Garbled Text Generator: Make It or Fix It

Garbled text is text that shows the wrong characters. Most of the time it happens because text was saved in one character encoding and opened in another, so an apostrophe shows up as ’ and a name like Zoë or José comes out with extra symbols. The technical name is mojibake, from the Japanese for "character change".

This garbled text generator makes that kind of damage on purpose. It does not just sprinkle random symbols on your words. It runs your text through the same byte-level mix-up that breaks real websites, emails and spreadsheets, so the result looks like a genuine encoding failure. That makes it useful for props, puzzles and stories, and for testing whether your own software mangles text.

If you want the stacked-marks horror look instead, the Zalgo text generator and corrupted text generator are better fits. If you want to understand where all of these effects come from, read how glitch text works.

Every character is stored as one or more bytes. UTF-8, the encoding most of the web uses, stores plain English letters as a single byte and everything else, like accents, curly quotes and emoji, as two to four bytes. Older single-byte encodings such as Windows-1252 and Latin-1 show one character per byte. If software reads UTF-8 bytes as Windows-1252, each byte of a multi-byte character gets its own character, and that is the junk you see.

Take the curly apostrophe. In UTF-8 it is three bytes: E2, 80 and 99. Read as Windows-1252, those bytes are ’. The table below shows the sequences you are most likely to run into, computed with the same method the tool uses.

Latin-1 and Windows-1252 differ in one range. Windows-1252 puts printable characters like the euro sign and curly quotes at byte values 128 to 159, while Latin-1 leaves them as invisible control codes. Browsers treat the "Latin-1" label as Windows-1252 (the WHATWG Encoding Standard (opens in new tab) lists "iso-8859-1" and "latin1" as labels for windows-1252), so the first style is the one you will meet most often. The Latin-1 style is here for the case where those hidden characters matter, such as when you are debugging a database.

The Accent Share slider decides how many letters get an accent before the error hits. At 1, a few letters change. At 10, about half of the vowels, n and c in your text do. Curly quotes and dashes are always converted when the option is on, since they are the most common real trigger.

1. Pick a mode. Make Garbled Text creates mojibake. Fix Garbled Text repairs it.

2. Type or paste your text. To garble, use sentences with apostrophes, quotes or accents. To fix, paste the broken text exactly as you see it.

3. Choose an encoding error and a level. The output updates as you type. Press Reroll for a different pattern of accents.

4. Copy the result. Press Copy, then paste it where you need it. Use "Try fixing this output" to see the round trip.

When fixing, the tool tries each word as Windows-1252 and then Latin-1, decodes the bytes as UTF-8, and repeats up to four times for text that was garbled more than once. It tells you how many words it repaired and how many layers it found.

Almost every case comes down to two parts of a system disagreeing about the encoding. The usual suspects:

• A web page with no declared encoding. Put <meta charset="UTF-8"> in the first 1,024 bytes of the page and save the file as UTF-8.
• A database or connection set to the wrong encoding. In MySQL, the old utf8 setting only stores up to three bytes per character, so use utf8mb4.
• CSV files opened in a spreadsheet. Excel can guess the wrong encoding for UTF-8 CSV files. Importing the file and choosing UTF-8, or saving it with a byte order mark, usually helps.
• Copying through several tools. Each hop between an email client, a word processor and a web form is another chance to mix encodings.

Once text has been saved through the wrong encoding and had unknown bytes replaced, it cannot be recovered from the damaged copy. Keep the original whenever you can.

• Testing software. Check that your forms, imports and exports keep accents and curly quotes intact.
• Fiction and games. Props for alternate reality games, damaged files, hacked notes and found documents.
• Puzzles and teaching. Show students what an encoding error looks like, then have them fix it.
• Bios and display names. Paste short garbled text into a display name or bio. Usernames and handles usually reject it.

If you want stacked marks, scrambled letters or decorative Unicode instead of encoding errors, these pages cover them.