Extract Text from PDF

Three formats, and the two unusual ones are the point

Most extractors give you plain text and stop. Switch tabs here and the same extraction comes out three ways.

Plain Text is the straightforward one, with a --- Page N --- marker between pages so you can see where each one starts. Useful when you are pasting into an email, a document or a search box. If you do not want the markers, they are one find-and-replace away.

Markdown gives you the text with each page as a heading, ready to drop into a notes app, a wiki, a static site or a README. Obsidian, Notion, GitHub and most documentation systems take Markdown directly, which makes this the fastest route from a PDF into a knowledge base.

JSON is the one worth knowing about if you write any code at all. You get the filename, the page total, a timestamp, an array of pages each with its number, content, word count and character count, and a total word count for the document. That is a structure you can loop over immediately — no parsing, no guessing.

Worth noting why that matters: searching for a PDF-to-Markdown or PDF-to-JSON converter turns up page after page of paid developer APIs wanting a key and a subscription. For a handful of documents, that is a lot of friction for something a browser can do.

A scanned PDF will give you nothing, and it tells you so

Check this first, because it is the commonest reason the job fails.

There are two kinds of PDF. One holds real text, which you can select with your cursor. The other is a picture of text — a scan or a phone photo — where the page is one image and the file contains no text at all.

This reads the text layer, so a scan has nothing to give it. You will get a clear message saying no text was found rather than an empty file, which at least saves you wondering.

The test takes a second: open the PDF and drag your cursor across a sentence. Words highlighting individually means you are fine. A blue box over the whole page means it is a scan.

For a scan you need optical character recognition, which is a different operation — Image to Text (OCR) reads text out of an image and will tell you whether the text is recoverable.

Word order is right. Column order is not.

Being specific about this, because it is the difference between usable output and nonsense, and it comes down to one design decision.

A PDF does not store sentences. It stores fragments of text with coordinates, and often in an order that has nothing to do with reading order. A naive extractor joins them in file order and produces scrambled words.

This one groups the fragments by their vertical position into lines, and then sorts each line by horizontal position before joining it. So within any line, the words come out in the order you would read them. That is the single most important thing an extractor can get right, and plenty do not.

The same design has a consequence you need to know about. Because a line is "everything at this height on the page", a two-column layout merges both columns into one line — you get a line from the left column immediately followed by the line beside it from the right column, joined together.

So an academic paper, a newspaper page, a two-column report or a brochure comes out interleaved and unreadable, and no setting fixes it. If your PDF is in columns, extract it a column at a time by cropping first, or use a desktop tool that understands layout.

Tables have a related problem. A table row is a single line at one height, so the cells arrive joined by spaces with nothing marking the column boundaries. For a PDF that is mostly tables, PDF to TXT Converter will not do better — a spreadsheet converter is the right tool.

What the Markdown headings actually detect

Useful to know before you rely on the structure, because it is narrower than "heading detection" suggests.

A line is promoted to a heading when it is entirely in capitals and reasonably short. Font size is not looked at.

Which means two things. If your document's headings are set in capitals — common in reports, contracts, manuals and legal documents — they are found reliably. If they are in Title Case, like "Chapter One: Beginnings", none of them is detected and you get a flat document.

And it goes the other way: any short all-caps line becomes a heading whether or not it is one. A "NOTE:" callout, a caps running header, a line of acronyms, a Roman numeral — all get promoted. You will see a few headings that should not be there.

One more thing to fix if the Markdown is going anywhere structured: the page numbers outrank the real headings. Pages come out as second-level headings and detected headings as third-level, and there is no top-level heading at all. For a wiki or a static site you will want to promote the document title and demote the page markers — a minute's work in any editor, but do it before publishing or your table of contents will be upside down.

Also worth expecting: each line of the PDF becomes its own paragraph in the Markdown, because a PDF records printed lines rather than paragraph breaks. The text is all there and in order; it reads loose until you rejoin the paragraphs.

Practical notes

The whole document is extracted — the page count is shown when you upload so you know what you are getting.

The stats line gives you the total words, characters and pages, which is often all someone actually wanted from a PDF in the first place.

The download buttons are separate from the format tabs, so you can take more than one format from a single extraction without running it again.

A password-protected PDF will not open — remove the password first. A very long document will take a while and hold the text in memory as it works.

Nothing is uploaded. That matters for this tool in particular: the documents people extract text from are overwhelmingly contracts, invoices, reports, medical letters and unpublished papers, none of which belongs on a stranger's server for the sake of copying some words out.

Related tools

PDF to TXT Converter is the simpler sibling — plain text only, without the format tabs, which is what you want if you just need to read or paste the words. Markdown to HTML Converter turns the Markdown output into HTML if that is where it is heading. Image to Text (OCR) is the route for scanned documents.