Why "for LLMs" flips the ranking
Most converter comparisons rank tools on format breadth and how the output looks to a human. For AI work the ranking inverts, because three different things matter: token economics (Markdown encodes the same structure with far less overhead, so the same document costs fewer tokens), retrievable structure (Markdown headings become natural chunk boundaries and tables stay queryable in a RAG pipeline), and native format (ChatGPT and Claude emit Markdown themselves, so they read it back more reliably than any other input). A general-purpose converter optimizes for none of these — it optimizes for covering every possible conversion.