OK, so we’ve all been there. The LLM‘s spit out some useful text seems like it’s formatted in markdown then we need to paste it somewhere and it just comes out unformatted. Is this a magic intermediate step or perhaps a prompt to get it to spit out something better? I know I’ve got it to create PDFs or as in a word format in the past any suggestions I just want to cut and paste I’ve tried stuff like Google Docs and Draft to no avail.
I’ve used various prompts to have the output reformatted. I even put it into my profile or pre-prompt instructions, depending on the tool.
I haven’t had much luck as the results are still some hybrid of formatting/non-formatted text that I have pretty much given up.
With DevonThink, where I am usually pasting the output, the result of direct copy/paste is “good enough” for me right now as i am saving mainly for an archive and use smaller copy/paste for useful snippets.
There is one prompt that has helped - I ask the LLM to not format any results in tables and to use bulleted lists instead, as much as possible.
That, at least, makes manually reformatting a little easier, when I do attempt to do so.
Ask the LLM to provide the answer as an RTF/plain text/markdown document that can be downloaded?
–
JJW
Here are a few approaches that work for me with ChatGPT, depending on the output:
- Ask for the reply a markdown code block
- Open the conversation in a web browser and highlight everything then copy with ⌘C or Edit>Copy instead of using the copy button. This works in some cases where the ChatGGPT app adds hard breaks within its response so you can’t highlight an entire response at once.
- I use Bear.app, which has excellent markdown support. In some cases, Edit>Paste From> html maintains formatting that is otherwise lost with regular paste.
One of these approaches works in every case, except tables. If the response includes tables and other text, the table will lose its formatting. In this case, I have to highlight as described, either in the app or web, then copy/paste each table. So I will copy and paste everything into Bear in one pass, then overwrite the tables, one at a time.
I often want my results as plain markdown text to paste into Obsidian. Even when I ask for exactly that, more often than not I get regular text rather than markdown. I have had better success recently with explicitly asking for ‘raw’ markdown text rather than ‘plain’.
If I have a markdown file I want to convert, I use MarkEdit. It gives you a terminal command for Pandoc that will turn the currently open md file into a word doc, complete with Heading, Body, etc all formatted. It even opens the terminal for you at the right location, so you just press Cmd+V and Enter. (You can of course just use Pandoc itself, but I can never remember the commands).
If I’ve copied markdown and want it in a different format, I use Text Workflow (Text Workflow — gTrigonakis) which takes input text and converts it to a different output via rules. So it can selectively strip markdown as needed.
I wouldn’t buy it just for this one task as its (I think) about £9, but its helpful for all sorts of little text formatting jobs.
Gemini gives you the option to export its response to a Google Doc, which you can then download in any number of formats.
Neither ChatGPT nor Claude can export responses directly to a Google Doc, but you can paste their responses into a MD file, open it Google Docs, and download it in the desired format.
I’m not a big Google Docs user, but I find this maneuver more straightforward than pandoc.
If you use the Chat assistant in DEVONthink 4 – either the popover or inspector – the natural output is Markdown, specifically MultiMarkdown. It also has the benefit of being able to create the document for you directly in the database.
However, in web interfaces or bespoke AI apps, you can clarify with a phrase like “Output in MultiMarkdown syntax” to use it in DEVONthink.
https://discourse.devontechnologies.com/t/importing-md-with-indented-task-lists/85419/2
2 Likes
When all else fails, “Paste and Match Style” (exact wording depends on app) works in most apps.
Katie
1 Like