Editing a scanned PDF

Open it and try to select a single word with the mouse. If the highlight jumps across the whole page as one block, or nothing highlights at all, there is no text in there. The other check is faster still: press Control F or Command F and search for a word you can plainly see on screen. No result means no text layer.

Mixed documents are common and they confuse people. A typed contract with one signed page photographed and dropped back in will behave normally on nine pages and refuse to co-operate on the tenth.

A scan is a photograph of a page. The file holds one large image per page and no characters at all. What looks like the letter a is a group of dark pixels that happens to be shaped like one. There is no text object to select, no font, no word boundaries, nothing for a cursor to sit inside. No editor anywhere can put you into text that does not exist.

Turning the picture back into characters is OCR, optical character recognition. Software looks at the shapes, compares them against what letters usually look like, and writes down its best guess. It is a guess, which is why the result always needs a read through.

Be clear which one you actually have, because they are not the same job. If you want to search the document, copy text out of it, or let something else read it, you want the file made searchable. If you want to change a line that is printed on the page, you want that page turned into editable text.

The first is the free OCR tool on this site. The second happens inside the editor and is a Pro feature. Both run on your own machine.

Drop the scan in and every page is read, then rebuilt as an image with an invisible layer of recognised text sitting exactly over the words. It looks identical. Search finds things. Copy and paste works. So does anything downstream that needs to read the file.

Two trades to know about. The file gets bigger, because every page is now a picture at roughly 144 dots per inch. And any page that was already real text becomes a picture of text. So run it on scans, not on documents that were fine to begin with. Recognition is English only today, and it costs nothing.

Open a scan in the editor and it notices the page has no text, then asks whether to read it. It never runs on its own, because recognition is slow and it should be your call. Say yes and the recognised lines come back as blocks you can click and retype, exactly like text that came with the document.

It works one page at a time, so you spend the time only where you need it rather than on all sixty pages. This is a Pro feature at $3.99 a month. Expect to correct it: recognition mixes up zero and capital O, one and lowercase l, and it struggles badly with handwriting, stamps, faint carbon copies and anything sitting at an angle.

Step back from the tooling for a second. You usually do not need the old text to become editable. You need the page to end up saying the right thing.

Whiteout a box over the wrong part, then use the Text tool to type the right part in the space. Signing a scanned form, filling in a printed application, adding a date, correcting a number: this is faster than any recognition step, it works on the free plan, and there is no guessing involved because you typed it. Zoom in first so the new line sits on the ruled line rather than floating above it.

Smart Fill covers a specific and very common job: your PDF has real form fields, and the details it wants are printed on something you can photograph, a licence, a utility bill, an ID card.

Take the photo, and the recognition runs in your browser. The text it reads is then sent to our server for the step that works out which value belongs in which field, and the filling happens back in your browser. So the picture stays on your device and the recognised text does not. That is the honest description, and worth knowing before you point it at an identity document. The free plan covers one fill, and the PDF has to have real fields in it: a form that was printed to PDF with nothing but ruled lines needs the Text tool instead.

This is the cheapest improvement available and almost nobody makes it. Scan at 300 dots per inch, keep the page flat and square to the glass, use even light with no shadow from your own phone, and choose grayscale rather than colour for plain text. Recognition accuracy on a clean 300 dpi scan is in a different league from a phone photo taken at an angle in a dim room. If you can put the page through a scanner again, do that before you fight the software.

One thing scans are actually good for. Because the page is already pixels, the redact tool paints black straight onto them and there is nothing underneath to recover. If the scan was made searchable earlier, that recognised text layer is dropped when the page is rebuilt, which is exactly what you want under a black box. Honest limit worth planning around: large scans are slow everywhere. A hundred page scanned file takes a while to open, and recognition adds time on top, page by page.