User-engagement optimizing "tricks" used by social networks and news sites will stop working eventually, and leave users with a bad taste of your company. Using data mining and A/B testing we are basically enrolling all users in a constant social science experiment so they can help us better hack their own brain. But following it blindly is incredibly short-sighted.

Yes it works, but every time you trick someone it ekes away at their trust and happiness using your platform. But this occurs over many such interactions so it won't show up in your metrics until much later, and then only in general stats so it's nearly impossible to measure the effect with the time scales that we actually record and test. Heck, the users themselves probably won't even notice anything specific but just have a slowly growing distaste for interacting with your platform.

Anecdotally, I've noticed this happening to me and people I know on many news and social sites. (Though, my sensitivity to this is turned up really high.) But this effect can help explain the cyclic nature of social networks, and perhaps Facebook's recent decline in the US.

Do your users come away from your site feeling good about their engagement? Assuming more is better may be hurting you in the long run.