AI Prompt Builder
Free tools · AI & Prompts
Build a prompt that leaves nothing to guess
Pick the model, the role and the rules. The builder assembles a structured prompt you can paste anywhere, and it does it here, in your browser: no model, no account, nothing uploaded.
- Nothing is uploaded
- Works offline
- Structure per model
Where will you paste it?
The structure changes; your words do not.
The task
One sentence is enough if it says what “done” looks like.
Shape of the answer
Rules it must respect
These are the lines a model forgets when nobody writes them down.
Example of a good answer (optional)
One sample teaches more than three paragraphs of instructions.
Start from a preset
Fills the fields; you can change everything afterwards.
Write the task first (step 02) and the prompt will build itself.
What the builder is actually doing
A good prompt is not longer: it is ordered. These five blocks are the order that survives pasted instructions, long chats and model changes.
The fifth block is the shape of the answer: format and length. Same prompt, different model: the structure changes (tags for Claude, headings for ChatGPT and Gemini, short lines for a local model), but the content stays yours.
Related tools
Measure it, adapt it, or draw with it.
Frequently asked questions
- Does this use AI?
- No. It assembles a structured prompt from your choices inside your browser. No model runs here, nothing is sent anywhere and no key is needed; the AI is whatever you paste the prompt into.
- Which model should I pick?
- The one you are going to paste it into. ChatGPT and Gemini read headed sections well, Claude reads XML tags well, and a local model does better with short, direct lines. Picking a different model only changes the structure, never your words.
- Can I reuse the same prompt with placeholders?
- Yes. Write anything like {{topic}} in a field and the builder keeps it as written, so the same prompt can be filled in later or shared with a colleague.
- Where can I see the exact token count?
- The counter at the top is a rough estimate (characters divided by four). For the exact number per model, paste the result into the Token Counter, which runs the real tokenizer on your device.