ChatGPT to PDF
ChatGPT wrote it. Now make it look published.
Paste the draft into Unplain and it comes back as a document: composed cover, contents page built from your headings, real pagination, your branding, and a closing call-to-action page. Roughly five minutes, no design tool, and the words stay exactly as you wrote them.
Two different things get called "ChatGPT to PDF"
Worth separating before you go further, because only one of them is what Unplain does.
Saving the conversation
You want the thread itself: your prompts, its replies, in order, as a file you can keep or share. That is a transcript export, and there are free browser extensions that do it well. Use one of those. Unplain will not help you, because it has no interest in the conversation, only in the content.
Publishing what it wrote
You asked for a guide, a report or a lead magnet, you got something good, and now it needs to look like a document rather than a chat reply. The prompts are irrelevant and the formatting is the whole problem. That is this page.
Canvas already exports a PDF
It does, and for a document you need once you should use it. Canvas will hand you a PDF, a Word file or Markdown in one click, and nothing here beats that for speed.
What it hands you is your text in a default layout. No cover. No contents page with page numbers that stay correct. Nothing carrying your colours or your logo, and nothing connecting it to the last document you made. For one memo that is fine. For the fourth report in a series, sitting next to the other three, it stops being fine.
The short version
Canvas is an export. Unplain is a publication. If the document is a one-off, export it and move on. If you will make another one like it next month, the difference is whether they match.
Getting a draft out of ChatGPT cleanly
One instruction saves the most trouble: ask for Markdown. Copy the response, paste it into Unplain, and the structure comes with it.
Headings become the contents page
Unplain builds the table of contents from your heading levels, so ask ChatGPT for a clear H2 per section rather than bolded pseudo-headings. A bold line is styling; a heading is structure, and only one of them can be listed with a page number.
Lists, tables, links and emphasis carry over
Markdown tables are laid out and split across pages at row boundaries. Links stay clickable in the exported PDF, which matters if the document ends in a call to action.
Code blocks and quotes survive
Both render as written. Nothing needs reformatting between the chat and the document.
Em dashes and the tells stay yours to fix
Unplain lays out what you give it. If a draft reads like it was generated, it will read that way in a nicely designed PDF too. Edit before you export, or afterwards in Unplain's editor.
If you are exporting from Canvas instead
Download it as Markdown rather than PDF and upload the .md file. Markdown keeps the structure Unplain reads; a PDF has already thrown it away and turned your headings into large text.
Doing it more than once
The first document is the easy one. The problem arrives at the fourth, when they need to look like a set and you can no longer remember which accent colour you used in January.
Unplain saves a Brand Theme: colours, fonts, logo. Every document after it inherits them without anyone restating anything, so the tenth matches the first because it is the same stored record rather than a fresh description. Free covers one theme; Pro allows up to 25, which is the case for agencies running several clients.
Revisions work the same way. The document stays a project in your account, so when one sentence is wrong you click it and type. You are not re-prompting a model and paying for a whole file to be built again, and nothing you were happy with quietly changes.
If you use Codex, skip the copying
Unplain runs an MCP server, and Codex is one of the two clients verified against it. Connect it once and you can ask Codex to write the report and create it in Unplain in a single instruction, with the PDF coming back without anything passing through your clipboard.
The consumer ChatGPT app is not an MCP client for Unplain, so from there it stays paste or upload. That is a two-second step, and it is the whole difference.
Frequently asked questions
How do I turn ChatGPT output into a PDF?
Ask for the piece in Markdown, copy it, and paste it into Unplain. It builds the cover, a contents page from your headings, pagination, branding and a closing call-to-action page, then exports a PDF. You can also upload a .md or .docx file if you exported one from Canvas.
Canvas can already export a PDF. Why use anything else?
For a one-off, do that instead. Canvas gives you your text in a default layout: no cover, no contents page with page numbers, no branding, and nothing tying it to the last document you made. Unplain adds those, and keeps them consistent across a series, which is the case Canvas is not built for.
I just want to save the conversation, not publish it.
Then Unplain is the wrong tool and a transcript exporter is the right one. Several free browser extensions save a whole ChatGPT thread, prompts included, as a PDF. Unplain takes finished content and designs it as a document; it has no interest in the conversation around it.
Does ChatGPT's formatting carry over?
Yes. Headings, bold, italic, lists, blockquotes, tables, code blocks and links all render into the designed PDF. Ask ChatGPT for Markdown and there is no reformatting step at all.
Can I connect Unplain to ChatGPT directly?
Not the consumer ChatGPT app. Unplain's MCP server is verified with Claude Code and Codex, so if you use Codex your assistant can create and export documents in your Unplain account without any copying. From ChatGPT itself, paste or upload.
Does Unplain write anything with AI?
No. ChatGPT produces the words and Unplain publishes them. The only AI in Unplain is optional cover artwork.
Is it free?
The free plan exports documents up to 5 content pages, where the cover, contents and closing pages do not count, with a "Made with Unplain" badge. Enough to run a real draft through and see the output before deciding anything.