AI Prompt Generator

Why does this generator work?

Every prompt it builds follows the structure prompt-engineering professionals use: expert role + concrete task + context + audience + tone + output format + quality criteria. It's the difference between telling the AI "write me an email" and handing it an agency brief.

Use it as a starting point: generate, paste into Claude (or whichever AI you use), and refine with our prompt library. Want the AI to always apply your rules without repeating them? The next level is skills.

This generator works on text prompts. If what you want to describe is an image, the rules change — style, lighting and framing matter more than role and tone — and there is a separate tool for that: the AI image generator, which also produces the image without leaving the page.

Why a long prompt beats a short one

Almost everyone's instinct when starting out is to write as little as possible: "summarise this", "write me an email". And it works, more or less, for trivial tasks. The problem shows up as soon as the task has criteria: the model then has to guess the tone, the length, the technical level and the format, and it guesses wrong more often than it guesses right. Not because it's stupid, but because there is no correct answer to "write me an email" without knowing to whom and why.

A well-built prompt isn't long for the sake of it — it's one that removes the decisions the model would otherwise have to make on your behalf. Each field in the generator above corresponds to one of those decisions. The role fixes the point of view and the vocabulary. The task says what has to be produced. The audience determines what can be assumed and what needs explaining. The tone sets the register. The length solves the eternal problem of getting three paragraphs when you wanted three sentences. And the format decides whether you get a list, a table or running text.

The practical test is simple: take one of your own one-line prompts, run it through the generator filling in those fields, and compare the two answers. The difference is almost never in the quality of the prose; it's in how many times you have to ask again.

The four mistakes that ruin a prompt

Asking for two things at once. "Analyse this data and write me the report" produces a weak analysis and a generic report. The model splits its effort. Separate them: do the analysis first, read it, then ask for the report based on the conclusions that convinced you. Chaining two good prompts always beats one that tries to do everything.

Describing what you don't want instead of what you do. "Don't use corporate language" is far less effective than "write it as if you were explaining it to a colleague over coffee". Prohibitions leave the space open; positive descriptions close it. If you have to forbid something, forbid it as well as describing the alternative, not instead of it.

Giving no example at all. A single example of the output you want — even an imperfect one scribbled in thirty seconds — is worth more than three paragraphs of instructions. It's the technique the technical literature calls few-shot, and it's probably the best effort-to-result trick that exists. If you have an earlier piece you liked, paste it in and say "in this style".

Treating the first answer as final. The conversation is the tool, not the isolated prompt. "That's good, but the second paragraph is redundant and the ending is soft" is a perfectly valid instruction and usually improves the result more than rewriting the original prompt from scratch.

When this generator isn't the right tool

This generator builds text prompts: writing, analysis, summaries, code, planning. There are two cases where it doesn't fit, and it's worth knowing which.

If what you want is an image, the rules are different: image models don't follow conversational instructions, they follow dense descriptions where word order matters and lighting or framing must be stated explicitly. That's what the image generator is for, and it produces the image right there.

And if what you need is for the AI to do the same task over and over to your standards — reviewing contracts, triaging tickets, drafting replies in your company's voice — the prompt stops being the right place to keep those instructions. What you want then is a skill: a set of standing instructions Claude loads only when they're relevant. The guide to creating your own skills explains how to turn a prompt that works into a reusable skill.

What to do with the prompt once you have it

Copy it and paste it wherever suits you: it works the same in Claude, ChatGPT, Gemini or any other model, because it doesn't rely on proprietary tricks — just on telling the machine precisely what you want. The "try in Claude" button is there for convenience, not because the prompt is exclusive to it.

If you're going to use it more than once, save it. The prompt library collects the ones that have worked best, sorted by category, and gives you a starting point for your own; and if you're unsure which model to use for the task you've just described, the model selector settles it in four questions and the cost calculator tells you what it would cost via API if you're thinking of automating it.