I was under the impression that they do not have a benefit over placebo for most patients - only for the most depressed patients. And also some risk of worsening things for younger patients.

I think I got this impression from reading Bad Science, which had references IIRC. Example column from the blog version:

http://www.badscience.net/2008/01/washing-the-numbers-sellin...

http://www.amazon.com/Bad-Science-Ben-Goldacre/dp/000728487X