PDF to Markdown for notes and AI tools: what carries over and what flattens

Why Markdown

Markdown is plain text with a few marks for structure: # for a heading, - for a list item. That makes it the format notes apps such as Obsidian are built on, it stays readable in any text editor forever, and version control can show what changed line by line. For AI assistants it has a further use: the headings tell the model where a section starts and what it is about, which raw text loses.

What you give up is everything about how the PDF looked: fonts, positions, page layout, colours. Markdown keeps the words and their structure, nothing else.

What carries over, and what flattens

We built a 4-page test report with the usual parts (a title, numbered headings, a bulleted and a numbered list, a table, a footnote, a 2-column passage, a web address, an italic sentence, and a running header and page number on every page) and converted it with the default settings. The result line counted the headings, paragraphs and list items it found and said “Left out 8 running lines (headers, footers, page numbers)”. We checked the table and the clause numbers again on 11 October 2026, after the converter learned to write tables.

Here is the start of the file it wrote, unedited:

# Quarterly Report

## 1. Summary

Sales rose in every region. See https://example.com/q3 for the data. This sentence is in italics for emphasis.

Key points

- Revenue up 12 %
- Two new stores
- Costs flat
1. Hire in Graz
2. Open in Linz
3. Review prices

### 1.1 Results by region

| Region | Q2 | Q3 |
| --- | --- | --- |
| Vienna | 1,200 | 1,350 |
| Graz | 800 | 910 |
| Linz | 640 | 700 |

Headings, lists and tables are the part that works well, and they are the part that matters most for notes and for AI tools. Clause numbers come through as printed: a contract’s 4.2.1 stays 4.2.1 and (a) stays (a), each clause on its own line, instead of being renumbered as a Markdown list. The converter puts no blank line between them, so a Markdown viewer that follows the strict rules shows a run of clauses as 1 paragraph; add a blank line between clauses if yours does. The structure is worked out from how the text looks, so a large pull quote can turn into a heading; the tool says so in its result (“check the structure”).

The column result depends on the file. Our test file writes its 2 columns line by line, left, right, left, right, which is what the converter followed. Many PDFs write each column in one go and come out in the right order, but you only know by looking.

Steps: from PDF to notes

1. Drop the PDF on the converter above (several at once is fine; each gets its own .md).
2. Running lines: leave on “left out” to drop headers, footers and page numbers that repeat on most pages. Set “kept” if the header carries something you need.
3. Page markers: switch on if you want to find your way back to the PDF. Each page then starts with <!-- page 2 -->, a comment that Markdown viewers hide.
4. Press Convert to Markdown and download. report.pdf becomes report.md.
5. Read the result line and the warnings. It names pages with no text (“has no text on pages 3, 4: probably scanned”) and reminds you that headings, paragraphs and tables are worked out from font sizes, spacing and positions, so check the structure.

Then into the place it goes:

• Obsidian. A vault is a folder of Markdown files. Obsidian’s help says you can drag the file into the File explorer pane, or move it into the vault folder with Explorer or Finder.
• Notion. Notion’s help gives Settings, Import, Text & Markdown, then the .md file. It notes a 5 MB limit per file on the Free plan and that import works on desktop and web, not in the mobile app. Its help lists headings and lists among what imports well.
• An AI assistant. Open the .md in a text editor, copy the sections you need, and paste them. The headings help the model keep sections apart.

The Obsidian and Notion steps follow their own help pages, checked on 11 October 2026.

For AI use: Markdown or plain text?

Pick Markdown when structure matters: long documents you will ask questions about, where “what does section 4 say” needs a section 4 to exist, or a set of documents you will split by heading. Pick PDF to text when the text is all you need, for example to count words, translate, or search, or when the PDF is a single block of prose with no headings to find.

Converting here also means you decide what leaves your device. The PDF is read in your browser; nothing goes anywhere until you paste text into an assistant or import a file into a cloud notes app, and then only the text you chose. A contract or a medical letter can be converted, cut down to the paragraphs you need, and only those shared.

The honest part: where it stops

A scanned PDF gives an empty file. Its pages are pictures, and there is no text to find: our scanned test page produced a Markdown file of 1 byte and the warning “has no text at all: probably scanned. Run OCR PDF first to make the text readable, then convert it.” Do exactly that: OCR PDF, then convert the result. The guide to making scans searchable covers the OCR step.

Tables are found from how the text lines up, so check each one. A table with cells in clear columns, or drawn as a ruled grid, comes out as a Markdown table; a frame drawn around a block of plain text does not count as one. For a table of numbers you will work with, PDF to Excel keeps amounts as numbers. PDF to Word keeps tables too, if Word is where the document is going.

Heavily designed pages (brochures, magazines, slide decks saved as PDF) put text in boxes all over the page, and the result reads as a heap of fragments. Bold and italic did not survive in our test, and links and pictures do not either. So convert 1 file, read it next to the PDF, and only then convert the other 49. For the wider choice between Word, Excel and text, see PDF to Word, Excel or text.