How to edit text in a scanned PDF
A scan has no text in it, only a picture of text, which is why editors refuse to let you click a word. Here is how recognition turns it back into something you can retype.
Run OCR on a scanYou open a scanned contract, try to click a word, and nothing happens. No caret, no selection, no way in. This is not the editor being difficult. There is genuinely no text in the file to click.
When a page is scanned or photographed, the result is one large image. To your eye it reads as words; to the software it is a grid of coloured dots with no more structure than a photograph of a beach. There are no characters, no fonts and no word boundaries — nothing to select, because nothing in the file knows that the shapes are letters.
How to tell whether your PDF is a scan
Open it in any viewer and try to select a line of text with your mouse. If you get a text cursor and can highlight individual words, the file has real text. If you get a selection box that covers the whole page as one block, or nothing happens at all, it is an image.
PDF Clay checks this for you on upload. It looks at how much of each page is covered by an image and how many characters are present, and when a page is mostly picture with almost no text it offers to run recognition instead of leaving you clicking at nothing.
What recognition actually does
Optical character recognition renders each page at 300 dots per inch, examines the shapes, and works out which characters they represent. It then writes those characters back into the PDF as a real text layer positioned exactly over the picture — and marks that layer invisible.
That invisibility is the clever part. You still see the original scan, pixel for pixel, so the document looks untouched. But the text is now in the file: selectable, searchable, and editable. Every "searchable PDF" you have used works this way.
Pages that already contain text are copied through untouched. Re-recognising a page that was never scanned can only make it worse, because it would replace crisp text with a picture of text.
Editing after recognition
Once the text layer exists you can click a word and retype it like any other PDF. There is one difference worth knowing about. On an ordinary page, deleting text removes the characters and leaves the background alone. On a scan the words are part of the image, so removing the invisible text layer would leave the original still visible underneath your replacement. Instead, the area is painted with the colour sampled from the paper around it, and the new text is drawn on top.
This is why edits on a scan look best on plain backgrounds, and why a heavily patterned or discoloured page can show a faint patch where the original word was. It is a real limitation of working with a picture rather than text.
What affects the quality of the result
- Resolution. A scan made at 300 dpi or better recognises well. A phone photo at an angle, in poor light, does not.
- Straightness. Skewed pages produce noticeably more errors than square ones.
- Typeface. Clean printed text is read accurately; handwriting essentially is not.
- Contrast. Faded photocopies and grey-on-grey forms are the hardest cases.
Recognition is never perfect. Read the result before relying on it, particularly for numbers — confusion between similar characters is the most common failure, and a wrong digit in an invoice total is worse than an obvious mistake in a word.
Limits here
Recognition is limited to 50 pages per run, because each page costs real processing time and memory. If you have a longer document, split it first and run the parts separately. English is the language currently installed.