See what your AI sees

Try a Unicode tag payload · zero-width characters · a bidi override · homoglyphs · clean text

What it looks for

• Unicode tag characters (U+E0000–U+E007F) — a full ASCII alphabet that renders as nothing. Subtract 0xE0000 from each and the hidden sentence falls out. This is the channel used by the signup form in the survey this tool came out of.
• Variation selectors (U+FE00–U+FE0F, U+E0100–U+E01EF) — the same trick with a different alphabet, usually hung off an emoji so there is something to attach to.
• Zero-width characters — ZWSP, ZWNJ, ZWJ, word joiner, BOM, soft hyphen. Long runs of them encode data one bit at a time.
• Bidirectional overrides (U+202A–U+202E, U+2066–U+2069) — "Trojan Source": the characters are all visible, but they are displayed in a different order than they are stored, so code can read one way and execute another.
• Private use area (U+E000–U+F8FF and the two supplementary planes) — renders as a box or as your font's guess, means whatever the sender decided it means.
• Homoglyphs — Cyrillic а, Greek ο, fullwidth Latin, sitting inside otherwise-ASCII words. Not invisible, just not the letter you think it is.

What it is not

Not a safety verdict. Plenty of text with none of these is hostile, and a few legitimate things use them — ZWJ builds emoji families, ZWNJ is required in Persian and Hindi, and variation selectors do real typographic work. It tells you what is there. Reading it is still your job.

Everything runs in this page. Nothing is sent anywhere, there is no request to my server after the page loads, and you can save the file and use it offline. Whatever you paste is presumably sensitive enough that you wanted to check it, so it should not have to travel.