The AI-generated Markdown cleanup checklist (12 items)
Twelve specific things to clean up in AI-generated Markdown before you ship the document. What each one looks like, why it matters, and how to fix it.
1. Filler intros
The "Sure! Here's a draft of…" and "Of course — here's…" openers. Useful in a chat, dead weight in a document. Strip the first paragraph unless it contains substantive content.
2. Filler outros
"Let me know if you'd like me to…", "I hope this helps!", "Feel free to adjust." The last paragraph of an AI response is almost always conversational closing. Cut it.
3. Excessive horizontal rules
AI assistants insert--- separators liberally — between every section, between every example. In a document they read as visible junk. Keep them only where a strong section break genuinely helps; remove the rest.
4. Mixed bullet styles
A single list with*,- and• mixed across items. Normalize to one bullet character throughout the document. Picking- tends to play best with downstream parsers.
5. Smart quotes that shouldn't be smart
Curly quotes (“ ”) are fine in body prose; they're a disaster inside code samples, command-lines and JSON. Replace smart quotes inside code blocks with straight ASCII (" ") before you ship anything technical.
6. Em-dashes instead of hyphens
AI prose uses long em-dashes (—) as transition marks. That's fine. The mistake is when they appear in identifiers, slugs, file names or command-line examples, where they break tools. Audit anything that should be a hyphen.
7. Invisible Unicode characters
Zero-width spaces (U+200B), byte-order marks, soft hyphens. They don't render but they breakgrep, diffs, and copy-paste into terminals. Strip them.
8. Broken heading hierarchy
AI assistants start documents with## instead of#, or skip from# to###. Fix the hierarchy so each document has exactly one# H1 and no skipped levels. Outline-aware tools (Docs, Notion, Confluence) rely on this.
9. Unclosed code fences
A model that runs out of tokens mid-response leaves the closing``` off. Pasting the source into another document turns subsequent paragraphs into one giant code block. Always check that fence counts are even.
10. Tables with ragged columns
Body rows that have a different number of cells than the header. See thetable-repair article for the full treatment. The short version: count, pad or truncate, normalize the delimiter row.
11. Over-bolding
Some assistants bold every other phrase. In a document, that reads as emphasis on nothing in particular. Pick the two or three things that genuinely need to stand out and unbold the rest.
12. Trailing AI disclaimers
"As an AI…", "I don't have access to real-time data", "I cannot verify…". These belong in chats, not in your finished deliverable. Search and remove on every document.
Run the checklist automatically
Each item on this list is mechanical. None of them require judgement once you've decided that the cleanup should happen at all. That's exactly the kind of work that should be automated: paste your Markdown intoMarkdown Tidy, hitClean, and all twelve items are handled in one pass. You get the time back to do the work that actually requires you.
Related reading:8 common ChatGPT formatting artifacts ·Why ChatGPT Markdown breaks in Google Docs
Related articles
Markdown to Word (.docx): the right way to convert
Three working methods to convert Markdown to a Microsoft Word .docx — Pandoc, a web converter, and an editor export. Plus when DOCX beats PDF.
Markdown code blocks: fenced, indented, language hints, escaping
Everything about Markdown code blocks — fenced vs indented, syntax highlighting, escaping backticks, nesting, and the rules that differ between renderers.
MD file: what it is, how to open it, how to convert it
An MD file is a plain-text Markdown document. This guide covers how to open .md files, view them, and convert them to PDF, Word, or HTML.
Markdown lists: ordered, unordered, nested, and task lists
Every way to write a list in Markdown — unordered bullets, ordered numbers, nested lists, task lists, and the rules that decide what renders.