What can we do? We find some evidence that fact-checking might eliminate liar's dividends. Another takeaway: While the public may be less willing to buy false claims of "deepfakes", rebuttals against real deepfakes may also be ineffective.

False claims of misinformation are also more effective than apologies and are at least as effective as simple denials. Of note, we find no reductions in trust in media and mixed evidence that the claims work by reducing belief in the scandal.

Yet, we also find support gains *across partisan subgroups*, rather than simply concentrated amongst co-partisans or core supporters of politicians. There's no evidence of a backlash effect amongst out-partisans either.

For those worried about deepfakes, we find, reassuringly, that claims of misinformation are mostly ineffective against video evidence of scandal. However, some recent evidence (Study 4) suggests that this could change, especially as awareness of deepfakes further increases.

Based on 5 survey experiments (total n > 15k), we find claims of misinformation increase politician support compared to non-response. For scandals reported via text, these claims are effective when using strategies we call "informational uncertainty" or "oppositional rallying".

We examine *misinformation about misinformation*, or politicians "crying wolf" over fake news. Can politicians maintain support after a scandal by falsely claiming that real news stories are fake news or deepfakes? Does that produce a benefit, or a "liar's dividend"?

8/ In a nutshell, our results highlight the significance of the text-based approach as a valuable complement to existing methods for detecting false news. It's more effective at uncovering falsehoods.

7/ Here's the kicker: our text-based approach outperforms the others in producing fewer false positives and only a small number of false negatives.

6/ Interestingly, demographic and political predictors of misinformation-sharing behavior remain consistent across these approaches.

5/ But here's the twist - there's little overlap between the approaches we compared. Our text-based approach leads to different conclusions about who shares misinformation and the type of false content they share.

4/ Surprise! Sharing false news among politicians is actually a rare occurrence. Less than 1% of their social media posts contain misinformation. Actually, it is not quite surprising; this finding is consistent with a handful of studies elsewhere.

3/ We've pitted our text-based approach against established domain-based and Facebook URL methods, focusing on Brazilian political leaders as a case study. We examine more than 900 politicians and about 4 million posts from Twitter, Instagram, and Facebook.

2/ Our method combines fact-checkers, supervised learning, natural language processing, and human review to spot falsehoods in textual content.

1/ We develop a new approach to detecting false stories, particularly those circulating as text without hyperlinks in the Global South.