URL: /rules/content/mojibake

---
title: "Encoding Corruption"
description: "Detects mojibake and replacement characters in visible text"
---

Detects mojibake and replacement characters in visible text

| | |
|---|---|
| **Rule ID** | `content/mojibake` |
| **Category** | [Content](/rules/content) |
| **Scope** | Per-page |
| **Severity** | warning |
| **Weight** | 5/10 |

## Solution

Mojibake means the bytes and the declared encoding disagree. Fix the declaration first: serve `<meta charset="utf-8">` as the first thing in the head, and send `Content-Type: text/html; charset=utf-8`. If the declaration is already utf-8, the source itself was double-encoded, usually by a CMS migration that read utf-8 bytes as latin1 and re-encoded them. Re-import the affected content from the original source rather than search-and-replacing the visible symptoms, because the same corruption is usually present in fields you cannot see, such as meta descriptions and alt text.

## What it checks

Three families of corruption, in visible text only:

| Kind | Example | Meaning |
|---|---|---|
| `utf8-as-latin1` | `It’s`, `café` | utf-8 bytes decoded as latin1 or cp1252 |
| `replacement-char` | `caf�` | the decoder gave up on those bytes |
| `double-encoded-entity` | visible `&amp;nbsp;` | the source said `&amp;amp;nbsp;` |

The check reads text with `<script>`, `<style>` and `<noscript>` excluded, never raw HTML, so an encoded sequence inside a script body is not reported. A page with no body is skipped rather than passed.

**Severity depends on certainty.** The replacement character is unambiguous corruption and fails. The other families warn.

## Cause attribution

The finding cross-checks `<meta charset>` so the message names the actual fix rather than just reporting odd characters:

- **No charset declared**: the browser guessed. Add `<meta charset="utf-8">`.
- **Charset is utf-8**: the declaration is right, so the source is double-encoded. Re-import the content.
- **Charset is something else**: correct the declaration, then re-check whether the source is also double-encoded.

## False positives

Correctly-decoded accented text is not corruption. Copy such as `Un café à Paris`, `Grüße aus Köln`, `jalapeño`, `25°` or curly quotes stays clean, because the rule matches multi-character *sequences* that only arise from a bad decode, not individual accented letters.

## Enable / Disable

### Disable this rule

```toml squirrel.toml
[rules]
disable = ["content/mojibake"]
```

### Disable all Content rules

```toml squirrel.toml
[rules]
disable = ["content/*"]
```

### Enable only this rule

```toml squirrel.toml
[rules]
enable = ["content/mojibake"]
disable = ["*"]
```