest improvement you can make.
- Before: "Write an email asking for a deadline extension."
- After: "I'm a freelance designer in Bengaluru. My client, a café chain, expects a logo on Friday. Their brand team sent feedback two days late, so I need until next Wednesday. Write a short, polite email asking for the extension, explaining the delay without blaming them, and offering to share rough drafts on Friday."
The second prompt produces an email you can send with light edits. The first produces a template you must rewrite.
2. Assign a role
A role sets the expertise level, vocabulary and viewpoint. Keep it relevant, and pair it with context; a role on its own does little.
- Before: "Review my business plan."
- After: "Act as an experienced small-business lender reviewing loan applications. Review my business plan for a cloud kitchen in Hyderabad and point out the five weakest assumptions a lender would question, in order of importance."
3. State the goal and audience
Tell the AI who will read the output and what you want them to do or feel.
- Before: "Explain inflation."
- After: "Explain inflation to a Class 8 student in India who has never studied economics. Use one example with prices of everyday items like milk or a bus ticket. The goal is for them to explain it to a friend afterwards. Under 200 words."
4. Show examples
When style matters, showing beats describing. One or two examples are often enough. This is sometimes called "few-shot" prompting.
- Before: "Write product descriptions for my handmade candles."
- After: "Write descriptions for these three candles in the same style as this example:
Example: *Monsoon Earth. Smells like the first rain on dry soil. Hand-poured soy wax, cotton wick, about 40 hours of burn. Made in small batches in Jaipur.*
Candles: (1) jasmine and sandalwood, (2) masala chai, (3) unscented beeswax."
Tip: if you give only one example, the AI may copy it too closely. Two different examples help it find the pattern rather than the words.
5. Set constraints
Constraints are rules: length, reading level, words to avoid, things not to include, facts not to change.
- Before: "Write a LinkedIn post about our new office."
- After: "Write a LinkedIn post announcing our new office in Pune. Under 120 words. No hashtags in the body, three at the end. Avoid the words 'thrilled', 'excited' and 'journey'. Mention that we are hiring two backend developers. Do not invent any numbers."
"Do not invent any numbers" is worth adding to almost any business prompt.
6. Specify the format
If you will paste the result into a spreadsheet, slide or document, say exactly what shape you want.
- Before: "Compare three laptops for me."
- After: "Compare these three laptops for a college student who does coding and light video editing: [models]. Give a table with columns: model, processor, RAM, battery life, weight, main strength, main weakness. Then one paragraph recommending one, with the reason. If you are not certain of a spec, write 'check' instead of guessing."
Useful format words: table, bullet list, numbered steps, JSON, headings, one paragraph, email, script with timestamps.
7. Ask for step-by-step reasoning
For maths, logic, planning and decisions, asking the model to work through the problem before answering tends to reduce careless mistakes. Many newer models do some of this automatically, but asking still helps you check the logic.
- Before: "Is it cheaper to lease or buy a delivery scooter for my business?"
- After: "Help me decide whether to lease or buy a delivery scooter. Work through it step by step: list the costs of each option over three years, state any assumptions clearly, calculate the totals, then give a recommendation. My numbers: [purchase price, lease rate, fuel, maintenance, expected resale]. Show the calculation so I can check it."
Always verify the arithmetic yourself for anything involving money.
8. Break big tasks into steps
Asking for a complete 3,000-word report in one go usually gives a shallow result. Split the job and review each stage.
- Before: "Write a full training manual for new staff at my restaurant."
- After (step 1): "I need a training manual for new waiters at a 40-seat South Indian restaurant. First, propose an outline with 8–10 sections and one line on what each covers. Don't write the sections yet."
- Then: "Good. Remove section 7, merge 3 and 4. Now write section 1 only, in simple English, under 400 words."
You stay in control of structure, and each section gets full attention.
9. Let the AI ask you questions
When you are not sure what details matter, ask the model to interview you first.
- Before: "Make me a diet plan."
- After: "I want a simple weekly vegetarian meal plan. Before you write it, ask me up to 8 questions you need answered to make it suitable for me. Ask them all at once, then wait for my answers."
This is especially useful for plans, resumes, cover letters and anything personal. For health, legal or financial matters, treat the result as a starting point to discuss with a professional, not as advice.
10. Iterate and give specific feedback
The first answer is a draft. The best results come from two or three rounds of precise feedback.
- Weak feedback: "Make it better."
- Strong feedback: "Keep the structure. Make the opening line a question. Cut the third paragraph. Replace the formal words with simpler ones a customer in a small town would use. The ending should tell them to message us on WhatsApp."
If a conversation gets long and the AI starts drifting, start a new chat and paste in a short summary of what you want plus the best version so far.
Putting it together: a reusable prompt template
You do not need all ten techniques every time. For anything important, this structure covers most of them:
- Role: Who should the AI act as?
- Context: What is the situation and background?
- Task: What exactly should it produce?
- Audience and goal: Who reads it, and what should they do?
- Constraints: Length, tone, must-include, must-avoid.
- Format: Table, list, email, script.
- Example (optional): A sample of the style you want.
- Check: "Ask me questions if anything is unclear" or "flag anything you're unsure about."
Full example: "Act as a customer support lead (role). We sell refurbished smartphones online across India, and customers often ask about warranty after delivery (context). Write a reply template for warranty questions (task) that a first-time online buyer will understand and trust (audience and goal). Under 100 words, friendly, no legal jargon, mention our 6-month warranty and that they can reply with their order number (constraints). Give two versions: one for email, one for WhatsApp (format). Flag anything you think is missing (check)."
Common mistakes
- Stuffing in "magic" phrases. Tricks like promising a tip or saying "this is very important" are no substitute for clear context.
- Asking several unrelated things in one prompt. Split them.
- Trusting facts without checking. AI can state wrong figures, dates, citations and laws confidently. Use a search-grounded tool like Perplexity for research and still open the sources.
- Pasting sensitive data. Remove personal, customer and confidential company information before you prompt.
- Giving up after one try. A two-line follow-up often fixes what a whole new prompt would not.
Does the tool matter?
The techniques work everywhere, but each assistant has its own strengths, limits on file uploads, and ways to save instructions (custom instructions, projects, gems and similar features). If one tool keeps misunderstanding a task, try the same prompt in another. Compare options on our ChatGPT alternatives, Claude alternatives and Gemini alternatives pages, or see side-by-side comparisons on /compare.
Quick picks
- One habit to start today: add two sentences of context to every prompt.
- For style: show an example instead of describing the tone.
- For anything complex: ask for an outline first, then build section by section.
- When unsure what to say: ask the AI to ask you questions.
- Ready-made starting points: browse our prompts library and adapt them with your own context.
Tools and plans change often — always check each product's own website for current features, pricing and terms. aixplor.in is independent; listings and rankings are not paid placements.