A campaign has written a few adverts about one policy. It can send one it believes in, test several and send the winner, or use personal data to send different people different messages. That last one is what "microtargeting" usually means.
Which moved opinion most? Tap them in order, best first.
Tap a card to place it. Tap it again to take it back.
One word, two jobs. Choosing whom to spend on, and testing which message works on everybody, is the half with evidence behind it. Writing a different message for each person is the half that gets sold, and the half that keeps failing to show up.
The same authors found "no evidence that targeting messages by more than one covariate yielded additional persuasive gains": more data about the voter bought no more persuasion.