Two people can ask the same AI the same question and get completely different answers. The difference is almost always the prompt.
Prompting isn't a mystical skill reserved for engineers. It's a handful of practical habits that anyone can learn in an afternoon — and they work in ChatGPT, Claude, Gemini, and every other chatbot. Here are the 12 techniques that matter most, from simple to advanced, each with a concrete example.
The basics: most bad prompts fail here
Before reaching for clever tricks, fix the fundamentals. Roughly 80% of disappointing AI output comes down to vague prompts missing one of these four things.
1. Be specific about what you want
Vague prompts get vague answers. The AI can't read your mind — it will fill in the blanks with its best guess, which is rarely what you wanted.
Before: "Write something about remote work."
After: "Write a 150-word LinkedIn post about the biggest challenge of remote work for new managers, in a conversational tone."
2. Give it context
Tell the AI who you are, who the audience is, and why you're asking. Context changes everything about the answer's tone and depth.
Before: "Explain compound interest."
After: "Explain compound interest to a 16-year-old who just opened their first savings account. Use a pizza analogy."
3. State the format you want
Bullet points or essay? Table or paragraph? 50 words or 500? If you don't say, you'll reformat it yourself — so save a step and ask upfront.
Before: "Compare solar and wind power."
After: "Compare solar and wind power in a two-column table: cost, efficiency, maintenance. Keep each cell under 20 words."
4. Define the tone
Professional, friendly, funny, formal — one adjective in your prompt can transform the output. Don't leave tone to chance when it matters.
Before: "Write an email asking for a raise."
After: "Write a polite but confident email asking my manager for a raise, citing my completed projects. Keep it under 120 words."
Intermediate: steer the answer like a pro
Once the basics are habit, these techniques give you far more control over quality and style.
5. Show it an example (few-shot prompting)
The fastest way to teach the AI your style: paste one or two examples of what "good" looks like, then ask it to match.
Prompt: "Here are two product descriptions I like: [paste examples]. Write three more in the same style for a stainless steel water bottle."
6. Assign it a role
Asking the AI to act as a specific expert focuses its answers and unlocks domain vocabulary it might otherwise hold back.
Prompt: "Act as a hiring manager with 15 years of experience. Review this cover letter and tell me the three weakest sentences."
7. Set constraints
Constraints are a gift. Word limits, forbidden words, required structure — they force the AI to be decisive instead of rambling.
Prompt: "Summarise this article in exactly 5 bullet points, each under 15 words. No jargon."
8. Ask for options, not one answer
When you want the best of something — headlines, names, ideas — ask for five or ten and pick your favourite. The AI's first answer is rarely its best.
Prompt: "Give me 10 headline options for a blog post about meal prepping on a budget. Make 3 of them funny."
Advanced: get reasoning, not just answers
These last four are for when the task is genuinely hard — planning, analysis, decisions.
9. Ask it to think step by step
For logic, maths, or multi-part problems, telling the AI to show its working dramatically improves accuracy. It catches its own mistakes mid-reasoning.
Prompt: "A store offers 20% off, then an extra 10% off the sale price. What's the final price of an $80 jacket? Think through it step by step."
10. Iterate instead of restarting
Your first prompt doesn't have to be perfect. Treat the AI like a colleague: react to its draft, then refine. "Make it shorter," "more formal," "add an example" — iteration beats perfectionism.
11. Ask the AI to ask YOU questions
When you're not sure what information the AI needs, flip the script. This is magic for complex tasks like trip planning or project specs.
Prompt: "I want to plan a week-long trip to Japan. Before you suggest anything, ask me the questions you need answered to make a great plan."
12. Request a critique of its own answer
After getting an answer, ask: "What are the weaknesses of what you just wrote?" or "Play devil's advocate." The AI will often spot flaws it glossed over — saving you an editing pass.
Frequently asked questions
Do these techniques work in every AI chatbot?
Yes. Be specific, give context, and set constraints — the core techniques are tool-agnostic and work in ChatGPT, Claude, Gemini, and most others. Advanced tricks like step-by-step reasoning work best in the more capable models, but the basics help everywhere.
How long should a good prompt be?
As long as it needs to be and no longer. A two-sentence prompt with clear context and format beats a 500-word mega-prompt full of contradictions. Start short, then add detail only where the output disappoints you.
Is prompt engineering a real job skill?
Knowing how to direct AI tools effectively is increasingly valuable at work — but it's a skill, not a job title. Employers care that you can get reliable results from AI, not that you know the jargon. The techniques above cover 95% of what matters.
What if the AI still gives bad answers?
Three usual culprits: the task genuinely needs information the AI doesn't have (paste it in), your instructions contradict each other (reread the prompt), or the model is too weak for the task (try a more capable model or break the task into smaller steps).
Key takeaways
- Most bad AI output comes from vague prompts — specificity, context, format, and tone fix the majority of cases.
- Show examples, assign roles, and set constraints to steer quality and style.
- For hard tasks: ask for step-by-step reasoning, iterate on drafts, and let the AI ask you clarifying questions.
- Prompting is tool-agnostic — these habits work in ChatGPT, Claude, Gemini, and beyond.