Text copied out of ChatGPT is rarely plain. It carries non-breaking spaces, curly quotes, em dashes, the ellipsis character and sometimes characters you cannot see at all. StripShot strips the hidden ones, normalizes the typographic ones if you want, points out the phrases that read as generated, and never sends a word to any model.
Last updated September 2026
Paste ChatGPT output below and click Scan. StripShot lists hidden characters (zero-width, non-breaking spaces, directional marks), typographic punctuation and look-alike letters, shows where they sit, and lists the writing tells. Strip removes the characters. You edit the words.
Scans are always free. Cleaning counts toward the daily free strips; Pro is unlimited.
| Character | Code point | Why it is there | Action |
|---|---|---|---|
| Narrow no-break space | U+202F | Documented in output from some models in 2025; typographic spacing elsewhere | Replaced with a space |
| Non-breaking space | U+00A0 | Typographic spacing, common around numbers and units | Replaced with a space |
| Curly quotes | U+2018, U+2019, U+201C, U+201D | Smart typography | Straight quotes (option) |
| Em dash, en dash | U+2014, U+2013 | House style of most chat models | Hyphen (option) |
| Ellipsis character | U+2026 | One character instead of three periods | Three periods (option) |
| Zero-width space, joiner, word joiner | U+200B, U+200D, U+2060 | Copy residue or deliberate fingerprinting | Removed |
| Directional marks | U+200E, U+200F | Appear around mixed-direction text | Removed |
| Look-alike letters | Cyrillic and Greek | Copy residue or deliberate substitution | Latin letter (option) |
After a scan, StripShot lists phrases and patterns that show up disproportionately in generated prose: rich tapestry, delve into, in today's fast-paced world, a testament to, chatbot openers, along with sentence-length variation and em dash density. It is a style read that helps you find what to edit. It is not a detector verdict, it does not rewrite anything, and a low score is not proof of anything. Cleaned text is still text you are responsible for disclosing where disclosure is required.
Cleaning characters is hygiene. It makes text behave like text. It does not change authorship, and it does not remove a statistical watermark, because those live in the choice of words. StripShot will not rewrite your text and will not send it to a model to be paraphrased. If you want to know exactly what each vendor embeds and what can be removed, read the AI text watermark guide.
Paste the output
Copy from ChatGPT and paste into the box above. The counter shows how many hidden characters it already sees.
Scan
Click Scan text. You get a list of every hidden and typographic character with its code point, a highlighted view of where each one sits, and the writing-tells panel.
Choose options
Keep or drop punctuation normalization and look-alike fixing. Invisible characters and unusual spaces are always cleaned.
Strip and copy
Click Strip. The cleaned text is re-scanned and shown. Copy it, then edit the flagged phrases yourself if you want to.
Sometimes. In April 2025 several people documented narrow no-break spaces (U+202F) appearing in output from some ChatGPT models. OpenAI described it as a side effect of training rather than a watermark and said it would be addressed. Independently of that, pasted ChatGPT text routinely carries non-breaking spaces, curly quotes, em dashes and the ellipsis character, which are typographic rather than hidden but still trip up exact-match searches, code and plain-text systems.
OpenAI has said it built a text watermarking method and has not deployed it publicly. ChatGPT images are a different matter: they carry C2PA Content Credentials, which StripShot's photo tool removes. For text, what this page removes are characters, not a statistical watermark.
No, and StripShot does not claim that. Detectors and statistical watermarks work on word choice and sentence structure. This tool changes characters, not words. The writing-tells panel shows you the phrases detectors key on so you can edit them yourself if you want to.
Because they are a style choice that many people do not want, they break plain-text formats and code, and heavy em dash use has become a widely noticed tell. Normalize punctuation is an option; turn it off to keep them.
Cyrillic а, е, о, р, с and Greek ο, α look identical to Latin letters but are different characters. They appear in copied text from some sources and in text that was deliberately altered to slip past filters. Words containing them fail spellcheck, search and matching. The cleaner swaps them back.
No. Everything runs in your browser tab. StripShot never sends text to a language model, never rewrites it and never stores it.
Yes. The cleaner is vendor-neutral. Paste any text. The AI text watermark guide covers what each vendor does and does not embed.
What Claude, Gemini and ChatGPT embed and what can be removed.
The full list of hidden code points and how they are handled.
The C2PA credentials on DALL-E images.
When ChatGPT is the listed author of a Word file.
Pricing
No card needed to start. Upgrade anytime.