They know much more. What you liked or disliked, which ads you clicked on, which ads you closed immediately, which you waited till they finished. When you added people as friends. Which network were you in at the time. What browser/system/device were you using. How fast was your internet. How much battery you had. How often are you using their site, how it changes through the day and through the year, have you changed your timezone.

Each of these is insignificant and noisy, but when aggregated it allows 99.9% predictions on stuff like "who are you voting for", "how much do you earn", "are you expecting a child", "are you buying a house/car/trip/anything significant soon", "how's your health", "are you sleeping well", "where do you live", "are you on a trip", etc.

People get ads about health issues before they realize they have these issues, because the Algorithm knows better.

When you have enough noisy uncertain data it doesn't matter that it's uncertain and noisy.

"Did Jane smith go to <some school/university>?"

"Did Jane attend <event>?"

"Do you work with Jane an FooCorp?"

etc etc etc