“How to Write Better AI Prompts for ChatGPT and Gemini”
Most bad AI answers are bad prompts in disguise. Learning how to write AI prompts well is the highest-leverage skill in the AI era: the same tool gives dramatically better results to someone who asks well. These 10 proven techniques work on ChatGPT, Gemini, Claude, and basically every major AI assistant. None of them require technical knowledge, and you'll feel the difference on your very next prompt.
How to write AI prompts: start with strong context
The model can't read your mind, so the first three techniques are about giving it what it needs up front.
1. Give context, not just commands. "Write an email" gets you a generic email. "Write a polite follow-up email to a client who missed our Tuesday call, keep it under 100 words, friendly but professional" gets you something usable. Every relevant detail narrows the gap between what you want and what you get. Before hitting send, ask yourself: who is this for, what's the tone, and what does good look like.
2. Show an example of what you want. This is the single most powerful technique most people never use. Instead of describing the format, paste a short example. "Write product descriptions like this: [example]" beats three paragraphs of formatting instructions. Models are excellent mimics. One good example often does more work than a page of directions, and it takes ten seconds to add.
3. Assign a role. "Act as a hiring manager reviewing this resume" or "You are a patient math tutor" changes the lens the model uses. Role prompts work because they activate patterns from the model's training: it has seen countless examples of how tutors explain and how managers evaluate. Keep roles specific rather than grandiose. "Experienced plumber explaining to a homeowner" beats "world's greatest expert."
Shape the answer before it arrives
4. Break big tasks into steps. Asking for a complete business plan in one prompt usually produces something shallow. Instead, work through it in stages: first the market analysis, then the financials, then the marketing section. Each step gets the model's full attention, and you can correct course between steps. Think of it like managing a smart assistant: small, clear assignments beat one giant vague one.
5. Ask for structure explicitly. Models default to walls of text unless told otherwise. If you want a table, say so. If you want pros and cons as bullet points followed by a recommendation, spell it out. "Compare these three options in a table with columns for price, features, and best for" takes five seconds to type and saves you reformatting the answer yourself.
6. Set constraints and length. "Explain photosynthesis" gets a textbook chapter. "Explain photosynthesis in 3 sentences a 10-year-old would understand" gets exactly what you need. Constraints on length, reading level, and format are the fastest way to make outputs usable. People consistently under-constrain their prompts, then complain the answer wasn't what they wanted.
7. Make it think step by step. For anything involving reasoning, math, or complex decisions, add "think through this step by step" or "show your reasoning." This simple phrase measurably improves accuracy on hard problems across every major model. It forces the model to work through the logic visibly instead of jumping to an answer, which also lets you spot where it goes wrong.
Treat the first answer as a draft
8. Iterate instead of restarting. Your first prompt is a draft, not a final. When the answer is close but off, don't start over. Say what to change: "shorter," "more formal," "add an example for small businesses," "you missed the pricing question." Each round of feedback sharpens the result, and three quick iterations beat one perfect prompt almost every time. The best prompters are the best editors.
9. Ask the AI to improve your prompt. Here's a meta trick that works surprisingly well. Paste your draft prompt and add: "Before answering, suggest how I could improve this prompt to get a better result." The model will often point out missing context or ambiguity you didn't notice. It's like having an editor for your instructions, and it teaches you to write better prompts on your own over time.
Know when the tool is the problem
10. Switch tools when the prompt isn't at fault. Sometimes the prompt isn't the problem. Questions about this week's events go to the tool with the best real-time access. Long document analysis goes to the one with the biggest context window. Image tasks go to the best image model. Part of prompting well is knowing which assistant to ask. A great prompt in the wrong tool still disappoints.
The throughline across all ten: specificity and iteration beat cleverness. You don't need secret prompt formulas or expensive courses. Give context, show examples, constrain the output, and refine in rounds. Do that consistently and you'll get more from free AI tools than most people get from paid ones. Try it on your next real task and compare the before and after. The difference will sell itself.
Do longer prompts always work better?
Not always, but detailed prompts usually beat vague ones. The key is relevant detail, not word count. A 50-word prompt with clear context, an example, and constraints outperforms a 300-word ramble. Focus on answering: what's the task, who's it for, what format, and what does good look like.
What's the biggest prompting mistake beginners make?
Asking once and accepting whatever comes back. Beginners treat the first answer as final, while experienced users treat it as a starting draft. The real skill isn't writing the perfect prompt, it's the two or three follow-up refinements that turn a decent answer into exactly what you needed.
Do these techniques work on all AI chatbots?
Yes. These are general principles about how large language models respond to instructions, so they apply to ChatGPT, Gemini, Claude, Grok, DeepSeek, and others. Individual models have quirks, but context, examples, roles, constraints, and iteration improve results everywhere.
