ChatGPT to Word

When you have a good conversation with ChatGPT, you want to keep it. Maybe you asked it to explain something and the explanation was actually clear. Maybe you worked through a problem together and want the result somewhere permanent. Maybe you want to send it to someone who doesn't use ChatGPT.

ChatGPT has a share button, which gives you a link. That's fine, but links are fragile. They live on someone else's server. You can't print them. You can't email them to your boss as an attachment. You can't highlight them or scribble in the margins.

So I made ChatGPT to Word. You paste a share link, pick which messages to include, and download a .docx or .pdf. That's it.

How to use it

First, share your conversation in ChatGPT. There's a share icon near the top of the chat. Click it, then copy the link. It'll look something like https://chatgpt.com/share/abc-123-def.

Go to chatgpt-to-docx.web.app and paste it in. The app fetches the conversation and shows you all the messages with checkboxes. Uncheck the ones you don't need, like the first few messages where you were still figuring out what to ask, then hit download.

The Word file preserves the formatting. Headings stay as headings. Code blocks stay as code blocks. Bold text stays bold. It's not a raw text dump.

Why

I use ChatGPT for research and I kept wanting to save the useful parts. Copy-pasting into a Google Doc loses all the formatting. Screenshots are awkward to share. A clean Word document is the boring, correct answer.

There's no sign-up. Nothing is stored. You paste a link and get a file.

The tricky part

ChatGPT's share pages don't have the conversation sitting in the HTML. They're single-page apps, so the data is embedded in a serialized format that only loads with JavaScript. The backend function fetches the raw page and decodes that format to pull out the actual messages.

This means if ChatGPT changes how their pages work, the tool breaks. I have a daily test that checks whether parsing still works and opens a GitHub issue if it doesn't. So far it's been reliable, but that's the nature of building on top of someone else's product.

Source on GitHub.