A hidden character is not proof of AI origin
People often use AI watermark as shorthand for zero-width spaces, soft hyphens, direction controls, unusual spaces or punctuation that appeared after a copy and paste. Those are real Unicode characters, but their presence does not identify the app or person that produced the text.
The Unicode Standard gives these characters ordinary text-layout jobs. U+200B can mark a word or line-break opportunity; U+00AD marks a possible hyphenation point; bidirectional controls help mixed left-to-right and right-to-left text display in the intended order. A zero-width character may be unwanted in one English paragraph and essential in another language or sequence.
Start by diagnosing, not deleting
Keep an untouched copy, then scan the text and identify each code point before removing it. The useful question is not Is this invisible? but Does this character serve a purpose here? Check words that will be searched, identifiers that must match exactly, URLs, code and passages using Arabic, Hebrew or another script with directional behaviour.
In Pasteful, enable only the cleanup categories you need and inspect the before-and-after diff. Removing a stray zero-width space from an English identifier may fix a match. Blanket removal across an entire multilingual document can change joining, line breaking or display order.
- Preserve an original copy
- Inspect the code point and its surrounding text
- Exclude code and identifiers unless the character is confirmed wrong
- Review right-to-left text with a fluent reader
- Copy only after the diff matches the intended change
Open the invisible-character remover · Read Pasteful's cleanup rules
Do not confuse cleanup with provenance
Text provenance needs evidence tied to the source, not a guess based on typography. OpenAI's current provenance documentation lists supported signals for images and audio and describes text provenance as a future goal. It does not present stray Unicode characters as a verifier for ChatGPT text.
Removing a character therefore cannot establish human authorship, defeat a reliable detector or erase a documented source record. It only changes the text in front of you. Treat claims such as guaranteed undetectable or proof that a document is human-written as unsupported.
Use the smallest safe cleanup
If the problem is a broken search, an unexpected line break or text that fails validation, remove only the confirmed artifact and test the exact destination again. Keep typographic punctuation when it is intentional; a curly quote or em dash is not hidden and does not need to be normalised merely because it may have come from an AI draft.
Pasteful performs the selected cleanup in the browser tab and shows the changes before export. That privacy boundary does not turn the result into authorship evidence. It simply makes a mechanical repair inspectable.
Review the local-processing privacy model · Replace punctuation only when needed
Paste it into Pasteful. It stays in this tab.