Convert many PDFs to text at once, and catch the scans that come out empty

Convert a folder of PDFs to text, step by step

1. Drop the folder on the tool above, or press or a whole folder under the drop zone and pick it. Subfolders are included. Files that are not PDFs are left out, and the page says how many (“Left out 3 files this tool does not take”).
2. Check the list. Each file shows its page count and size; a file that needs a password to open shows “locked”.
3. Pick what goes between pages under Between pages: Empty line (the default), Page 2 (a line naming the page that follows, useful when you need to cite pages later), or Page break (a form feed character that printers and some text tools split on).
4. Press Get the text. The progress line reads like “Reading the text · file 12 of 200 · page 2 of 3”.
5. Read the result. It says “200 files are ready”, lists every .txt with its size, and names each file’s character count, for example “invoice.pdf: Took 172 characters of text from 3 pages, in the order a PDF reader finds it, as UTF-8”. Files that could not be done are listed under “1 file was not done; the others were:” with the reason.
6. Save. Download all (200) as zip gives you getpdf-files.zip. In Chrome and Edge, Save to a folder writes each .txt straight into a folder you pick, with no zip in between. Nothing already in the folder is replaced: if report.txt is there, the new one is saved as report-2.txt. An empty folder is still the tidiest choice.

The text is UTF-8, so accents, Polish, Czech, Greek and Cyrillic come through as they are: “Łódź” stays “Łódź”.

How many files can one batch take?

The batch runs each PDF as its own job, one after another, and only 1 file is open in the engine at a time. That is why the count barely matters. We measured the same code the page runs, on our test laptop:

• 200 invoices of 3 pages each (0.8 MB of PDFs) became 1.9 million characters of text in 6.7 seconds, about 33 ms a file.
• 1 file of 500 pages became 21,891 characters of text in 173 ms.

In the browser tab, expect a little more time per file than that. What does not scale is a single gigantic file. The engine has to hold the whole PDF in the browser’s memory while it reads it; if a file does not fit, that file is listed as “too big for this browser to open”, with what the device can hold, and the batch carries on with the next one. Text itself is small, so hundreds of results wait in the tab without trouble until you save them.

The silent failure: scans that come out empty

A scanned PDF converts “successfully” into nothing, because its pages are pictures of text, not text. These are real results from our test files:

The tool does not hide this. Each such file gets a warning in the result, worded like “scan-lease.pdf has no text at all: probably scanned. Run OCR PDF first to make the text readable, then convert it”, or for a mixed file “has no text on page 4”. A page that is only a photo or a diagram gets the same note, so read it as “nothing to read here” rather than proof of a scan.

To find them in a big batch, look down the size column in the result list: anything of a few bytes, or far smaller than files of a similar page count, is your list for OCR.

Send the scans through OCR, then convert again

1. Collect the PDFs the warnings named.
2. Open OCR PDF, drop them all, pick the language they are written in, and press Make it searchable. It adds an invisible text layer and leaves the page pictures as they were.
3. Drop the OCR results on PDF to Text again.

OCR is a guess, and the tool says how sure it is for each page. On a scanned bank statement it averaged 94 % confidence in our tests. The guide to making scans searchable covers languages, confidence and handwriting.

Text or Markdown for an AI pipeline or a search index

PDF to Text keeps each printed line as a line and keeps everything on the page, including the running header and the page number on every page. That is right for search indexes and for diffing, less right for anything that reads meaning.

PDF to Markdown takes the same batch and rebuilds the structure: headings from larger font sizes, paragraphs from line spacing, lists from bullets and numbers, tables as Markdown tables, and words broken by a hyphen at a line end joined again. Running lines (headers, footers and page numbers that repeat on most pages) are left out by default. On our 4-page test lease, the plain text counted 1,132 words and the Markdown 1,092: the 40 words of difference are the header and “Page 1 of 4” lines that every chunk of a pipeline would otherwise carry. If you feed the files to a language model or a notes app, Markdown is the better batch.

Spot-check 1 hard file before trusting the batch

The tools take the text in the order the file stores it, which is the order the creating program drew it. For letters, reports and most exports that is the reading order. For some 2-column layouts it is not: on a test page whose program drew the 2 columns line by line across the page, the text came out with both columns mixed, line by line, and Markdown and Word followed the same order. In the .txt, tables come out as lines of text; PDF to Markdown writes them as Markdown tables.

So before you rely on 2,000 files, open the text of the most complicated one (a newsletter, a 2-column paper, a form) next to its PDF. If it reads right, the batch is probably fine; if it does not, the guide to jumbled PDF text explains what you are seeing and what can and cannot fix it.

Two more limits. A PDF with a password to open is skipped in a batch (“It has a password. Run it on its own to type the password.”), and the folder structure is not kept: every .txt lands in 1 zip or 1 folder.