Read the report as character evidence
A useful hidden-character check should identify what it found. “Narrow no-break space, U+202F, count 3” is something you can verify in another Unicode inspector. A label such as “AI text detected” would claim more than the same evidence supports.
The cleaner inspects your original input first. It then applies the selected settings and shows a separate output. A character can be marked Removed, Replaced, or Kept. Kept entries can be expected: for example, some joiners and presentation selectors form part of supported emoji sequences.
- Paste a sample into Input. The text is processed locally in your browser.
- Read each reported code point and its action. Consult the Unicode character reference if its purpose is unclear.
- Compare the cleaned output with the source. Copy only after checking that the changes suit your document.
Three results and what they mean
A special space is replaced
An entry for U+202F means the sample contains a narrow no-break space. Replacing it with U+0020 can change line wrapping. It does not establish whether the source was a chatbot, a typeset document, or another copy-paste step.
A zero-width character is removed
An entry for U+200B can explain why a visible word failed an exact search. Removing it may resolve the mismatch. Other join controls can be meaningful in writing systems, so avoid treating every invisible character as unwanted.
The report is empty
The supported checks found no matches. This is a limited result from a defined rule set. The tool is not an exhaustive Unicode security scanner, a homoglyph detector, or an authorship classifier.
Why a character checker cannot verify every watermark
A scheme that inserts removable characters is different from a scheme that changes how words are selected. Google DeepMind describes SynthID-Text as watermarking generated text through token selection. Inspecting code points alone does not verify such a signal.
Likewise, a normal space or em dash does not establish a watermark. The Unicode General Punctuation chart documents these as ordinary characters with legitimate uses. For a practical cleaning workflow, see ChatGPT watermark claims and hidden-character removal.
Choose settings for the text you actually have
AI Clean is intended for prose cleanup and can change punctuation, compatibility characters, and join controls. Open Advanced Controls to disable transformations you do not want. Script Join Controls and BiDi Controls deserve particular attention for Persian, Indic, Arabic, Hebrew, or mixed-direction text.
For code, identifiers, or scientific notation, compare the complete output against your original. A character cleaner is a troubleshooting aid; its output should not be treated as proof that a command or received document is safe. The methodology page describes the order of operations, and data handling details explain the browser processing and site analytics.