Strip Claude's <thinking> extended reasoning blocks, markdown formatting, and hidden Unicode — clean output in one click.
Cleaned text appears here
Paste input → click Clean Text
Stats appear after cleaning
0
Total fixes
0
Clean runs
Your breakdown appears after your first clean →
What gets cleaned
Anthropic Claude exposes <thinking>…</thinking> reasoning blocks when extended thinking is enabled, wraps structured content in XML artifact tags, and uses heavy Markdown throughout. TextPurify removes all of it so you get clean, paste-ready prose.
Claude's extended thinking feature generates internal reasoning wrapped in <thinking>…</thinking> tags. These are hidden in Claude.ai but appear verbatim when copying from the API or certain third-party apps.
<thinking>Let me reason through this…</thinking>Claude uses ## headings, ### subheadings, - bullet lists, and **bold** extensively. These render as raw symbols outside Claude.ai in email, Word, or plain-text editors.
Claude outputs typographic "smart quotes" which fail in JSON, HTML attributes, code strings, and some database fields that expect straight ASCII quotes.
Claude uses the em-dash (—) as a structural separator and the Unicode ellipsis character (…), both of which cause compatibility issues in plain-text contexts and string comparisons.
Invisible characters embedded in Claude responses cause broken word counts, encoding errors in downstream systems, and unexpected search/replace failures.
Claude wraps code, documents, and structured output in <antArtifact> XML tags with metadata attributes. TextPurify strips the tags while preserving the content inside.
When Claude's extended thinking mode is enabled (available on Claude 4 Sonnet and Opus), it generates an internal reasoning chain wrapped in <thinking> tags before the final answer. These blocks are hidden in the Claude.ai interface but appear in full when you access Claude via the API or copy from some third-party apps.
Paste your Claude text into TextPurify. The Claude preset has 'Remove <think> blocks' enabled by default, which removes any <thinking>…</thinking> content. Click Clean Text and only the final answer remains.
No. The <thinking> block contains Claude's internal reasoning only. The final answer always follows after the closing </thinking> tag and is never removed.
Yes — TextPurify handles output from all Claude model versions: Claude 3 Haiku, Sonnet, and Opus, plus Claude 4 Sonnet and Opus.
Never. TextPurify is entirely client-side. Your text stays in your browser and is never transmitted anywhere.
More AI text cleaners