Several of the examples they cite from political violence to Holocaust denialism I saw a lot of friends share. As with most things polling related, if one poll sounds too good to be true it probably is.
Opt-in polling differs from probability-based polling in that the respondents sign up for the survey, rather than being randomly sampled from the population like a traditional poll. Opt-in polls often use financial rewards to get people to participate. Response bias is an issue with all polling, but with opt-in polling you get a self-selection that is even harder to assess and control for.
As they note, bogus respondents are typically overrepresented in young adult and Hispanic samples.