Paste the answer and get it back in the shape you would have typed it. Every pass at once, and a list of what was actually wrong — because most of it you cannot see.
A paragraph out of a chatbot rarely has just one thing wrong with it. It has em dashes and curly apostrophes and a stray asterisk where the bold used to be, and somewhere in there a zero-width space that will make a spreadsheet lookup fail next Tuesday.
Which is a nuisance when you have to work out which is which before you can fix any of it, and worse when one of them is invisible. So they all run in one go, in the order set out above. You do not have to know what you are looking at to get rid of it.
The report under the fields is the part worth reading. Most of what gets cleaned here is either invisible or so small you would scroll past it, so text that comes back looking identical is the normal case, not a sign that nothing happened.
Each line of the report links to the tool that deals with that one thing on its own, where the defaults are explained and you get finer control. Use this page when you just want it fixed; go to the specific one when you want to decide exactly how.
It will not rewrite a sentence. Nothing here is a language model: it is find-and-replace with the edge cases thought through, which is why it is instant, free and gives the same answer every time.
So it will not make AI writing sound human, and it will not get anything past a detector. Detectors score sentence rhythm and word predictability, and punctuation is not part of that arithmetic. What this does is stop the text looking like it was pasted from somewhere else, which is a different problem and usually the one you actually have.
Both are kept. Fenced code comes through with its contents intact, and nothing inside it is treated as formatting, so an asterisk in your code stays an asterisk. A table comes back as rows with a tab between the cells, which reads sensibly and pastes into a spreadsheet with the columns where they should be. If you would rather drop either of them, the Markdown to Plain Text tool has the switches.
Read the report rather than the text. Most of what this removes is either invisible or a single character you would scroll straight past, so identical-looking output is the normal case. If the report is empty, your text really was clean and you have ruled something out.
Coverage against control. This page runs everything with sensible defaults and is the one to use when you just want the text fixed. The single-purpose pages let you decide the awkward cases: whether an en dash in a range should go, whether guillemets are a mistake or correct French, what a link should turn into. Each line of the report links to the relevant one.
No to both, and be suspicious of anything claiming otherwise. There is no language model here - it is find-and-replace with the edge cases thought through, which is why it is instant and free. Detectors score sentence rhythm and how predictable each next word is; punctuation is not part of that arithmetic. What this fixes is text that looks pasted from somewhere else, which is a different problem and usually the one you actually have.
It is JavaScript running in your own browser, so nothing is uploaded, logged or sent anywhere. Open your browser's Network tab and watch while you type if you would rather not take my word for it.