You open a scanned document, try to search for a word, and nothing is found. Nothing is broken: the file contains a photograph of a page, and a photograph has no words in it — only pixels that happen to look like words.
Text recognition, usually called OCR, is what turns those pixels back into characters. How well it works depends much more on the scan than on the software.
OCR reads the shapes in the image and outputs text. That text is what you can search, copy and edit. Depending on the tool you either get the plain text, which is what you want when the point is to reuse the content, or a new PDF with an invisible text layer placed over the original image, which is what you want when the page must still look like the scan.
Either way the original image is not improved. A blurred scan stays blurred; only the recognition of its characters is added.
Before blaming the software, check the input. In rough order of impact:
Recognition engines work from a language model, so telling them the right language improves accuracy noticeably. A document that mixes scripts — German with English quotations, or Persian with Latin names — needs an engine that can read both in one pass, otherwise one of the two comes out as gibberish.
Abzar's OCR covers English, German, Persian, Arabic, Russian, French and Spanish in a single automatic pass, and lets you narrow it to specific languages for a faster, more accurate run when you know what is on the page.
Standard OCR is trained on printed type. Handwriting varies far too much between people for the same approach to work, so expect poor results — neat block capitals sometimes come through, cursive rarely does. If a handwritten document matters, typing it is faster than correcting recognition errors.