12 Prompt Writing Techniques That Actually Improve AI Results
Twelve practical prompt techniques for ChatGPT, Claude and Gemini, with a reusable template you can copy for everyday work.
The 3 that matter most
- Context beats clever wording. Say who you are, who it’s for and what good looks like.
- Show, don’t describe. One example of the output you want is worth a paragraph of adjectives.
- Iterate. The first answer is a draft. Specific feedback gets you the rest of the way.
Most disappointing AI answers come from short, vague prompts. The model fills the gaps with the most generic answer possible. These twelve techniques work in ChatGPT, Claude, Gemini and any other assistant, and they don’t require any special syntax: just clearer instructions, the kind you’d give a smart new colleague on their first day.
1. Give it a role and an audience
“You’re a tax accountant writing for small-business owners who hate jargon.” A role sets the expertise; an audience sets the level and tone. Together they change the answer more than any other single line.
2. State the goal, not just the task
“Write an email” is a task. “Write an email that gets a busy client to approve the budget by Friday” is a goal, and the model will choose content, length and tone to reach it.
3. Paste the source material
Ask about the document you provide, not about what the model remembers. Grounded answers are more accurate and more specific. Long PDFs and spreadsheets can usually be uploaded directly.
4. Show an example
Paste a paragraph you like and say “match this style and length”. For structured output, show one finished row or item and ask for more in the same format.
5. Ask for a specific format
A table with named columns, five bullets, a 150-word summary, a numbered checklist. Format instructions are followed very reliably and make answers easier to use.
6. Set constraints
Word limits, words to avoid, reading level, what not to include. “No buzzwords, no exclamation marks, UK spelling” saves a round of edits.
7. Break big jobs into steps
Instead of “write my report”, ask for an outline first, agree on it, then draft section by section. You stay in control and quality goes up.
8. Ask it to think before answering
For analysis and decisions: “Think through the options step by step, then give your recommendation.” Many models now reason automatically, but asking for the reasoning still helps you check it.
9. Ask clarifying questions first
“Before you start, ask me up to five questions that would improve the result.” This is the fastest way to discover what context you forgot to give.
10. Give feedback like an editor
“Shorter” is weak feedback. “Cut the intro to one sentence, make the second point a table, and end with a clear next step” gets you there in one turn.
11. Ask it to critique itself
“What’s weak, missing or possibly wrong in this answer?” Then ask it to fix those points. Two passes beat one long prompt.
12. Save prompts that work
Keep a library of prompts you reuse, with [brackets] for the parts that change. Most assistants also let you save instructions: custom GPTs in ChatGPT, Projects in Claude, Gems in Gemini.
Template: “You are [role]. I need [deliverable] for [audience] so that [goal]. Use the material below. Format: [format]. Constraints: [length, tone, things to avoid]. Ask me questions first if anything important is missing.”
Try these techniques in
GeminiVerifiedChatbots · FreemiumGoogle’s AI assistant connected to your Google apps
Want ready-made prompts for your industry? Our prompt packs collect tested prompts for lawyers, consultants, finance, marketing, real estate and HR.
Frequently asked questions
Do prompts work the same in every AI tool?
Mostly. The techniques here are universal. Each model has its own style, so you may tweak wording, but context, examples and format help everywhere.
Are long prompts better?
Longer isn’t better by itself; complete is. Include everything a capable colleague would need, and nothing they wouldn’t.
Should I use prompt templates from the internet?
They’re a good starting point. The best results come when you add your own context, examples and standards.
