The battle loads in this browser.
Free · In this browser · No AI call
Prompt Battle
A head-to-head round of 8. Each battle shows a task and two prompts. Pick the one that gets the better result, then read a short sample of what each prompt tends to produce and the principle that won. When you want to write your own, the builder scores the slots as you fill them. Nothing you type is uploaded.
- Read the task A cover letter, a bug, a meal plan, a summary, a product line.
- Pick A or B Buttons, or the A and B keys. Then the samples open.
- Build one Role, task, context, constraints, and format, with a live checklist.
Battle 1 of 8
Score 0 of 0
Best on this device: none yet.
Principles mastered on this device: 0 of 10.
Press A or B, or use the buttons. The samples are written ahead of time.
What makes one prompt stronger
A short prompt can still be the better one when it carries the fact the model would otherwise invent. Role names the reader or the expert. Context is the empty list, the budget, the rain, the city. Constraints delete the answers you do not want: the shellfish, the extra apology, the word premium. Examples show a finished sentence in the voice you mean. Format is the container, so an open question has a slot to land in.
Iteration asks for the next draft and the fact still missing, instead of a single jump to a polished line. Audience is the person who already has the problem. Specificity puts a number on the page. Steps put the work in order. Sources say which notes the model may use, and tell it to stop when those notes run out. The samples on this page were written to show those differences. They are not a live reply.
Use the prompt somewhere else
Copy the builder result into a tool you already trust. The AI humanizer rewrites stiff paste, the Upwork proposal generator turns a job post into a draft, and the AI detector flags patterns in text you paste. The code explainer walks through a snippet. Those tools send text to an AI. This page does not. For a different kind of quiz, the Tech Myth or Fact deck stays in the browser too.
FAQ
Does this call an AI?
No. The battles, the samples, and the builder checklist all run in your browser. The page does not send a prompt to an API, and it does not wait on a model.
Are the samples live model output?
No. Each sample is a short piece written ahead of time to show what that prompt tends to produce. A real model can still wander. The sample is the lesson, not a guarantee of the next reply.
How does a round work?
Each round is 8 battles drawn from a deck of 20. You pick prompt A or prompt B, then see both samples and the principle that decided it. Battles you have already seen wait until every battle has been shown, then the deck starts over.
What does the builder checklist score?
It scores the five slots on this page: a specific role, a task with a deliverable, context the model cannot guess, a constraint, and a format. The score updates as you type. It does not judge the writing with an AI, and it does not save what you typed.
What is saved on this device?
Only your best score, the ids of battles you have seen, and the ids of principles you have mastered. The prompts you picked, and anything you type in the builder, are not stored and are not sent anywhere.
Will a perfect score write the prompt for my job?
No. A perfect round means you spotted the stronger prompt in these examples. Your real task still needs the facts only you have. The builder is a way to put those facts in the slots before you paste the prompt into a tool you trust.