Free Tool + Guide October 2026

Scanned PDF to Markdown
AI OCR — Tables Included

Drop a scanned PDF and get clean Markdown. Pages without selectable text are read by an AI vision model, tables as Markdown tables. Free for the first 2 pages of each file — no account, no install.

🎉 Payment confirmed! Your conversions are unlocked — drop your files again to convert them.
Checkout cancelled — your free conversions are still available.

Drag & drop your file here

or click to browse  ·  multiple files supported

PDF · DOCX · PPTX · XLSX · TXT · HTML  ·  Max 20 MB each

Scanned PDFs are read by AI on every plan (free: the first 2 pages of each file).

Converting… 0%
 · 
Copied!

All files packaged into a ZIP archive.

5 free conversions left today

Why scanned PDFs come out empty

Most PDF-to-Markdown converters, including Microsoft's MarkItDown, read the PDF's text layer: the characters stored in the file. A scanned PDF has none. Each page is a picture of text, so the converter finds nothing and returns an empty file, or one that holds only page numbers. MarkItDown users report it on GitHub: "Markitdown produces nothing?" and OCR on non-text-layer PDFs generating only page numbers.

How RawMark reads a scanned page

1

Find the scanned pages

Every page is checked for selectable text. Pages that have it go through MarkItDown, as before.

2

Render at full resolution

Each scanned page becomes an image 1,568 px on its long side, whatever page size the PDF declares, so small print stays legible.

3

Read with AI

A vision model (Anthropic Claude) transcribes the page into Markdown: headings, lists, paragraphs, and tables row by row.

Partly scanned PDF? Each page is handled on its own and the result keeps the original page order.

What we tested

On 12 October 2026 we built six documents with known content, printed them to PDF, then turned each into a scan: rasterized at 150 dpi, slightly rotated, with grain added. Then we compared RawMark's output with the original text, word by word and row by row.

📊

Tables

An income statement, an invoice, a 7-column timetable and a results table: every row of every table came back exactly as printed.

📄

Dense text

A two-column page in 9 pt type, 571 words: all 571 came back, in the right order.

Limits we state plainly: these were printed English documents. We have not measured handwriting, faded photocopies, photos taken at an angle, or other languages. AI transcription can misread, so check figures that matter.

Plans and limits

  • Free: the first 2 scanned pages of each file, within 5 free conversions a day.
  • Paid plans: up to 50 AI-read pages per file. The Unlimited plan includes 2,000 AI-read pages a month.
  • Files: PDF up to 20 MB. Pages beyond the limit are marked in the result, never skipped silently.

Paid plans can also turn on AI accuracy, which reads every page of a digital PDF with AI: useful when MarkItDown merges the columns of a table. See pricing.

Frequently asked questions

Why does MarkItDown return an empty file for my scanned PDF?
MarkItDown reads the text layer of a PDF. A scanned PDF is a set of page images with no text layer, so there is nothing to read and the output is empty or holds only page numbers. RawMark detects those pages and reads them with an AI vision model instead.
How do I know if my PDF is scanned?
Open it and try to select a sentence. If you can highlight words, the page has a text layer. If you can't select anything, the page is an image (a scan or a photo).
Is it free?
Yes, for the first 2 scanned pages of each file, within the 5 free conversions a day. Paid plans read up to 50 pages per file; the Unlimited plan includes 2,000 AI-read pages a month.
Are tables kept as tables?
Yes. The model is asked to write every printed table as a Markdown table, one row per printed row. In our tests every row of every table came back exactly as printed. Check figures that matter: AI transcription can misread, especially on faint or skewed copies.
Does it read handwriting or languages other than English?
We have tested printed English documents only. The model can read other languages and some handwriting, but we have not measured how well, so check the result before relying on it.
What happens with a PDF that is partly scanned?
Each page is handled on its own: pages with selectable text go through MarkItDown, pages without it are read by AI, and the result keeps the original page order.
Are my files stored?
No. RawMark deletes the file as soon as the conversion ends. The images of the pages read by AI are sent to Anthropic's API for transcription; under Anthropic's Commercial Terms they are not used to train models. Details are in the privacy policy.