AI Prompt Builder

How to build a prompt#

The builder only arranges your words; it does not call an AI model. Nothing you type leaves the page.

What makes a prompt work#

Assistants guess whatever you leave out. Most weak answers come from missing context, not from missing magic words. A good prompt answers the questions a new colleague would ask:

• What is the task? One clear instruction beats three vague ones.
• What does it need to know? Paste the email, the data or the draft instead of describing it.
• Who is it for? "A CFO with two minutes" and "a new intern" need different answers.
• What should the answer look like? Format and length save a round of edits.
• What should it avoid? Constraints such as "no jargon" or "do not invent numbers" go in their own list.
• Can you show an example? One sample of the output you like is often worth a paragraph of description.

Prompt structures#

• Labeled sections put each part under a heading. It is easy to scan and works everywhere.
• XML tags wrap each part in tags such as context and constraints. Anthropic's guides recommend tags for Claude, and other models read them well.
• Plain message writes the same parts as ordinary sentences.
• CO-STAR (Context, Objective, Style, Tone, Audience, Response), RTF (Role, Task, Format), RISEN (Role, Instructions, Steps, End goal, Narrowing) and RACE (Role, Action, Context, Expectation) are named frameworks people use in prompt courses.

The structure matters less than the content. If a framework has no slot for something you filled in, the builder still adds it under its own heading, so nothing gets lost.

Limits#

The templates cover common work tasks, not every situation; edit them freely. Very long prompts take up context that the answer needs, so check the size with the Token Counter if you paste whole documents.