That’s a good point. I would imagine they break it up into pieces - in a reCAPTCHA sorta way - and any given person sees a sentence or a piece of a sentence.

An alternative would be to strip out all obvious known words and only leave unknowns (i.e., names) and then have those fragments reviewed (in a reCAPTCHA sorta way).

Finally, for images, cover all faces and the one by one decide which should remain covered and which should not.

LOTS of work but there are workflows to mitigate the ability for reviewers to connect more than they should.