Text tool

Prompt Builder

Type the request you actually mean, and get back a proper brief: a role, the situation, who it is for, how it should sound, what shape the answer takes, and the rules to follow. Pick the AI you are working with and the structure changes to suit it. Everything is assembled in your browser, so your prompt never leaves the device.

Your prompt never leaves your browser. There is no account, no upload and no model behind this, so there is nowhere for the text to go even in principle.

Your request

0 characters

Try one of these:

Which AI are you writing for?

What kind of request is it?

Detected: Writing

Optimised prompt



    


    

    

    

    


    

    

    Before
    

    

    

    

    After
    

    

    

    


    

    Try a different angle
    

    

    


    





    


    Ctrl or Cmd with Enter rebuilds. Ctrl or Cmd with Shift C copies.
    

    

    






    

    


    




    

    
    
    

    Why a lazy prompt gets a lazy answer
    

    Ask for “an email” and the model has to guess almost everything: who it is writing to, what the email is meant to achieve, how formal to be, how long it should be, and what to leave out. It will guess, and it will guess differently depending on the version you happen to be running. Everything that is not specified is a decision handed to a system that has no reason to make the same one twice.
    

    A prompt that names those things removes the guessing. It is a longer piece of text and a slightly slower first reply, and after that the answers stop drifting.
    


    What this tool actually does
    

    Nothing here calls a model. It detects what kind of request you have made from keywords, picks a matching template, and fills it in from a form you can see and change. The output is string assembly, which is why it is free, needs no key, works offline once the page has loaded, and can tell you exactly what it did.
    

    Because the assembly is visible, the quality score is honest. Six rules, all shown: does it give the AI a role, explain the situation, state a clear task, ask for a specific format, set limits, and name an audience or tone. The score is a checklist, not a verdict.
    


    Four AI styles from one template
    

    Each AI reads a brief differently. ChatGPT is given a system-style brief with clear headings. Claude gets the same content in XML-style tags, which it follows closely. Midjourney gets a single visual line with parameters, because a wall of prose means nothing to an image model. Anything else gets plain labelled lines, which survive being pasted into any chat box.
    


    What to fill in
    

    Anything left empty shows as a marker such as [RECIPIENT] rather than being quietly dropped, and the markers are highlighted so a half-finished prompt cannot look finished. The usual reason to leave one is that you do not know the answer yet, and it is useful to see what the prompt needs before you decide.
    


    Keyboard
    

    Ctrl or Cmd with Enter rebuilds the prompt. Ctrl or Cmd with Shift C copies it. Everything else is reachable by tabbing, and the options are real switches rather than icons, so they announce their state.
    

    


    

    Further reading: Read the full guide: Prompt Builder
        
    
    Common questions
    Is my prompt sent to an AI service?
    No. There is no model and no account here, so there is nowhere for your text to go. The tool assembles a better prompt from what you type using its own templates, in this tab. Your draft is only as private as this page.
    Does it write the answer for me?
    No, and that is the limit of what it can do. It restates your request in a form an assistant is more likely to follow, with the role, the format and the constraints spelled out. What comes back is still whatever that assistant decides, which is why the tool never claims to know the answer.
    What actually makes a prompt better?
    Three things, in order: saying who the assistant should be, saying exactly what to return and in what shape, and saying what to avoid. A request with those three in it is mostly answered properly, where "write something about X" gives the assistant room to guess at all three.
    Should I pick an output format?
    Yes, if you want something you can use straight away. Asking for a table, numbered steps or JSON stops the assistant choosing its own shape, which is the usual reason a result is nearly right but needs rewriting before it can be pasted somewhere.
    What are the avoid rules for?
    They cut the stock answers. Without them an assistant pads a reply with restatement, "it is important to note" and offers to help further. Listing what you do not want is often more effective than adding more of what you do.
    Can I keep what I have written?
    Yes. Recent prompts are kept in this tab so you can go back to one, and the finished prompt can be copied or downloaded as a text or markdown file. Nothing is stored on a server, so clearing the tab clears it.
    		
    

			
			
			
			
	
    

		
		
			
    


    


    
	
	
	
	
		
    
		
    Leave a Reply