Invisible Unicode Characters Guide
Learn to find and remove invisible Unicode characters.
NOCLAU GUIDES
Text cleaning guides
Learn how hidden Unicode characters and formatting affect your text.
Zero Width Spaces Guide
Understand and remove zero width spaces safely.
Unicode Control Characters Guide
Identify Unicode control and formatting characters.
Em Dashes in AI Writing
When and how to replace Em Dashes in text.
Claude Text Formatting Guide
Review copied assistant text, inspect characters, and preserve document meaning before publishing.
Preserve Emoji and Multilingual Text During Cleanup
Compare cleaning modes without damaging joined emoji, connected scripts, or mixed-direction text.
Copy and Paste Text Troubleshooting
Use a three-sample comparison to investigate text mismatches, isolate changes, and validate a repair.
Choose a text-cleaning workflow
Choose a guide by the problem you are trying to solve, then use its example and acceptance check with a small working copy of your text. This library separates character detection, deliberate cleanup, and publishing review. A finding is useful when it helps explain an observed behavior, such as a failed lookup or an unexpected line break. It is not a score for writing quality, authorship, or the likelihood that a passage came from an AI assistant.
Scroll the table sideways to see every column.
| Your task | Start here | Acceptance check |
|---|---|---|
| Exact-match failure | Copy-and-paste troubleshooting | Repeat the lookup after saving |
| Known U+200B | Zero-width spaces | Check expected word or identifier |
| Direction or control finding | Unicode control characters | Review the intended display or format |
| Several languages or emoji | Preservation guide | Language and destination review |
| Copied assistant draft | Claude text formatting | Check content, structure, and selected edits |
Start with the symptom
If two values look identical but do not match, start with the copy-and-paste troubleshooting guide. It walks through a typed sample, a pasted source sample, and a deliberately changed copy. That comparison helps locate which step changes the text. A reproducible mismatch is a better starting point than applying every cleaning rule to an entire document and hoping one of them works.
If you already see a U+200B finding, open the zero-width-spaces guide. It explains when deletion is appropriate and when an ordinary word boundary needs to be inserted manually instead. If the character type is still unclear, begin with the invisible-Unicode guide, which separates hidden separators, unusual spaces, joiners, and direction-related marks. Each page leads to a tool suitable for its specific task.
Follow an inspection-first workflow
Keep the original somewhere outside the tool, then inspect a small copy. Read the finding's code point and surrounding words. Ask what behavior you expect in the destination. A hidden mark inside a reference number may be accidental, while a joiner inside an emoji may belong there. Those findings should not trigger the same edit simply because neither mark is easy to see.
Inspection pages preserve input when you press Analyze text. Removal pages generate a proposed output based on explicit rules. Use the report to choose an edit and the diff to verify what it changes. The final test happens in the receiving application, where you should repeat the original lookup, paste, import, or layout check. Tool output is a candidate result rather than automatic acceptance.
Use the guides for different editing decisions
The control-characters guide is for direction, imported controls, and cases where automatic deletion would be difficult to justify. It explains why noclau reports selected controls while preserving them. Use it when a mixed-language paragraph displays unexpectedly or a restricted field contains a control. For code and structured data, follow the project's normal parser, editor, and validation workflow too.
The em-dash guide covers editorial choices rather than a promise about AI detection. It compares keeping punctuation, using a mechanical replacement, and rewriting a sentence so its relationship remains clear. The Claude-formatting guide connects inspection to publishing: source format, plain versus rich text, independent punctuation choices, and final review. It does not assume that a source name identifies how a character entered the passage.
Protect emoji and multilingual content
Before cleaning a multilingual document, read the preservation guide and create a representative sample. Include joined emoji, connected-script words, and mixed-direction punctuation that occur in the real work. Conservative mode preserves several joiners, but it is still a fixed rule set. It cannot decide whether an intentional line-break opportunity or soft hyphen should remain in a particular language.
If you do not read the language, ask a knowledgeable reviewer to compare the original and changed text. Provide code-point details as well as screenshots. A visually attractive result may still contain an incorrect word, and a lower finding count may simply mean useful characters were deleted. Preserving a detected character is often the correct outcome of a careful review.
Understand what the examples demonstrate
Examples in these guides are constructed to illustrate specific transformations. Bracketed labels such as [U+200B] make invisible characters readable in explanatory text; they stand for actual characters rather than literal strings to paste. Tool example buttons contain real demonstration characters. Save your own input before using a button that replaces it with an example.
The examples are not a dataset of Claude outputs, a measurement of character frequency, or evidence of a proprietary watermark. Expected behavior is tied to the current scanner and cleaning rules. When a guide discusses broader Unicode behavior, relevant standards are linked for further reading. Keep a distinction between a general standard, an application's implementation, and the result of your own destination test.
Know when the library is not enough
The scanner covers selected character categories rather than every Unicode code point. It does not perform general normalization, compare visually confusable letters, validate code, or reconstruct rich-text formatting. A value can still differ because of ordinary whitespace, case, accent representation, or a receiving system's rules. An empty report does not end an investigation when the original failure still reproduces.
For a formatting issue, inspect the editor's paste and style settings. For a data import, validate a small record with the appropriate parser. For an authorship or attribution question, review the document's provenance and relevant disclosure requirements. A character cleaner does not replace those processes. The guides help you choose a bounded next step rather than treating removal as a universal repair.
Build a repeatable review note
When a change solves the problem, record the source and destination, a short invented example, the rule selected, and the test that passed. For an identifier, write the expected exact value. For a paragraph, describe the accepted language and layout. Keep the original until the final result has been reviewed, especially when several people edit the same document.
If a guide's example behaves differently in your browser, use the Contact page to send a minimal reproduction. Include the page address and the reported code point instead of a confidential document. A clear before-and-after record makes both support and future editing easier. It also gives you a practical standard for deciding that cleanup is complete: the intended behavior works and the content still means what it should.