Skip to content
Hidden Character Inspector

Private browser-local Unicode inspection

Literal text, not authorship evidence

Hidden characters in AI text

Copied text can contain characters that take no visible space or have a formatting effect rather than a glyph. They can occur in generated text, but they also occur throughout ordinary human-created documents and software. Their presence alone cannot identify the author or source.

How hidden characters get into text

A web editor may insert a word joiner to control a line break. A rich-text system may retain a byte-order mark when converting formats. A document can use bidirectional controls for mixed right-to-left and left-to-right writing. Emoji can depend on a zero-width joiner or variation selector to display the intended sequence. Copying through several applications can carry those scalars into a plain-text field.

Some characters are also inserted deliberately for layout, identifiers or source-code behavior. The same code point can therefore be harmless in one context and surprising in another. A useful inspector must show the location and context instead of applying one global delete rule.

What can go wrong

Unexpected format controls can make visually identical strings compare differently, break a search, change a slug, confuse a compiler or move the cursor in surprising ways. Bidirectional controls can also alter display order. Those are reasons to review a character; they are not evidence of malicious intent or AI generation.

The workbench reveals each occurrence as an inert token, for example exam⟦U+200B ZERO WIDTH SPACE⟧ple. It reports scalar, UTF-16 and UTF-8 positions, and for imported files it also maps original byte positions. The surrounding text is rendered as text, never interpreted as HTML or Markdown.

Preservation is part of correctness

The inspector does not normalize the input. Joining controls, variation selectors, valid emoji sequences, combining marks and provenance-sensitive characters are preserved. Bidirectional controls and unknown future format characters receive prominent review treatment but are not offered as edits in this launch.

That conservative posture matters for Persian, Arabic, Indic scripts, emoji and any text whose language rules are not obvious to the person running a cleanup. An invisible character is not automatically disposable merely because a screen does not draw it.

Primary references

These sources describe the external standards or mechanisms discussed above. The property’s cleanup decisions remain narrower than those specifications.

Scope reminder: This workbench inspects hidden Unicode and copy/paste artifacts. It does not determine AI authorship, remove SynthID or strip Content Credentials.