What formatting artifacts do AI models leave in pasted text?
Why copying from ChatGPT, Claude, or PDF documents breaks formatting in Word, Gmail, Notion, and CMS editors.
When you copy raw Markdown or rich text from an AI assistant, hidden formatting symbols and typographical quirks often come along for the ride. The reference table below shows the exact patterns our cleaner detects and strips locally in your browser:
| Artifact Type |
Raw Pasted Input |
Cleaned Output |
Why It Matters |
| Markdown Bold & Italic |
**Key takeaway:** *always* test |
Key takeaway: always test |
Prevents literal asterisks from appearing in plain-text emails, social posts, and docs |
| Heading Hashes |
### 1. Introduction |
1. Introduction |
Removes #, ##, and ### prefixes without deleting the heading text |
| AI Em-Dashes (—) |
Clear writing—especially online—wins |
Clear writing, especially online, wins |
LLMs overuse em-dashes; replacing them with commas or hyphens restores natural pacing |
| Zero-Width Unicode |
wordcount test (U+200B, U+00A0) |
wordcount test |
Prevents invisible characters from breaking code compilers, regex searches, or character limits |
| Broken PDF Line Wraps |
Hard line break at column\nedge inside a sentence. |
Hard line break at column edge inside a sentence. |
Merges broken single newlines from PDFs while keeping double newlines as paragraphs |
After cleaning your draft, check its Flesch readability score and keyword phrase frequency in the text analyzer, or estimate prompt size in the LLM token counter.