Skip to content
TabShedKitsMy desk

Converting a PDF to Markdown for Notes, Wikis and Chatbots

Updated 5 min readBy

Turn PDF text into Markdown with headings, lists and tables. Use the "For LLMs" option to drop running headers, and know when columns or scans will fail.

PDF to MarkdownConvert a PDF to Markdown for notes and LLMs
Open PDF to Markdown
On this page
  1. How the conversion decides what is a heading
  2. Step by step
  3. The two options, and when each helps
  4. Preparing a PDF for a chatbot
  5. Notes and wikis
  6. Where it goes wrong
  7. A worked example: a 40-page policy document
  8. Batch conversion
  9. Going back
  10. Privacy
  11. FAQ

PDF is a good format for reading and a poor one for working with. You cannot paste a PDF into a note app and keep its headings, and if you feed raw PDF text to a chatbot or a search index it arrives full of page numbers, repeated headers and broken lines. Markdown is the opposite: plain text with structure. The PDF to Markdown tool converts one to the other in your browser. This guide explains what it can reconstruct, how to prepare text for a language model and what usually goes wrong.

How the conversion decides what is a heading

The tool reads the text of every page along with its position, size and font. From that it infers structure:

  • Larger text becomes headings.
  • Bold and italic runs keep their style.
  • Bullets and numbers at the start of a line become lists.
  • Rows of aligned columns become Markdown tables.
  • Links are kept.

Because it is inference and not knowledge, it works best on documents that use size and weight consistently, such as reports, manuals and articles, and worst on designed pages where a heading is the same size as the body text but in a different colour.

Step by step

  1. Drop one PDF or many. Each is read on your device the moment you add it.
  2. Check the preview. Switch between the formatted preview and the raw Markdown.
  3. Choose options: "For LLMs" and "Page separators".
  4. Copy the Markdown, or download a .md file. Several PDFs download together as a ZIP.

The two options, and when each helps

OptionWhat it doesUse it when
For LLMsRemoves lines that repeat near the top or bottom of pages: running headers, footers and page numbersPasting into a chatbot, building a search index or a knowledge base, where repeated lines waste space and confuse retrieval
Page separatorsAdds a line between pagesYou need to cite page numbers or find where a page ended

They pull in opposite directions: For LLMs strips page furniture, and separators mark the page boundaries. If you want to quote pages later, keep separators on; if you only need the content, leave them off.

Preparing a PDF for a chatbot

  1. Cut the PDF to the part you need with Split PDF; a smaller input is easier for any model, and you do not pay with attention on boilerplate.
  2. Convert with For LLMs on.
  3. Read the raw Markdown once. Look for sentences cut in half by a column break, tables that turned into runs of text and strange characters from ligatures.
  4. Fix those by hand, then paste.

The effort in step 3 is worth it: a language model treats whatever you give it as the truth, so a table that lost its columns will produce answers about the wrong numbers.

Notes and wikis

For note apps and wikis, the main value is headings and lists. After conversion, check the heading levels: a document with a very large title and slightly smaller chapter headings may end up with level 1 and level 2 correctly, while a PDF with uniform headings may flatten them all to the same level. Adjust the pound signs and your outline works in the app's sidebar.

Where it goes wrong

  • Scanned PDFs. A scan is a picture with no text layer, and this tool does not do OCR. You will see a notice when no text is found. Run OCR PDF first.
  • Multi-column pages. Pages with two or more text columns may come out with columns mixed together, reading across instead of down. Check these pages and reorder paragraphs by hand.
  • Complex tables. Tables are guessed from alignment, so merged cells or tables without spacing may need a quick fix. For a table you want in a spreadsheet, PDF to Excel gives you a column strip to correct.
  • Images and diagrams. Only text, links and structure are converted; pictures are not kept. To save pictures or pages as images, use PDF to Image.
  • Hyphenated line ends. Words that the PDF split with a hyphen at the end of a line may still look split in the text. A find and replace on the Markdown cleans it up.

A worked example: a 40-page policy document

Suppose you want a chatbot to answer questions from a company policy PDF. Drop the file and open the raw Markdown tab. Scan the first three pages: do the section titles start with pound signs, are bullet lists real lists, and has the table of contents been turned into a long list of titles with dots? The contents list is useless for a model, since it repeats the headings; delete it. With For LLMs on, the footer that said "Confidential - Page 12" on every page should be gone; if it is still there, its position varies between pages, and a find and replace removes it. Then look at the pages that have tables, such as a leave entitlement table. If the table became a Markdown table with the right number of columns, keep it. If it became a run of text, rebuild it by hand in a few minutes, because an answer about leave days is exactly the type of question that depends on that table.

Finally, keep the original PDF alongside the Markdown. When the chatbot cites a rule, you can verify it against the page. Page separators help here: they let you tell which PDF page a passage came from.

Batch conversion

Drop several PDFs to get a ZIP with one .md file for each. This suits a folder of manuals you want to index. Convert one first and inspect it, because the same quirks repeat across every file in the same template.

Going back

If you later want a PDF from the edited Markdown, Markdown to PDF does that, and Markdown to Word gives an editable file. Neither will reproduce the design of the original PDF, because the conversion keeps structure, not appearance.

Privacy

Everything runs in your browser. Your PDFs never leave your device and the tool works offline once the page is loaded. That is valuable when the document is internal and you are about to give it to an AI system you control, since you decide what leaves your device afterwards.

FAQ

Does it keep images?

No, only text, links and structure.

What does For LLMs do?

It removes repeated headers, footers and page numbers.

Why is a table wrong?

Tables are inferred from alignment; fix them by hand if they have merged cells.

Can I convert a scanned PDF?

No, use OCR PDF first.