Your AI text still carries traces. Find them and remove them all

How it works

1. Paste or upload

   Drop in an answer from ChatGPT, Claude, Gemini or Copilot. The scan runs on your device the moment the text appears.

2. See every trace

   Each finding is marked in red right inside your text, with its Unicode code and a one-line reason it gives the text away.

3. Remove and paste

   Invisible characters go free. Pro removes the rest down to zero, re-checks the result and copies it ready for Word, Google Docs or email.

What the scanner looks for

Seventeen groups of traces in text, plus Word metadata in documents. Each one is a real character or fragment we can point to in your text — never a guess about who wrote it.

• Shadow watermarks

  Pro
  U+200B

  Empty invisible characters with no width that let scanners spot a copied AI answer instantly.

• Script-assembly traces

  Free
  U+FEFF

  A system encoding marker in the middle of a phrase shows the text was put together by a script, not written by a person.

• AI clipboard markers

  Free
  U+200D

  Hidden joiners that travel with text copied from a chatbot window and give away its source.

• Machine spacing matrix

  Free
  U+00A0

  Specific indent bytes from the AI training base that physically cannot be entered during ordinary manual typing.

• Direction marks

  Free
  U+200E

  Text-direction controls in left-to-right text — invisible and easy to detect.

• Hidden Unicode tags

  Free
  U+E0000

  Invisible tag characters that can spell out hidden instructions for AI.

• Variation selectors

  Free
  U+FE0F

  Selectors outside emoji can carry hidden data in otherwise plain text.

• Hidden layout commands

  Free
  U+00AD

  Invisible system hyphens and formatting codes that physically cannot be typed on a keyboard.

• Hidden line breaks

  Free
  U+2028

  Unicode line and paragraph separators that editors treat differently.

• Markdown leftovers

  Free
  ** ##

  Asterisks, hash headings and pipe tables are chat formatting, not document formatting.

• Link tracking tags

  Pro
  utm_source

  utm_source=chatgpt.com names ChatGPT as the source of the link.

• Citation artifacts

  Pro
  oaicite

  Leftovers like oaicite and 【4:0†source】 appear only in answers copied from AI.

• LLM syntax signature

  Pro
  U+2014

  A hard join of words with a special character, whose mathematical frequency detectors use to work out the structural handwriting of a language model.

• Training-dataset artifacts

  Pro
  U+2015

  Internal markup codes from AI databases that are absent from the keyboard layout and prove machine output.

• Regional token conflict

  Pro
  U+201C

  The neural network inserts paired bytes from its base dataset against the language’s norms, leaving a trace of generation that is obvious to scanners.

• Compressed monolithic tokens

  Pro
  U+2026

  The algorithm packs a group of marks into a single Unicode glyph to save tokens, giving away an automatic text dump.

• Unicode spoofing and glyph swaps

  Pro
  U+0430

  A mix of bytes from different language tables inside words triggers instant alarm in verification and anti-spam systems.

• Document metadata

  Pro
  docProps

  Author, company, app, template and edit history stored inside the Word file.

Before and after

Plus two invisible characters you cannot see in either version. The scanner shows them as code labels.

Paste it where it needs to go

Markdown from the chat turns into real formatting for the place you paste into.

• Word

  Headings, lists, tables, bold and links arrive as real Word formatting.

• Google Docs

  Same structure, tuned for how Docs handles pasted HTML.

• Email

  Simple formatting that survives Gmail and Outlook, without big headings or tables.

• Plain text and Markdown

  Clean text with paragraphs and lists kept, or tidy Markdown without hidden characters.