pdftotext -layout splits table phrases, so grep misses them

When auditing a PDF for a phrase, grep over pdftotext -layout output can give a false negative if the phrase sits in a table cell. Cell text wraps across lines and interleaves with neighbouring columns, so a hyphenated term becomes 'no AI-' on one line and 'generated text' on the next, and the full-phrase pattern never matches. Search for each fragment separately, or run pdftotext without -layout and normalize whitespace and line-end hyphens first. Treat a no-match as unproven until you have done that, especially when using it to claim a PDF does not contain some text.