# What's the best way to give a PDF to ChatGPT or Claude?

Source: https://convcc.com/answers/pdf-to-markdown-for-chatgpt/ (updated 2026-11-17)

Convert it to clean Markdown first: the model then gets the text in reading order with headings, lists and tables, without page headers and page numbers repeated on every page. That is fewer tokens and no sentences mixed from two columns.

## What goes wrong with the raw PDF

- Two-column pages read across, so sentences from both columns get interleaved.
- The same header and footer repeated on every page, which the model reads as content.
- Tables flattened into runs of numbers with no idea which column each belongs to.

## For a retrieval (RAG) pipeline

- Split by headings, so each chunk is a section that makes sense alone.
- Keep each table whole in one chunk.
- Use JSON if you need the page number of each block to cite it.
