Find out which pages of your PDF need OCR

How it works

1. Upload the PDF you cannot search. Nothing is written into it. If you already know roughly where the scanned section is, set a range — the report then answers a narrower question instead of handing you a list of two hundred pages.
2. Read the three lists. The notes name the pages that already carry real text, the pages that are pictures and would need OCR, and the pages that are blank. A total character count tells you how much readable text was found across the whole document.
3. Take the list to something that can OCR. Your scanner’s own software, Acrobat’s Recognise Text, or the free ocrmypdf command-line tool. Extract just those pages first if you are paying per page or waiting on a queue.

What the options do

The form above lists the options this tool is running today, straight from the server. This is what they are for.

Pages to check

Which pages to look at, as 1-20 or all. Narrowing it is a way of asking a sharper question — are the appendices scanned, is page 14 the problem — rather than reading a long list to find out. Pages outside the range are not mentioned in the report at all.

Characters that count as a real text layer

How much text a page has to carry before it is counted as text rather than as a picture. Twenty is right for prose, because a scanned page often carries a stray character or two from a stamped page number or a header added afterwards, and two characters are not a text layer. Raise it when a badly recognised document is reporting pages as text with nothing usable on them; lower it for sparse tables and title pages where a genuine page really does have ten characters on it.

What this tool does not do

• It does not add a text layer, and it does not change your file: the PDF you download is the one you uploaded, byte for byte.
• It cannot tell you whether an existing text layer is any good. A page recognised badly years ago counts as text here, because there is text on it — the character count is the only clue you get.
• It does not straighten or turn the pages. Deskew handles a crooked scan and rotate handles one that is on its side, and both are worth doing before an engine reads it.
• A blank scanned sheet is reported as blank rather than as needing OCR, because there is nothing on it for an engine to find.
• It does not extract the text it finds. PDF to text does that, for the pages that already have some.

What happens to your file

The file is uploaded because OCR PDF runs on our server rather than in the browser tab. It is kept for at most 2 hours and the result you download for 24 hours, then a scheduled job deletes both — a job has to be able to read the file while it runs, and you have to be able to retry a failed conversion without re-uploading. Neither needs longer.

Nothing about the contents is logged, and no one reads your documents. The retention page lists the same numbers the code enforces, and the privacy policy says who can reach the storage they sit in.

Questions

Why can this not do the OCR itself?

Because character recognition needs an engine and its language data, and neither is installed on this server. Shipping a tool called OCR that quietly hands your file back unchanged would be worse than saying so plainly, so it does the part that needs no engine and names the pages that do.

What should I use to actually add the text layer?

Your scanner’s software will often do it at the moment of scanning, which is the best time. Otherwise Acrobat’s Recognise Text, or ocrmypdf, which is free, runs on Windows, macOS and Linux, and takes a page list like the one this report gives you.

How does it decide a page is a picture rather than text?

It pulls the text out of the page and counts the characters that are not whitespace. Below your threshold it looks for an image on the page: if there is one, the page is a scan; if there is not, the page is blank.

Should I run OCR over the whole document or only the listed pages?

Only the listed ones. Recognising a page that already has real text usually makes it worse, because it lays a second, guessed copy of the words underneath the good one and searches then match both.

Every page came back as text, but I still cannot search the document.

Then the text is there and something else is wrong — the reader’s search, a font with a broken character map, or a text layer that was recognised into gibberish. Run those pages through PDF to text: if what comes out is unreadable, the layer is bad and re-recognising it is the fix.

Does the check leave anything behind in my file?

No. The file is opened, read and handed back as the same bytes. The report exists only in the notes beside the download.