Text cleanup guide · October 14, 2026

How to Check Claude Text for Hidden Characters

Direct answer: Copy the text, paste it into Claude Text Cleaner, and read the Removed characters count. A count above zero means this browser-only cleaner found characters on its removal list. Compare the clean result with your original before copying it. A zero means no supported characters were found; it does not prove anything about the text's source or a model watermark.

Check pasted text locally

Read the report in three steps

1. Copy the passage from Claude or another AI tool. On the cleaner page, paste it into Paste Text. The page checks it as you type or paste; you can also click Clean Text.
2. Read Removed characters and Changed categories. The report also shows original and clean lengths, counted as Unicode code points. The category tiles tell you which supported groups changed.
3. Compare Clean Result with the source. If the edits are right for your document, click Copy. The tool needs no login or upload; the pasted text is processed in your browser, not sent to a server for cleaning.

A small hidden-character example

The bracketed notation below names a code point; it is shown visibly so you can understand an otherwise invisible edit. It is an illustration, not evidence that Claude inserted that character.

U+200B is a zero-width space. In this example the report would show one removed character under Zero-width characters. A visual scan of the unannotated text might miss it. For a hands-on check of the report, click Load Sample on the tool page; that sample includes several supported character categories.

What the count can and cannot tell you

The cleaner checks each pasted Unicode code point against a defined list: zero-width characters, bidirectional controls, tag characters, variation selectors, private-use characters, soft hyphens, and em dashes. It removes matches and counts them locally. This is a practical check for those characters in the exact text you pasted, not a universal scan for every possible invisible character.

A positive count does not establish where a character came from. It could have entered during copying, editing, formatting, or from another source. A zero count does not establish that Claude did or did not write the text. Claude's embedded text watermark is described by Anthropic as adding no hidden characters, so this report cannot test for that mark.

Before you use the cleaned version

• Check em dashes: the tool removes them too, so a sentence may need a replacement space or punctuation.
• Check emoji and non-English text: variation selectors and bidirectional controls can affect display. Review these passages against the original.
• Keep your source copy: if an edit changes meaning or layout, use the original to restore it.
• Proofread the destination: a document editor or publishing system can apply its own rich-text formatting after you paste.

Quick questions

Related guide: Claude watermark remover vs AI detector.