Claude Text Cleaner

Mixed-direction text guide · October 29, 2026

Does Claude Text Cleaner Remove Bidi Isolate Pairs?

Yes. The browser cleaner deletes U+2066, U+2067, U+2068 and U+2069 from pasted text, counting each under “Bidi controls.” If text contains U+2067 RIGHT-TO-LEFT ISOLATE around a phrase and U+2069 POP DIRECTIONAL ISOLATE at its end, both marks are removed. The letters remain, but the intended display order around punctuation or numbers can change. Keep the original when the direction controls are intentional.

Try the browser cleaner

Here “Hebrew phrase” is שלום (four code points). The labels show logical order; actual visual order depends on the surrounding text and application.

Why does the pair matter?

W3C's explanation of bidirectional controls shows that U+2067 starts an isolated right-to-left run and U+2069 closes it. The Unicode Bidirectional Algorithm also defines U+2066 (left-to-right isolate) and U+2068 (first-strong isolate). Unlike the single U+200E or U+200F direction marks, these controls delimit a range. Deleting them can let punctuation and nearby text interact differently with the right-to-left phrase. This cleaner applies a fixed code-point rule; it does not reconstruct the intended layout.

A public English-language question about embeddings and isolates asks what these controls do and requests examples. That discussion is about Unicode generally, not a report of this site's behavior.

How to review a cleaned result

  1. Save the original. Keep a copy of mixed Hebrew, Arabic, English or numeric text before cleaning.
  2. Check the report. Paste a copy into the browser cleaner and look at “Bidi controls.” The count includes isolate marks and other supported direction controls; it does not identify which ones were present.
  3. Compare at the destination. Inspect punctuation and numbers in the app where the text will be used. If the isolated phrase must keep its display order, use the original or add direction handling appropriate to that destination.

For the labeled sequence above, the production cleaning rule removes U+2067 and U+2069, leaving six of eight Unicode code points and counting two Bidi controls. This is a rule-derived example, not a claim that this exact sequence was tested in the live UI. The cleaner processes pasted text in the browser. A removal does not establish text origin, detect a watermark or predict an AI detector outcome. For single marks, see the U+200F guide and U+200E guide.