Direct text rewriting and formatting cleanup
This Claude watermark remover gives you a direct editing workflow for Claude-assisted text. Paste a draft, choose how extensively it should be rewritten, protect the details that must stay exact, and compare the new version with the source. The tool changes wording and structure; it does not claim to verify that a model-level watermark existed or was removed.
Claude text rewriting workbench
Direct rewrite · no detector required
10 free AI rewrites every 24 hours · Checking remaining…
Rewrite Text sends your passage to the configured AI provider. Clean Formatting stays in this browser.
Compare every change, copy or download the result, regenerate with the current settings, or restore the source.
Start by pasting the passage you want to edit. It can be an article, email, report, Markdown document, technical note, Claude Code output, JSON example, or a draft that Claude helped you revise. The character and word counters update in your browser, and the original remains available while you work. You do not need an account, and you do not need to run a detector before using the rewriting controls.
Next, choose Light, Balanced, or Strong. Light makes restrained wording improvements. Balanced, the default, restructures sentences and some paragraphs while keeping the complete meaning. Strong changes the expression and organization much more substantially. Tone, target length, and preservation settings let you describe the edit without asking the model to invent facts. Citations, URLs, numbers, dates, code, and JSON can be kept explicit in the rewrite instructions.
Selecting Rewrite Text sends the passage and your settings to the configured non-Claude editing provider. Substantial rewriting changes the visible token sequence, phrasing, sentence structure, paragraph flow, and statistical properties of the passage. Anthropic states that heavy editing or paraphrasing may make a mark undetectable, but no public third-party mechanism currently exists to verify whether a particular rewrite removed the model-level mark. The result is therefore presented as a rewritten draft, not as proof of removal.
Review the output beside the source or through the unified diff before using it. Strong edits deserve especially careful checking because large structural changes can affect nuance even when the instructions require factual preservation. Confirm names, figures, dates, quotations, links, citations, commands, and technical constraints yourself. Copy the approved result, download it as plain text or Markdown, regenerate it with the same settings, or restore the original and try a different approach.
Editing controls
Light is intended for material that already says what you mean. It corrects clear grammar problems, reduces repetition, and adjusts a limited number of sentences while retaining paragraph order, voice, and most original wording. It suits emails, careful copyediting, and drafts where even small changes must remain easy to audit.
Balanced is the default because it produces a visibly fresh expression without rebuilding every paragraph. It can split, combine, and rephrase sentences and reorganize selected passages for flow. Facts, meaning, names, figures, dates, quotations, URLs, and technical content remain protected instructions rather than optional suggestions.
Strong changes sentence patterns and paragraph organization across the draft. It is useful when a long passage needs a substantially different expression, but it also requires the closest review. Check every fact, number, citation, quotation, code sample, and technical limitation before publishing or relying on the result.
Formatting cleanup is a separate, local operation. It can normalize inconsistent line endings, trim trailing whitespace, replace selected Unicode spaces, remove soft hyphens and conservative zero-width spacing controls, review bidirectional controls, and apply NFC normalization. Code fences, inline code, URLs, email addresses, citations, valid JSON, meaningful joiners, variation selectors, and emoji sequences are protected where possible. Higher-risk changes appear in a preview before you apply them, and you can undo the result or restore the source.
Hidden Unicode and a model-level watermark are not the same thing. Formatting characters can come from document editors, websites, email clients, multilingual text, or copying between applications. Removing one does not establish that Anthropic placed it, and cleaning formatting does not verify or remove a statistical mark. You can use Clean Formatting without calling the AI rewrite service at all.
Rewriting and verification are separate claims. This page keeps the boundary visible so that an editing result is never presented as an authorship or watermark verdict.
Frequently asked questions
It sends the text and your selected editing controls to the configured non-Claude rewrite provider. The rewrite changes wording, sentence construction, paragraph flow, and other statistical properties, while the preservation settings instruct the editor to retain important details. It is not an official Anthropic detector and does not report that a model-level watermark has been confirmed or removed.
No independent site can make that promise with the public tools currently available. Anthropic explains that heavy editing or paraphrasing may make a mark undetectable, but there is no public third-party verifier that can confirm removal from a particular passage. Treat the output as a revised draft and review it for quality and accuracy rather than as a verified removal result.
The page is currently free and does not require registration. Anonymous visitors receive ten successful AI rewrites during a rolling 24-hour window, and the tool shows the remaining allowance before submission. Formatting cleanup runs locally and does not consume an AI rewrite. Reasonable safeguards may pause unusually frequent automated attempts.
The phrase “removal” describes the editing intent, not a verified watermark outcome. For Rewrite Text, the passage and selected settings are sent to the configured AI provider, then the result is returned for comparison; the site does not place the passage in the URL or send its body to analytics. Local formatting cleanup stays in the browser. Review facts, citations, numbers, dates, URLs, code, and JSON before using any rewritten draft.
Yes. The controls work for many kinds of drafts, including emails, articles, reports, documentation, Markdown, and technical material. Use the preservation options, compare the diff, and make your own final editorial decisions. You remain responsible for appropriate disclosure and for following workplace, publishing, academic, and platform rules; the tool does not promise to bypass any detection system.
Anthropic has not yet released a public third-party watermark detector or verifier. This tool rewrites text to change its statistical properties. It cannot independently verify the presence or removal of a model-level watermark. Heavy editing or paraphrasing may make a mark undetectable per Anthropic's documentation, but no mechanism exists to confirm removal.
← Back to Claude Watermark Detector