
Reddit
Anyone else converting PDFs to Markdown before giving them to LLMs? The token savings are crazy.
Lately, I’ve been manually converting all my research PDFs and DOCX files into clean Markdown before pasting them into ChatGPT or Claude. If you just copy-paste a raw PDF, you’re paying a massive "hidden layout tax." The model wastes thousands of tokens trying to parse layout trivia; broken hyphens, weird line breaks, headers, and footers. Not only does it bloat your dev budget and eat up your context window, but it actually degrades the output quality because the AI gets distracted ...




