PDF to Markdown for AI: why plain text beats uploading the PDF
PDF is a format designed for printing. It records where each letter sits on a page, not how the text is structured. That's perfect for a printer and awkward for an AI model, which needs the words, in order, with a sense of what's a heading and what's a list.
Markdown is the opposite: a plain-text format where structure is written with a few simple symbols.
# Supply Agreement
## Clause 1: Payment
Payment is due within **30 days** of invoice.
- Late payments carry 1.5% monthly interest
- Disputes must be raised in writingAI models are trained on enormous amounts of Markdown, from documentation and code repositories to chat transcripts. They read it fluently and often reply in it.
Benefits of converting first
- Smaller. A PDF holds fonts, images and layout data. Markdown holds only the text and its structure. The same contract can shrink from megabytes to kilobytes.
- Fewer tokens. No page images, no repeated headers and footers. See how to use fewer tokens with documents.
- More reliable. You can read the Markdown yourself before sending it, so you know exactly what the assistant sees. With a PDF, you're trusting the tool's own extraction, which can scramble columns or skip text.
- Editable. You can delete irrelevant pages, fix an OCR mistake, or add a note before you paste it.
- Works everywhere. Plain text can be pasted into any chat, any model, any app, and doesn't depend on whether a tool accepts attachments.
How to convert a PDF to Markdown
- Open the converter and drop your PDF onto it.
- Pages that contain real text are read directly. Pages that are only images (scans) are run through OCR in your browser. Pick the document's language in the OCR box first.
- Preview the result, then download the
.mdfile or copy it.
Each page is marked with a heading such as ## Page 3, so you and the AI can refer to page numbers. Pages converted with OCR are labelled (OCR) so you know which parts to double-check.
What converts well, and what doesn't
| PDF content | Result |
|---|---|
| Typed text, letters, contracts, reports | Exact text, instantly |
| Scanned pages and photocopies | OCR text: check names and numbers |
| Multi-column layouts | Usually correct reading order; check complex pages |
| Tables | Text in rows; the grid lines are not kept |
| Charts, photos, signatures | Not included (they are images) |
If a table matters, ask the assistant to "rebuild this as a Markdown table" after pasting. Models are good at this when the numbers are in the right order.
Scanned PDFs
A scanned PDF is really a set of photos. To get text out of it you need OCR (optical character recognition). See how to convert a scanned PDF to text for tips on getting accurate results.