I’ll just copy/paste my reply to a different comment:

Scanning every jpeg on the internet with OCR would require far more compute than simply doing a regex for plaintext.

Also that white line can really fuck with most lightweight OCRs