Skip to content
Hidden Character Inspector

Private browser-local Unicode inspection

Hidden-character inspection, not an AI detector

Inspect hidden characters in AI text

Find zero-width characters and copy/paste artifacts, remove selected items in your browser, and see every change. This does not remove SynthID or other statistical watermarks, and recognized Content Credentials are protected.

Private by design: Your text is processed in this browser and is not uploaded.

“AI text watermark” can mean different things

The workbench handles only the first row. The distinction is part of the product, because a successful hidden-character inspection cannot answer a question about another mechanism.

MechanismWhat it isLaunch behavior
Invisible Unicode or copy artifactA literal character such as a zero-width space, soft hyphen, control or stray byte-order mark.Reveal, classify and offer selected conservative cleanup.
Statistical watermarkA generation-time token-selection signal, such as SynthID Text.Not inspected or removed.
Content CredentialA C2PA provenance structure embedded in or wrapped around text.Structurally checked; possible credentials stop every transform.
File metadataProperties, XMP, comments or other data in a document container.Unsupported; DOCX and PDF are not accepted.
AI-authorship classifierA model or score estimating how text was written.Unsupported and never inferred from hidden characters.

What you can inspect

Reveal every occurrence

See an inert visible token, Unicode identity and exact location for each bounded finding.

Choose each deletion

Review conservative suggestions. Language, emoji and provenance-sensitive controls stay preserved.

Verify the change

Compare before and after, then read a receipt that accounts for every removed scalar.

Example reveal

exam⟦U+200B ZERO WIDTH SPACE⟧ple

The token makes the invisible scalar reviewable. Its presence does not say that the text came from AI.

The cleanup boundary

The tool can: inspect pasted text and local UTF-8 text/Markdown files, reveal hidden characters, apply selected permitted deletions and generate local output plus a change receipt.

The tool cannot: remove SynthID, validate or strip Content Credentials, decide whether writing is human or AI, process PDF/DOCX metadata, paraphrase text or guarantee a detector result.

No normalization is applied. Possible provenance stops processing before cleanup, and every selected deletion is revalidated in a separate local worker run.

Understand the result before using it

What is an AI text watermark?

Separate hidden Unicode from statistical signals, provenance, metadata and classifiers.

Read the taxonomy

Why SynthID is different

Learn why a token-selection watermark is not a run of removable invisible characters.

Read about SynthID Text

Why provenance is protected

See what the bounded C2PA structural preflight checks and what it does not validate.

Read about Content Credentials

Which characters stay

Review the conservative policy for bidi, joining, emoji, variation and unknown controls.

Read the Unicode guide