Why this exists
Text is not as simple as it looks. A paragraph that renders perfectly on screen can contain characters with no width, characters that reverse reading order, letters from one alphabet impersonating another, and, occasionally, an entire hidden message encoded in characters that display as nothing at all.
Most of the time this is harmless clutter picked up from a word processor or a PDF. Sometimes it breaks things in ways that are genuinely difficult to diagnose: a database lookup that fails on two strings that look identical, a CSV that will not parse, a search that returns nothing for a term plainly visible on the page. And occasionally it is deliberate.
Existing tools mostly answer a yes-or-no question. We wanted one that answers the useful question instead: what exactly is in this text, and does it matter?
How the detection works
There is no heuristic and no model involved. The engine walks the text one Unicode codepoint at a time and classifies each one against explicit character tables. That means every finding is exact and verifiable: you get the codepoint, the official Unicode character name, and the count. Nothing is inferred, so nothing is guessed wrong.
Characters are grouped into nine categories and weighted by how unusual they are in ordinary prose. A curly quotation mark is legitimate typography. A zero-width non-joiner in the middle of an English word is not. The score reflects that difference rather than treating every finding as equally alarming.
Two hidden-data encodings are decoded automatically. Unicode tag characters
(U+E0000 to U+E007F) map one-to-one onto ASCII, so any run of them
is converted straight back to readable text. Zero-width binary steganography is also decoded,
but only reported when the result is clean printable ASCII of a believable length. A false
claim that your document contains a secret message would be worse than saying nothing.
Your text stays with you
The engine is JavaScript that runs on the page you are already looking at. Paste text into either tool and it is processed in your browser's memory. There is no upload, no server-side copy, no database row, and no log line containing your content.
This is verifiable rather than a promise: open your browser's developer tools, switch to the Network tab, and use the tool. No request carries your text. The only optional request is an anonymous counter recording that a run happened and which categories were found.
A server-side copy of the same engine exists for the public API and for visitors with JavaScript disabled. Text sent that way is processed in memory and discarded; it is never written anywhere. Both copies are tested against the same suite to guarantee they return identical results.
What these tools will not do
We would rather be useful than oversell, so plainly:
- This does not defeat AI detectors. Detectors score statistical patterns in word choice and sentence structure. They do not look for special characters. Removing a zero-width space changes nothing about how a detector scores your writing.
- This does not prove text was AI-generated. Word, Google Docs, PDFs and ordinary web pages all introduce the same characters. Findings tell you what is in a document, not where it came from.
- This does not rewrite or paraphrase. The tools change characters, never words. Your sentences come out exactly as you wrote them.
Editorial standards
Everything published on this site is written by people who work with this material directly. Claims about Unicode behaviour are checked against the specification, and any codepoint we name is one the engine actually handles. Where something is uncertain or best-effort, the article says so rather than rounding up to a confident answer.
Articles are dated and updated when the underlying facts change. If you find something wrong, tell us and we will correct it.
Independence
This site is not affiliated with, endorsed by, or connected to Anthropic in any way. “Claude” is a trademark of Anthropic, PBC, used here only to describe what the tool is commonly searched for. The tools are general-purpose Unicode utilities and work on text from any source.
The site is free to use and carries no account system. Running costs are covered by advertising where it is enabled, which never affects what the tools report.
Get in touch
Bug reports, feature requests, corrections and questions are all welcome through the contact page. If you have found a character category the engine misses, that is the most useful message you can send us.