Clean text from ChatGPT and other AI chatbots

Paste an AI chatbot's answer to see the invisible characters and special spaces that came along with it, then copy a clean version.

Checked in your browser. Nothing is uploaded. Turn off your Wi-Fi and it still works.

Paste your text

Under the light
Inspection report—
    Cleaned text

    Ticked items are cleaned. This changes how the text is stored, not who wrote it.

    How it works

    Short answer

    Text copied from AI chatbots can contain invisible characters and special spaces, such as narrow no-break spaces (U+202F) and zero-width spaces. Paste the text above to see and remove them. This tidies the text but does not remove statistical watermarks, which are hidden in word choice rather than in characters.

    Text: characters that take up no space

    ON SCREEN results IN THE FILE r e s u l t s U+200B · 3 bytes · 0 px wide

    A zero-width space sits between letters but draws nothing. It can break search, word counts and code, and can be used to mark copied text. Tag characters go further and spell out invisible messages.

    What gets cleaned

    By default the cleaner removes three things that commonly ride along with chatbot text:

    • Invisible characters like zero-width spaces, joiners and byte order marks.
    • Special spaces like non-breaking (U+00A0) and narrow no-break spaces (U+202F), replaced with normal spaces.
    • Hidden messages written in invisible Unicode tag characters, if any.

    Two optional style fixes are off by default: straightening curly quotes, and replacing em dashes with hyphens. Tick them if you want text that reads less like a chatbot. For dashes specifically, the em dash remover has tips on better replacements.

    What it doesn't do

    Some AI companies now watermark text statistically. Google has done this for Gemini with SynthID since 2024, Anthropic has watermarked Claude's text worldwide since August 2026, and in October 2026 OpenAI announced a text watermark for ChatGPT users in the EU. These watermarks work by nudging which words get chosen. There's nothing to delete, so no character cleaner can touch them.

    That's also why we don't promise to beat AI detectors. If you used AI to help write something, the honest fix is to edit it into your own words and say how you used it where that's expected.

    Why clean AI text at all

    Invisible characters cause real problems once text leaves the chat window: broken searches and word counts in documents, failed form fields, odd spacing in emails, and errors in code or spreadsheets. Cleaning makes the text behave like text you typed.

    Questions

    Does ChatGPT put hidden characters in its answers?

    Sometimes odd characters appear. In 2025 users found narrow no-break spaces in long answers from some OpenAI models. OpenAI reportedly said they were a side effect of training, not a watermark, and they later stopped appearing. Copying from any chat interface can also bring non-breaking spaces and other formatting.

    Will this remove the ChatGPT watermark?

    No tool can remove a statistical text watermark by deleting characters, because it isn't made of characters. It's a pattern in which words were chosen. Our guide on whether ChatGPT watermarks text explains the details.

    Will cleaned text pass AI detectors?

    We don't claim that, and you shouldn't trust any tool that does. Detectors judge writing patterns. Cleaning only changes how the text is stored.

    Does this work for Gemini, Claude and Copilot text?

    Yes. The cleaner looks for characters, not for a particular chatbot, so it works on text from any source.