AI text paste cleaner

Why it arrives full of asterisks

The text an assistant produces is written in Markdown, and the window you are reading it in renders that Markdown as you go. Two asterisks around a word mean bold. Hashes at the start of a line mean a heading. Square brackets followed by round ones are a link.

What you see is the rendered version. What the clipboard gets is the instructions. Word, Google Docs, an email field and almost every text box on the web have never heard of Markdown, so they show you the instructions: the asterisks, the hashes, the brackets and the addresses, all of it.

That is the whole problem, and it is a formatting one rather than a writing one, which is why it can be fixed by rules rather than by another model.

Where it is going changes what should happen

The destination is the first control on the page because it is the question that decides the rest, and the answers genuinely differ.

WhatsApp is the only one with markup of its own, and it is not Markdown: one asterisk for bold, one underscore for italic, three backticks for code. So emphasis is translated rather than stripped, and blank lines are pulled out from between list items, because there a blank line costs a whole row of a small screen.

Word and Google Docs take plain text through the clipboard and show every Markdown mark literally, so the marks come off. Their curly quotes and dashes stay: both handle them correctly and they are what the writer meant.

Email is the same, except that the punctuation is flattened. Somewhere between your drafts folder and the person reading it, something will render a curly quote as a question mark, and it is not worth finding out which link in the chain it was.

Plain text takes everything off and leaves nothing behind, which is what a form field, a caption or a code comment wants.

It is rules, not a model

Nothing here asks anything. The cleaning is a fixed set of steps that run in your browser, so the same text with the same switches gives the same answer every time.

That is worth more than the speed. A cleaner that sent your draft somewhere to be rewritten could also change what it says, and you would have no way of telling by looking. This one removes marks and moves whitespace, and the panel has a toggle between the result and what you pasted so you can check exactly that.

Your text stays in this tab. There is no request in the page that could carry it anywhere, nothing is stored, and it keeps working with the network off.

The two things it is careful about

Code is left alone. Anything between triple backticks keeps its asterisks and underscores exactly as they are, and only the fence lines come off. Outside a code block, an underscore inside a word is not emphasis, so annual_report_2024.pdf and user_name survive, and so does 2 * 3 * 4. Mangling a snippet is the classic failure of a naive stripper and it is silent, which is what makes it worth guarding against.

Numbers stay numbers. Bullet characters are made consistent, but a numbered list keeps its numbering. Steps in order are not the same information as the same steps as bullets.

Two bracket shapes are left alone, which is the case where doing less is the harder choice. Citation markers get removed: footnote markers, superscript digits, labelled ones. But a bare [3] is how legal and academic writing numbers its own references, and a bracketed capitalised phrase looks like a source name and is far more often a blank you are meant to fill in, such as [Your Name] or [Company]. Assistants produce those constantly in exactly the letter and contract templates somebody would bring here, and quietly deleting the blank a document is waiting for is worse than leaving a citation in: the citation is visible and the missing placeholder is not.