Feed your AI clean text: turn any document into Markdown first

Ask an AI assistant to "read this PDF" and the answers often come back vague. The model receives a soup of columns, headers and page furniture. Models work best on clean, structured plain text — and the format every one of them natively reads is Markdown. MarkItDown, a free open-source tool from Microsoft (MIT licence), converts almost anything into it: PDF, Word, Excel, PowerPoint, HTML, CSV, JSON, XML, EPub, ZIP archives, images (with OCR) and audio (with transcription).

1. List the documents you actually answer questions from. Price list, service catalogue, standard contract, FAQ, last year's reports. Usually it's under twenty files.
2. Convert them in one pass. Install once with pip install 'markitdown[all]', then markitdown pricelist.pdf -o pricelist.md. Anyone technical — or your coding assistant — does the lot in an afternoon.
3. Spot-check the result. Tables and scanned pages are where every converter slips. Fix it once in the .md file and you reuse that fix forever.
4. Keep an "AI-ready" folder. Those .md files are what you upload, paste, or point an agent at — not the original PDFs.
5. Re-export when the source changes. A stale price list is worse than no price list, because the agent will quote it with confidence.

There's even an MCP server version, so an assistant can do the converting itself. Clean input is the cheapest quality upgrade available — it costs nothing and every model benefits.