Skip to the tool
CALCPARK

AI Prompt Generator

Write better prompts for ChatGPT, Claude or Gemini: fill in role, context, task, constraints, output format, examples and tone, preview the prompt live and copy it.

Runs on your device
Loading the tool…

How to use it

  1. Pick a preset close to your need (blog outline, code review, email reply, summary, explanation, translation) or start blank.
  2. Write the task first: one verb, a clear object and the size of the result (“Draft a 150-word reply”).
  3. Add the context the model cannot know and the constraints you would give a new colleague. Put strict rules here, not in the task.
  4. Describe the output format and, if quality matters, paste one or two examples of a good answer.
  5. Choose a section style, copy the prompt and paste it into ChatGPT, Claude, Gemini or any other assistant.

Example

Role “You are a senior TypeScript engineer reviewing a pull request”, context “The code runs in a statically exported Next.js app”, task “Review the code I paste below and find bugs and unsafe assumptions”, constraints “Only report real problems; quote the exact line”, format “Numbered list by severity with a suggested fix in a code block” and tone “Technical and precise” produce a six-part prompt of 585 characters (roughly 147 tokens). In Markdown style it starts with ## Role; in XML style with <role>. The “Code review” preset loads exactly this.

Why this structure

Prompt-engineering guides from the model makers agree on the same ingredients: give the model a role and the context it lacks, state the task clearly, add explicit constraints, specify the output format and show examples when the output must match a pattern (few-shot prompting). Separating the parts with headings or tags helps the model tell instructions from data – which is also why you should paste the text to work on after the prompt, not inside it.

  • Markdown headings (## Task) are understood by every chat model and remain readable to people.
  • XML tags (<task>…</task>) are the form Anthropic recommends for Claude; they are unambiguous even when your content itself contains headings.
  • Plain labels (Task: …) suit short prompts and voice assistants.

Token estimate: English text averages about four characters per token for GPT and Claude tokenizers, so the figure shown is characters ÷ 4, rounded up. Other languages and code use more tokens per character.

Sources: Anthropic – Prompt engineering overview (Claude docs), OpenAI – Prompt engineering guide, OpenAI Help – What are tokens and how to count them.

What the generator does not do

  • It does not call any AI model; it only assembles text for you to paste.
  • It does not check your prompt for quality – a vague task stays vague however nicely it is formatted.
  • Long documents belong after the prompt in the chat, not in the context field, to keep the instructions readable.

Questions

Do I have to fill in every field?
No. Empty fields are left out of the prompt. A task alone is a valid prompt; role, context and constraints are what most improve the answer.
Does the same prompt work for ChatGPT, Claude and Gemini?
Yes. All current chat models follow headed sections. Claude’s documentation prefers XML tags, ChatGPT and Gemini handle Markdown headings and XML equally well, so pick whichever you find easier to edit.
Where is my draft stored?
Only in your browser’s local storage, and only while "Remember on this device" is ticked. Unticking it deletes the saved copy. The share link carries just the section style.