AI Prompt Generator

About the AI Prompt Generator

Most disappointing answers from a language model are not the model's fault. They come from a request that never said who the reader was, how long the piece should run, what it must avoid, or what a good answer would even look like. Ask for "a blog post about cybersecurity" and you will get a competent, weightless article aimed at nobody, because that is the only thing the request described.

Writing the request properly is a chore, though, and it is the same chore every time: state the role, name the audience, set the constraints, describe the shape of the output, and say what must not appear. This page does that chore. You describe the job in whatever words come to mind, and it comes back as a finished instruction you can paste straight in.

The target matters, so it is a setting rather than an assumption. An instruction for a chat assistant is written as a command — do this, for these people, under these limits. A request for an image model is not a command at all; it is a description, and one that reads like a sentence spoken to a person will produce a worse picture than a list of subject, composition, lighting and lens. Video adds the dimension that separates it from a still: what changes over the length of the shot. A coding assistant needs the runtime, the interfaces it must not break, and permission to ask rather than assume. An agent with tools needs to be told when to stop and what to confirm before acting.

Where your idea implies a detail without stating it, the generated prompt fills it in — that inference is most of the value. Where a detail genuinely cannot be guessed and would change the answer, it leaves a marked placeholder rather than inventing a figure you would then have to notice and remove.

How to use it

1. 1Describe the job in the box. Ordinary sentences are fine — that is what the tool is for.
2. 2Pick what the prompt is for: a chat assistant, an image or video model, a coding assistant, or an agent with tools.
3. 3Choose how detailed you want it. Compact gives a paragraph; Full adds an output format, a worked example and a checklist.
4. 4Turn on “Give me three versions” when you want distinct angles to choose between rather than one answer.
5. 5Copy the result and paste it into whichever model you use. Fill in any bracketed placeholder before you send it.

Questions

Why not just type my question straight into the model?

You can, and for a quick question you should. The gap opens on work you intend to reuse or hand over: a prompt you will run every week, one that has to produce the same shape of answer each time, or one a colleague will run without you there to explain what you meant. Specificity is what makes those repeatable, and specificity is tedious to write from scratch.

Will the prompt work on any model?

The structure travels well between the major chat models, since they all respond to a stated role, explicit constraints and a described output format. Image and video models vary far more in how they read a description, so treat what you get as a strong starting draft and expect to adjust the vocabulary for whichever one you use.

What are the square brackets in the result?

Places where a detail is needed and could not be inferred from what you wrote — a date, a company, a word count. They are left visible on purpose, because a plausible invention that you fail to spot is worse than an obvious gap you cannot miss.

Is my idea stored anywhere?

It is sent to the AI provider that produces the answer, and it passes through this site's server on the way without being written to any database, log or file that we keep. We have no account for it to be attached to. The provider's own retention is theirs, which is the honest limit of what can be promised about anything sent off the device.

Can it write a prompt that gets around a model's safety rules?

No, and it will not try. Requests to impersonate a named real person or to bypass another model's own rules are refused. That is a deliberate limit rather than an oversight.