How to write AI prompts that get better answers

The prompts in the library above are ready to copy and paste, but understanding why they work will make every prompt you ever write better. Most disappointing AI answers trace back to the prompt, not the model. A vague prompt gets a vague answer β€” the model can only work with what you give it.

Five techniques that change everything

1. Give context, not just a task. "Write a product description" produces generic copy. "Write a product description for a $40 insulated water bottle aimed at trail runners, in a rugged no-nonsense tone, under 80 words" gives the model the constraints it needs to be useful. Context is the cheapest way to improve output quality.

2. Assign a role. Starting with "You are a senior email copywriter" or "Act as a Python mentor" focuses the model's knowledge and tone. It is not magic β€” it simply narrows the enormous space of possible responses toward the expertise you need.

3. Show an example. If you want a specific format, show it. Paste one example of the output you want and say "follow this format." This technique, called few-shot prompting, is dramatically more reliable than describing a format in words β€” especially for tables, structured summaries, and code.

4. Iterate instead of restarting. Your first answer is a draft, not a verdict. Follow up with "make it shorter," "more formal," or "explain the second point in detail." Each round narrows in on what you actually want, and it is far faster than rewriting the original prompt from scratch.

5. Ask for reasoning when it matters. For logic, math, or planning tasks, add "think through this step by step before answering." Making the model show its work catches errors early and produces more reliable results.

Before and after: a real example

Weak: "Write about our new app." This could mean anything β€” a tweet, a press release, a landing page. The model guesses, and the guess is usually wrong.

Strong: "Write a 120-word App Store description for a habit-tracking app for busy parents. Tone: warm and encouraging, not preachy. Mention the 2-minute daily check-in and the streak feature. End with a call to action to download." Every sentence removes ambiguity. The result needs almost no editing.

Common prompting mistakes

  • Asking two things at once. "Summarize this and also write a tweet about it" splits the model's attention. Do one task per prompt, then chain them.
  • No constraints. Without a length, format, or audience, you get the model's average answer β€” which is nobody's ideal answer. Always specify length and audience.
  • Accepting the first draft. The first response is the model's best guess at what you meant. Treat it as a starting point and refine.
  • Overloading with background. Dumping 10 pages of context and asking a vague question buries the actual task. Lead with the task, then add only the context it needs.
  • Forgetting the model can't see your screen. "Fix this" with no code pasted, or "make it better" with no draft attached, leaves the model guessing. Paste the material you are talking about.

Tips by category

Writing: specify tone, audience, and length every time. Ask for three variants of headlines or intros instead of one β€” picking beats rewriting.

Coding: paste the error message and the relevant code, state the language and framework, and describe what you expected versus what happened. For new code, describe inputs, outputs, and edge cases.

Image prompts: lead with the subject, then style, then details β€” "a lighthouse at dusk, watercolor style, soft washes, muted palette." Negative prompts ("no text, no watermark") help image models avoid common artifacts.

Business and productivity: give the model your raw material (notes, a messy draft, bullet points) and a clear target format. AI is best at transforming content you already have, not inventing it from nothing.

The library above puts all of this into practice β€” each prompt bakes in the context, constraints, and structure that make models perform. Use them as-is, or steal their structure for your own.