Remove ChatGPT's hidden characters and 'watermarks'
In April 2025, researchers noticed that ChatGPT models were inserting narrow no-break spaces (U+202F) into long outputs — invisible characters that survive copy-paste and mark text as AI-generated. OpenAI called it a quirk of training rather than a watermark, but the lesson stuck: AI text carries characters you cannot see.
CopyCleanse scans your text for every hidden character class — zero-width spaces, byte-order marks, narrow no-break spaces, soft hyphens — shows you exactly where each one sits, and strips them all. Use scan-only mode to check text without changing it.
Free, no sign-up, nothing leaves your browser.
Frequently asked questions
Does ChatGPT watermark its text?
OpenAI says no deliberate Unicode watermark exists, but models have shipped outputs containing unusual invisible characters like U+202F narrow no-break spaces. Whatever the intent, removing them makes your text clean and portable.
Will removing hidden characters make my text undetectable to AI detectors?
No. Detectors like GPTZero analyze writing style statistically, and Google SynthID watermarks token probabilities, not characters. CopyCleanse removes character-level artifacts — hidden Unicode, formatting tells, and tracking links — it does not rewrite your prose.
Is my text uploaded anywhere?
No. All scanning and cleaning runs in your browser. Nothing is sent to a server, logged, or stored.