Athens

How to Get ChatGPT Output into Substack Without Breaking Formatting

- Moritz Wallawitsch

You wrote something in ChatGPT. It looks great. Bold headings, numbered lists, clean paragraphs. You copy it. You paste it into Substack. And now your post looks like a broken config file.

The bold text shows up as **asterisks**. The headers are just lines starting with #. The bullet points are literal dashes. None of the formatting came through. Substack treated everything as plain text.

This is not a bug. It is a fundamental incompatibility. ChatGPT outputs markdown. Substack does not understand markdown. Every time you paste from one into the other, you get a mess.

Why This Happens

ChatGPT thinks in markdown. When it writes **bold**, it means bold. When it writes ## Heading, it means a second-level heading. Markdown is how every major language model formats its output. It is lightweight, readable, and universally supported by developer tools.

Substack is not a developer tool. Substack is a rich-text editor built for newsletters. It expects formatted text, not raw markup. When you paste markdown into Substack, it sees literal characters. The asterisks are asterisks. The hash symbols are hash symbols. There is no conversion layer.

Google Docs has a similar problem, but at least it preserves some formatting when you paste from ChatGPT's web interface. Substack strips almost everything. You end up with a wall of plain text peppered with markdown syntax that your readers will see verbatim.

The result: you spend twenty minutes reformatting a post that ChatGPT generated in thirty seconds. You manually bold every heading. You re-create every list. You fix every link. The "time saved" by using AI evaporates in the cleanup.

The Workarounds People Actually Use

Search for "ChatGPT to Substack formatting" and you will find a graveyard of workarounds. Every one of them works. None of them are good.

Ask ChatGPT for HTML instead

The most common suggestion: tell ChatGPT "give me this in HTML." ChatGPT will output your post wrapped in <h2>, <p>, and <strong> tags. Then you open Substack, switch to the HTML view (if you can find it), and paste the raw HTML.

This works. But now you are reading HTML in ChatGPT instead of formatted text. You cannot easily tell if the content is good because it is buried in tags. And if you want to make a quick edit, you are editing HTML by hand. One missing closing tag and Substack renders garbage.

Use pandoc on the command line

If you are technical, you might reach for pandoc. Save the ChatGPT output as a .md file, run pandoc input.md -o output.html, open the HTML file, copy the rendered content, paste into Substack. Some people pipe it straight to the clipboard: pandoc input.md | pbcopy on Mac or pandoc input.md | xclip -selection clipboard on Linux.

This produces clean output. But you are now writing blog posts through a terminal. You are saving files, running commands, and managing a pipeline just to get formatted text into a newsletter editor. For a writer who just wants to publish, this is absurd.

Paste through Google Docs as an intermediary

Another popular approach: paste the ChatGPT output into Google Docs first. Google Docs sometimes renders the formatting correctly (especially if you copy from ChatGPT's web interface rather than the API). Then you copy from Google Docs and paste into Substack. The rich-text clipboard data survives the second paste better than raw markdown.

This is a three-step copy-paste chain. ChatGPT to Google Docs to Substack. You are using a word processor as a format converter. The fact that this is a recommended workflow tells you everything about the state of the tooling.

Use a markdown-to-HTML web converter

Sites like Dillinger, StackEdit, or Markdown to HTML converters let you paste markdown in one pane and see rendered HTML in the other. You copy the rendered output and paste it into Substack. The formatting usually survives.

Like the Google Docs approach, this adds an extra tool to the chain. You are now managing three browser tabs: ChatGPT, the converter, and Substack. Every edit requires cycling through all three.

Use a browser extension

Extensions like "Markdown Here" let you type or paste markdown into any rich-text field and convert it in place with a keyboard shortcut. You paste the markdown into Substack, hit the shortcut, and the asterisks turn into bold text.

This is the least painful option. But it still assumes the workflow is: generate text in ChatGPT, copy it, paste it somewhere else, then fix it. The extension makes the "fix it" step faster. It does not eliminate the fundamental problem.

The Fundamental Problem

Every workaround above accepts the same premise: you write (or generate) in one tool and publish in another. The formatting breaks because these tools were not built to work together. Markdown is a developer format. Substack is a consumer product. They do not share a common language.

But the formatting is not even the biggest issue. The bigger problem is the copy-paste workflow itself. When you use ChatGPT to write a Substack post, you are outsourcing the writing. You get back a finished block of text. You cannot see what the AI chose to emphasize or de-emphasize. You cannot compare its version to yours. You accept the whole thing or start over.

This is not editing. It is delegating. And the readers can tell. AI-generated Substack posts have a particular flatness to them. The sentences are grammatically perfect but lack a point of view. The structure is predictable. The voice is generic. Readers subscribe to Substack writers for their voice, not for competent prose that any chatbot can produce.

The copy-paste era of AI writing trained writers to treat AI as a ghostwriter. But the best use of AI is not generation. It is editing. You write the draft. The AI helps you sharpen it. Your voice stays intact because the words started as yours.

The Real Fix: Write in Markdown, Edit with AI, Export Clean HTML

The solution is not a better paste tool. It is not a fancier conversion pipeline. The solution is to write in a tool that speaks both languages natively: markdown for the AI and clean HTML for Substack.

Here is what the workflow should look like:

  1. You write your draft in a markdown WYSIWYG editor. You see formatted text (bold headings, bullet lists, links), but the underlying format is markdown.
  2. You ask the AI to edit specific sections. Not "rewrite my whole post." Targeted edits: "tighten this paragraph," "make this opening more direct," "cut the filler from this section."
  3. The AI shows you inline diffs. Strikethrough for deletions, highlights for additions. You accept or reject each change individually.
  4. When the post is done, you export it as clean HTML. No markdown syntax in the output. Just proper <h2>, <p>, <strong>, and <ul> tags.
  5. You paste the HTML into Substack. Everything renders correctly. No reformatting. No cleanup. No extensions.

This is not hypothetical. This is how Athens works. You write in a markdown editor that renders formatting in real time. The AI edits your document directly, showing you exactly what it changed. When you are done, you export clean HTML that Substack accepts without mangling.

Why Markdown WYSIWYG Is the Key

The term "markdown WYSIWYG" sounds like a contradiction. Markdown is a plain text format. WYSIWYG means "what you see is what you get" - formatted output. But modern editors bridge the gap. You type naturally and see formatted text. Under the hood, the document is stored as markdown.

This matters for two reasons.

First, AI models read and write markdown natively. When the editor stores your document as markdown, the AI can make targeted edits to specific lines without any format conversion. No translation between rich text and markdown. No clipboard weirdness. The AI reads your actual document and writes changes directly to it.

Second, markdown converts to HTML perfectly. The conversion is deterministic. A ## Heading always becomes an <h2>. A **bold** always becomes a <strong>. There is no ambiguity, no edge cases that break. When you export from a markdown editor to HTML, the output is predictable and clean.

Substack's rich-text editor happily accepts clean HTML. The formatting renders exactly as intended. No asterisks. No hash symbols. No broken lists.

What About Substack's Built-in AI?

Substack has been adding AI features. You can generate post ideas and get writing suggestions inside the editor. For basic use cases, this might be enough.

But Substack's AI is limited in important ways. It cannot show you diffs of what it changed. It cannot edit a specific sentence while preserving the rest of your paragraph. It does not understand your full document as context when making suggestions. It is a lightweight assistant bolted onto a publishing platform, not an editing tool built from scratch.

If you write short newsletters with minimal editing, Substack's built-in tools may be fine. If you write long-form essays, research posts, or anything that requires serious revision, you need a dedicated writing tool.

A Practical Workflow for Substack Writers

Here is the workflow that eliminates the formatting problem entirely:

  1. Draft in Athens. Write your post in the markdown WYSIWYG editor. Use headings, lists, bold, links - everything renders in real time.
  2. Edit with AI. Select sections and ask the AI to improve them. Review the inline diffs. Accept the changes that make your writing better. Reject the ones that do not. Your voice stays because you wrote the draft.
  3. Export as HTML. When the post is ready, export it. The output is clean, semantic HTML.
  4. Paste into Substack. Switch to Substack's HTML view and paste. Everything renders correctly on the first try.

Total formatting cleanup time: zero. No pandoc. No Google Docs intermediary. No browser extensions. No manually re-bolding every heading.

Stop Converting. Start Writing.

The ChatGPT-to-Substack formatting problem is a symptom of a broken workflow. You are using a chatbot as a word processor and a newsletter platform as an HTML renderer. Neither tool was designed for the job you are giving it.

The workarounds (pandoc, Google Docs, browser extensions, HTML mode) all treat the symptom. They make the paste less painful. They do not fix the underlying problem: you are writing in the wrong tool.

The best AI writing tools for bloggers do not generate posts for you. They help you write better posts yourself. They store your document in a format that AI can edit natively. They show you what changed. And they export clean output that publishing platforms accept without breaking.

If you are a Substack writer spending more time on formatting than writing, the fix is not a better clipboard. It is a better editor. Try Athens and skip the conversion circus entirely.