Unless the PDF has every letter placed by independent PostScript like commands or - worse - text has been imported from an SVG file.
pdfgrep is at best probabilistic and I stand by my comment..
Agreed. I work on a solution for extracting data from random documents ( invoices, payslips, you name it ) for natives pdf in the wild ( not scans) we gave up : and just rasterize to send to an OCR software... It's also probabilistic, but way more reliable.