Free browser tool
Paste any text to see the invisible characters hiding inside it. The scan names every mark it finds, decodes hidden messages when they are there, and removes only the categories you choose.
Your text stays in this tab and is never submitted.
Scan some text to see the breakdown.
Your text with every hidden character replaced by its code point. Hover a chip to see what it is.
Nothing to show yet. Hidden characters appear here once the scan finds them.
The scan always reports everything it finds. Removal is your choice, one category at a time.
These have no place in ordinary prose. They are the channels used to hide data inside text.
Rarely needed in ordinary writing, so they are removed by default. Check the result if your text is right-to-left.
These characters have real uses in emoji, multilingual text, and typography. They are reported but left in place unless you turn them on.
What a character scan can and cannot do
Any tool that claims otherwise is selling you something. Here is the honest split.
Invisible characters, format controls, and smuggled payloads are countable. The scan names each one, shows you where it sits, and the counts after cleaning are yours to check.
Statistical watermarks live in word choice. The model biases which token it picks, so the signal is spread across the whole text and no character edit reaches it. Rewriting is the only lever, and a rewrite cannot exceed the quality of the model doing the rewriting.
Content Credentials (C2PA), EXIF, and XMP live in image, PDF, and document files rather than in text. Pixel and audio watermarks are a separate layer again. Pasting text here does not touch any of them.
Invisible does not mean harmless
A hidden character is one that occupies a position in the text but renders as nothing. Copy that text anywhere and the character travels with it. Some arrive by accident, pasted in from a PDF or a rich-text editor. Others are placed deliberately: the Unicode tag block and variation selectors each encode a byte per character, which turns ordinary-looking text into a carrier for a hidden message. The scan reports both kinds and tells you which is which.
This tool finds and removes characters. It is not an AI detector, it cannot tell you who wrote a text, and it cannot promise that any detector will change its verdict.
Use it to clean text you own and to inspect text you were given. Do not use it to conceal authorship where disclosure is required.
Provenance marking is arriving
The EU AI Act's transparency rules became applicable on August 2, 2026, and the accompanying code covers machine-readable marking for generated text, images, audio, and video where it is technically feasible. Providers are responding in different ways.
Google publishes SynthID Text, a statistical watermark applied while the model chooses its next token. Anthropic is rolling out text marking for supported Claude models and Content Credentials on some file exports. OpenAI's public provenance work currently shows up as labels and file-level credentials rather than a published text-sampling spec.
None of that is a hidden character you can delete. That is the distinction this tool is built around: character-level marks are countable and removable, model-level marks are not.
Characters are not provenance
This scanner reports characters. It cannot verify authorship, read file metadata, or remove a model-level watermark, and it does not claim to.
Read the European Commission guidanceThe original and the cleaned version stay side by side, so you decide what leaves.
Nothing is uploaded. The scan runs in this tab, on your machine.
Every hidden character is named, counted, and shown in position. If it was carrying a message, you see the message.
Turn categories on or off, compare the two versions, then copy the one you want.
Answers about what the scan detects, what it cannot touch, and how your text is handled.
LLM Pulse is the all-in-one AI search and GEO platform. Use these free tools to audit how AI engines see your brand, then track it all in one place.
See how often AI engines mention your brand across ChatGPT, Perplexity, and Gemini.
Check whether AI answers mention your brand for the prompts that matter.
Measure how discoverable your brand is across AI engines.
Score your domain on the files, formats and interfaces AI agents look for.
Test whether AI engines can crawl and render your pages.
Check whether AI crawlers are allowed to read your site.
Generate an llms.txt file so AI crawlers understand your site.
Check whether your site appears in Common Crawl, the open corpus that trains most AI models.
Check whether your site structure helps AI engines understand you.
Audit your structured data so AI and search engines understand your pages.
Grade how well a page is optimized for AI answer engines.
See if your content is ready to be cited in AI answers.
Validate your product feed for ChatGPT shopping results.
Remove hidden characters, odd spacing, and leftover formatting from pasted text.
Clean the characters here, then use LLM Pulse to track how ChatGPT, Perplexity, and Gemini actually describe your brand.