PDF to Markdown: Turn a PDF into Text for AI
Pull the text out of a PDF as clean Markdown, ready to paste into any AI chat or agent. The PDF is read on your device and never uploaded.
or drop it here
Accepts: PDF
Converted on this device. Your files are never uploaded.
Why Markdown can cost fewer tokens than a PDF
Many AI services read a PDF twice: they extract its text and also look at an image of each page, so they can follow charts and layout. OpenAI’s and Anthropic’s documentation both describe sending text and page images together, and Anthropic puts the text alone at about 1,500 to 3,000 tokens per page before the image is added.
A Markdown file is only text, so the page-image part goes away. How much that saves depends on the service and the document, and it isn’t always a saving: Google’s Gemini API counts a PDF page as a fixed number of tokens (258, or about 560 by default on Gemini 3) and, on Gemini 3, doesn’t charge for the PDF’s own text, so pasting the text there can cost more.
What the converter keeps
Headings are found from font size, lines are joined back into paragraphs (also across page breaks), and bulleted and numbered lists become Markdown lists. Links keep their address. Tables become Markdown tables: ones drawn with borders are read cell by cell from the lines, and ones without borders from how the columns line up.
Two-column pages, such as research papers, are read one column at a time. Headers, footers and page numbers that repeat on every page are removed, and you can mark where each page starts so you can still ask about page 7.
What it can’t do
Charts, photos and diagrams are left out, and a scanned page with no text layer gives no text at all; the result tells you which pages those were. There is no OCR here, so for scans upload the PDF itself, or run OCR first.
Tables with merged cells or several header rows can come out shifted, and formulas become plain text such as 10^20. Read the preview before you rely on a table.
Advertisement
Questions
Does this work with ChatGPT, Gemini and other AI tools?
Yes. Markdown is plain text, so any AI chat, agent or LLM can read it: paste it in or attach the .md file. What differs between services is how they charge for a PDF, so the token saving differs too.
How many tokens will I save?
There is no single number. Where a service reads both the text and an image of each page (OpenAI and Anthropic describe this), converting removes the image part; by Anthropic’s published formula, one A4 page image is up to about 1,550 tokens on most Claude models. Where a page is billed at a flat rate and the text is free, as on the Gemini 3 API, Markdown may not save anything.
When should I keep the PDF instead?
When the meaning is in the picture: charts, graphs, drawings, slides, handwriting, scans, or forms where the layout matters. The AI can only see those if it gets the PDF or an image of the page.
Is my PDF uploaded?
No. PDF.js, the PDF reader built into Firefox, runs inside this page and reads the file on your device. The reader is downloaded from this site the first time you convert, and the file itself is never sent anywhere. You can check this in your browser's network tab.
How is this different from Microsoft’s MarkItDown?
MarkItDown is a free open-source command-line tool that converts PDFs and also Word, PowerPoint and Excel files. It needs Python installed. This page does PDFs only, with nothing to install, and keeps headings, lists and tables from the page layout.
It says no text came out of some pages.
Those pages are pictures: a scan, a photo of a page, or a PDF made from images. There is no text to extract, so give those pages to the AI as a PDF or image, or run OCR on them first.
Advertisement