How to make a scanned PDF searchable
Text recognition reads the words off each page image, and they are written back into the PDF invisibly, in the place they were read from. The file still looks exactly like a scan, but find-in-page, copy and paste, and screen readers all work on it.
This is the one thing the web tools here do not do — it is in the Scanly Android app. Said plainly rather than pretending otherwise.
Why a scan is not searchable to begin with
A scanner produces a picture of a page. To a computer it is pixels: there is no letter 'e' anywhere in the file, only dark shapes. Searching it finds nothing because there is nothing to find.
What a text layer actually is
Every recognised word is drawn onto the page a second time, in a colour with zero opacity, positioned where the ink is. You see the scan; anything that extracts text sees the words. It is an old, completely standard technique, and it is why a searchable PDF looks identical to the scan it came from.
What recognition gets wrong
Handwriting, mostly. Also poor lighting, unusual layouts, tables, and any script the model was not trained on. Recognition is a machine reading a photograph, and on a bad photograph it reads badly. A good tool tells you how much it managed rather than implying it read everything.
The language limit worth knowing
On-device recognition usually ships one script. Scanly's reads Latin — so an English or Urdu-in-Roman page works, and a page in Urdu's own script exports as a perfectly good picture with nothing searchable in it. The app says so rather than letting you discover it in a viewer.