Skip to content

PDF to Text

Pull the text out of a PDF without the columns coming back interleaved, and be told plainly when a page is a scan with no text to extract.

Private · on device. Runs in your browser: your file is not uploaded. No sign-up.

Nothing is uploaded

A PDF does not contain paragraphs

This is the fact that explains every frustration people have with copying out of PDFs, and it is worth stating plainly: a PDF file contains no text structure at all. No paragraphs, no columns, no headings, no reading order. What it contains is a list of instructions along the lines of "draw the characters quick at coordinates 214, 588 in 11pt Times".

The order of those instructions is whatever the producing program found convenient. Often it matches reading order. On a two-column layout it frequently does not, the file may draw all of the left column, then all of the right, or alternate between them line by line, or emit them in an order that reflects the internal structure of the word processor and nothing else.

So when a converter gives you scrambled text, it has not failed to read the file. It has read the file perfectly, and the file did not contain what you assumed it did.

Rebuilding the structure from the geometry

The only reliable information about layout is where things are on the page. This tool uses that, in four steps.

Group text into lines. Pieces sharing a baseline belong on one line, within a tolerance, because subscripts, font changes and rounding all shift the baseline slightly.

Split lines at gutters. This is the step that matters. On a two-column page the left and right columns share their baselines, so grouping by baseline alone welds "left line 3" onto "right line 3". Splitting at any gap wide enough to be a column gutter keeps each piece inside one column.

Find the columns. A gutter is a vertical band that no line crosses, clear of the margins and wide enough to be deliberate. Full-width lines are set aside first, since a single heading crosses every gutter and would otherwise hide one that genuinely exists.

Read in order. Each column top to bottom, in turn, with full-width lines kept where they sit, so a mid-page heading separates the columns above it from the columns below rather than scrambling both.

Columns or a table? The one genuinely ambiguous case

A three-column newsletter and a three-column table are, geometrically, the same thing: vertical bands of text separated by gaps. There is no marker in the file distinguishing them, and the right reading order is opposite in each case, down the columns for the newsletter, across the rows for the table.

The tool guesses using the one difference that holds up: flowing text fills its column, while table cells are short relative to the space they sit in. That is a judgement, not a certainty, which is why the layout control above lets you override it. If an invoice comes out reading down the amount column, switch it to Rows. If a newsletter comes out interleaved, switch it to Columns.

Naming the ambiguity is more useful than pretending it does not exist. Every extractor faces it; most just pick one and give you no way to change the answer.

Scanned PDFs: the honest answer

A large share of "this converter is broken" complaints are about scanned documents, and the converters are not broken. A scan is a photograph. The words in it are arrangements of dark pixels, exactly as the words in a holiday snap of a road sign are. There is no text in the file to extract, so extraction correctly produces nothing.

What is unhelpful is returning an empty file without explanation. This tool checks each page and tells you which ones carry no text at all, so you know immediately whether you have an extraction problem or an OCR problem. They need entirely different tools, and knowing which you have saves trying five converters that were all going to fail.

The quick test you can run yourself: open the PDF and try to select a sentence. If words highlight, there is text. If a rectangle covers the whole area, it is an image.

What plain text cannot carry

Extraction gives you the words. It cannot give you bold, italics, font sizes, colours, tables as tables, images, or the position of anything on the page, none of that survives a conversion to plain text, by definition.

That is usually fine, because the reason people extract text is to search it, quote it, feed it to something else, or paste it somewhere they will restyle it anyway. If you need the formatting, you do not want text, you want the PDF, and tools like split or compress keep it intact. Going the other way, the HTML to PDF converter builds a PDF whose text stays selectable, so it never becomes this problem for the next person.

Frequently asked questions

Why does copying from a PDF mix up my columns?

Because a PDF has no columns. It stores individual pieces of text at x and y coordinates, in whatever order the program that made it happened to write them. There is no paragraph, no column and no reading order anywhere in the file. When a tool concatenates those pieces on a two-column page, you get line one of the left column, then line one of the right, then line two of the left, which is why the result is unreadable.

How does this one get it right?

By working out the layout from the geometry before reading anything. It groups text into lines, splits any line at a gap wide enough to be a column gutter, then looks for vertical bands that no line crosses. If it finds them, it reads each band top to bottom in turn. A heading spanning the full width is recognised as spanning and kept in place, so it does not get filed into one column or the other.

My PDF produced nothing at all. Is it broken?

Almost certainly not, it is a scan. A scanned document is a photograph of a page: the words are pixels, not text, and there is nothing to extract. This tool says so explicitly rather than handing you an empty file, which is what most converters do. Recovering the words needs OCR, which reads letters out of an image, and that is a genuinely different piece of software.

How can I tell whether my PDF is a scan?

Open it and try to select a sentence with your mouse. If you can highlight individual words, there is a text layer and this tool will extract it. If your selection draws a rectangle over the whole area instead, it is an image. A document can also be mixed, a digital contract with a scanned signature page, and this tool reports exactly which pages had nothing on them.

What does "join lines into paragraphs" do?

A PDF breaks lines wherever the column ran out, so raw extraction gives you a stack of short lines that reflow badly when pasted anywhere. This option joins them back into flowing sentences, keeping genuine paragraph breaks, keeping list items on their own lines, and repairing words that were hyphenated across a line break. Turn it off when the original line breaks matter, poetry, code, addresses.

Will tables survive?

The text will, in the right order, with cells separated by spacing. The table structure will not, because plain text has no tables. Set the layout to Rows for a table-heavy document: that keeps each row together across the page instead of reading column by column, which is what you want for an invoice or a statement and the opposite of what you want for a newspaper.

Is my document uploaded?

No. It is read in this browser tab and never sent anywhere. Worth noting for this tool in particular, since the documents people most want to extract text from, contracts, statements, reports, research under embargo, are the ones least suited to being uploaded to a free web service.

Can I get a Word document instead?

Not from here, deliberately. A .docx that is really just extracted text in a Word wrapper adds nothing you cannot get by pasting, and pretending the formatting survived would be misleading. Copy the text or download the .txt and paste it into Word, where you can apply your own styles.