Which version of markdown will be the most compatible for LLMs such as ChatGPT, Gemini...etc

[mindxw]

New to Pandoc. I want to use it for two types of file conversations:

• PDF to markdown
• EPUB to markdown

The markdown output file will be uploaded to use when prompting ChatGPT, Gemini and Claude.

When using markdown files for LLMs which type of markdown from Pandoc should I use. I read this older thread and didn't know Pandoc had several types of markdown to choose from.

#10739

[rnwst]

I suspect these LLMs have ingested lots of training data from each of the markdown formats, so it probably doesn't really matter.

[mindxw]

Thanks for the reply! I am new to Pandoc and markdown so which of the formats in Pandoc would you choose to be the most universally accepted. Any help to suggest one to standardize on would be appreciated.

[rnwst]

Depends on your expected content. Pandoc's Markdown is by far the most capable of the available markdown formats, and therefore you can expect the conversion to work best with that format. However, it is also a much less popular Markdown format than some other options such as commonmark or gfm, so your LLMs will have had much more training on those formats. Just pick one and experiment.

[rnwst]

For reference, ChatGPT apparently uses gfm in its output.

[pdfkungfoo]

ChatGPT by default uses Github-flavored Markdown (GFM) in its output. ChatGPT very likely is able to understand any other flavor of Markdown too. I'd guess it is similar for all the other LLMs. Why? In their huge pre-training data corpus they must have digested (and learned to understand) all sorts of Markdown dialects... Am Fr., 4. Apr. 2025 um 19:44 Uhr schrieb mindxw ***@***.***>:

…

New to Pandoc. I want to use it for two types of file conversations: - PDF to markdown - EPUB to markdown The markdown output file will be uploaded to use when prompting ChatGPT, Gemini and Claude. When using markdown files for LLMs which type of markdown from Pandoc should I use. I read this older thread and didn't know Pandoc had several types of markdown to choose from. #10739 <#10739> — Reply to this email directly, view it on GitHub <#10756>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABEHFOD6L4H7TCPQV72KRJD2X3AHLAVCNFSM6AAAAAB2PI3MDSVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZYGE3DONJYHE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>