Looking for a TextGrain remover? Start with the right distinction.
A useful removal tool must identify what it changes and what its result can establish. TextGrain Remover offers a working Unicode character cleaner and evidence-led guidance about statistical watermarks. The cleaner changes selected code points in a string. It does not test whether a TextGrain signal is present, and it does not certify that an OpenAI watermark has been removed. You can use it as a guest, inspect every flagged category and compare the output before copying.
That distinction matters when pasted text contains unexpected spacing, word-boundary markers or invisible formatting controls. These are concrete editing problems with a before-and-after result. A statistical signal in token choices is a different problem. A text can contain hidden characters without being AI-generated, or have a statistical watermark without containing the characters our cleaner flags. Keep those two questions separate when choosing a tool or evaluating a claim.
What each method can actually establish
| Method | Useful for | Not evidence of |
|---|---|---|
| Selected Unicode cleanup | Removing a flagged zero-width space or replacing an unusual space with an ordinary one. | Statistical watermark removal or human authorship. |
| Copying as plain text | Removing rich formatting that an editor carried along. | Erasing a pattern in unchanged words. |
| Rewriting or translation | Changing wording for a legitimate editing task, with a meaning review. | A guaranteed watermark-free result. |
| A matching watermark detector | Assessing evidence for the particular signal it is configured to detect. | The writer's identity, ownership or factual accuracy. |
| A general AI classifier | A classifier-specific assessment with its own error rates. | A TextGrain result or proof about a secret-key watermark. |
If your problem is strange pasted characters, the tool above gives you a specific, inspectable change. If your question is whether a passage carries TextGrain, read the detector access and limitations guide. If you are comparing broader removal claims, the methods comparison separates character cleanup from statistical-watermark handling.
What a trustworthy removal result needs
A claim about removing a particular watermark needs more than a cleaner's success message. You need to know which watermark was generated, whether the detector matches that setup, and what was measured before and after editing. Record the detector's assumptions and decision threshold. A score from an unrelated classifier cannot fill those gaps. A service that cannot name its test should not present a percentage as a verified TextGrain removal rate.
Compare changes in meaning as well as changes in the measured signal. Replacing words can alter a qualification, a quotation, a technical term or a factual statement. A lower detector score is not evidence that the edited passage is accurate or acceptable for your purpose. This site does not offer a paid rewriting model, a promise to bypass every detector or an automated decision about who wrote your text.
Our independently testable claim is narrower: selected characters change according to the categories you choose, while the original input remains available for comparison. The scan reports code points, counts and positions. Joining controls and variation selectors remain intact. The resulting text is not sent to an AI provider or a server-side detector.
How to Handle a Suspected TextGrain Watermark
- Cleaning a pasted document: inspect the flagged categories, keep uncertain characters, and replace unusual spaces rather than joining words together.
- Checking a claimed watermark: establish the originating product, date and configuration before interpreting a detection result. Do not assume every ChatGPT response has the same provenance setup.
- Reviewing an attribution dispute: combine the relevant detector's limits with contextual evidence and a human review. Our tool does not decide authorship, disclosure duties or ownership.
- Following implementation changes: consult the dated source links and the open-source release page. An announced plan is not an available repository or a tested installation procedure.
For a technical explanation, use What Is TextGrain?. For product and region questions, use Does ChatGPT Watermark Its Text?. These pages have different jobs: the overview helps you choose an action; the principles guide explains the mechanism; the detector guide explains what a result can and cannot mean.
TextGrain removal questions
Does this site remove the TextGrain statistical watermark?
No verified statistical-watermark removal capability is offered. The available cleaner removes or replaces explicitly selected Unicode characters. Its output names those changes; it does not label the result “TextGrain removed.” Use the tool for character cleanup, not as a substitute for a matching watermark test.
Does removing U+200B prove that text was watermarked?
No. U+200B is a zero-width space used in many text workflows and writing contexts. Its presence is not proof of AI generation. Removing one demonstrates a character edit, not a conclusion about the text's origin. The same caution applies to non-breaking spaces, soft hyphens and other formatting controls.
Can copy and paste alone remove a statistical watermark?
Copying the same words does not by itself change their statistical relationships. Some clipboard and editor workflows alter formatting or characters, but that is a different kind of change. Do not read a plain-text paste, an invisible-character count or a changed file format as a verified watermark-removal result.
Why preserve emoji joiners and other invisible characters?
Invisible does not mean unnecessary. A joiner can combine emoji into one displayed symbol, and a non-joiner can affect a writing system's spelling or appearance. Variation selectors also affect glyph presentation. We flag these categories for inspection but never remove them. Ordinary newlines, tabs and indentation remain unchanged.
What happens to special spaces?
Special-space normalization is off initially. If you choose it, supported special spaces become ordinary spaces rather than disappearing. That keeps adjacent words separated, although layout and non-breaking behavior can change. Review the comparison before copying, particularly for quotations, code, tables and multilingual documents.
Is the cleaner free, and where is my text processed?
Guests can inspect, clean and copy up to 100,000 UTF-16 code units per input. Processing happens in page memory in your browser. This site does not upload or store the entered text. Reloading or closing the page clears it. Optional analytics require consent and never receive the text itself.
Can a “not detected” result establish human authorship?
No. Interpret any result under the specific detector's scope and assumptions. Unrecognized, short, edited or differently generated text may not produce a usable signal. A negative result is not a reliable declaration about the writer. Our character cleaner does not produce a watermark detection result in the first place.
Does open-source code automatically unlock production detection?
No. An implementation, the key used for your own experiment and access to a production detector are distinct things. A public repository would not on its own disclose another system's production secret key. Verify the official repository, license, configuration and detector access separately before relying on an installation claim.
Sources, editorial scope and corrections
This independent site separates official statements from its own implementation tests. Product availability is based on the dated OpenAI announcement; technical explanations cite the technical report. The access guide distinguishes public image/audio verification from approved-organization text access.
Reviewed October 6, 2026. Tool behavior is checked independently using ordinary text, emoji, joining controls, special spaces and HTML-like input. We do not turn example edits into customer testimonials or a detector benchmark. Send a specific statement and attributable source through the contact page when reporting a correction. TextGrain Remover is not affiliated with OpenAI and cannot grant official access.