I have explained the details in other comments, have a look. But you can start by looking at pdftotext from Poppler, it is ready to go for 60-70% of cases with -layout flag, with bbox-layout you get even more details.
The advantage of the OCR method is that it effectively performs that visual inspection. That's why it is preferable for PDFs of disparate origin.
I did read your comment, because my intention here is to learn. I already described how tools such as pdftotext do not produce strings when each letter is positioned independently. I even gave an example of a few replies up.