@mdickens We mainly plan to use ads targeting different narrow audiences; and then to compare the impact different messages have on the engagement and on actions people than take on a website (we’ll also be asking them to complete surveys, though that won’t be very informative due to selection effects).
There are downsides (the feedback is somewhat low-resolution, social media algorithms might add noise as they won’t be showing the ads to random samples of people), but it seems much cheaper than using mechanical turks/focus groups and provides much shorter feedback loops.
(I also shared a bit on our longer-term strategy and how we'll use our results with Michael.)