Labelling tweets featuring false claims about election fraud as “disputed” does little to nothing to change Trump voters’ pre-existing beliefs, and it may make them more likely to believe the lies, according to a new study.

The study, authored by John Blanchar, an assistant professor from the University of Minnesota, Duluth, and Catherine Norris, an associate professor from Swarthmore College, looked at data from a sampling of 1,072 Americans surveyed in December of 2020. The researchers published a peer-reviewed paper on their findings this month in the Harvard Kennedy School’s Misinformation Review.

“These ‘disputed’ tags are meant to alert a reader to false/misinformation, so it’s shocking to find that they may have the opposite effect,” Norris said.

Participants were shown four tweets from Donald Trump that made false claims about election fraud and told to rank them from one to seven based on their truthfulness. A control group saw the tweets without “disputed” tags; the experimental group viewed them with the label. Before and after seeing the tweets, the subjects were also asked to rank their views on election fraud overall.

The study found that Trump voters who were initially skeptical about claims of widespread fraud were more likely to rate lies as true when a “disputed” label appeared next to Trump’s tweets. The findings meanwhile showed Biden voters’ beliefs were largely unaffected by the “disputed” tags. Third-party voters or non-voters were slightly less likely to believe the false claims after reading the four tweets with the tags.

Blanchar and Norris had expected in their study that the disputed tags would produce little change in Trump voters with high levels of political knowledge, given that previous research had shown politically engaged people can dismiss corrective efforts in favor of their own counterarguments. The researchers did not predict the opposite possibility: corrective as confirmation. The knowledgable Trump voters surveyed were so resistant to corrections that the factchecking labels actually reinforced their belief in misinformation.

“Surprisingly, those Trump voters with higher political knowledge actually strengthened their belief in election misinformation when exposed to disputed tags, compared to a control condition without tags,” Blanchar said. “Instead of having no impact, the tags seemed counterproductive, reinforcing misinformation among this group.”

Previous studies and research from disinformation experts have argued that directly challenging conspiracy theorists’ beliefs can be counterproductive, leading them to withdraw or double down on their convictions. While Blanchar and Norris state in the study that their findings do not necessarily prove this backfire effect is universal – since the sample size of Trump voters in the study was relatively low – they are more confident that disputed tags are less effective the more politically knowledgable Trump voters become.

Social media platforms have tried for years to create various kinds of labeling systems that signal to users when content contains false, misleading or unverified claims. Twitter/X formerly labeled some tweets with false information as “disputed”, a practice it has in recent years replaced with its “community notes” peer review feature and a more lax attitude toward content moderation overall.

A larger question that misinformation researchers have sought to answer is whether labels and factchecks attempting to debunk falsehoods are actually effective, in some studies finding the potential for these warnings to actually backfire. The field of research has implications for social media platforms, news outlets and initiatives aimed at preventing misinformation, especially at a time when political polarization is high and false claims of election fraud are pervasive.

The authors assessed political knowledge by asking participants 10 questions to test general understanding of US politics, such as: “What political office is now held by John Roberts?”

One limitation of the study is the unique time frame when it was conducted – the height of the 2020 election, when conservatives had more antagonistic views toward Twitter. Since the study was conducted, Twitter has not only gotten rid of the “disputed” tags but undergone a broader change in ownership, content moderation policy and user attitudes. After the Tesla CEO, Elon Musk, bought Twitter for $44bn in 2022 and renamed it X, the platform has brought far-right voices back on to the platform, including Trump himself, and taken a rightward turn that has led conservatives to see it in more positive terms.

“We can’t pinpoint why disputed tags backfired among Trump voters, but distrust of the platform may have played a role,” Blanchar said. “Given the conservative distrust of Twitter at the time, it’s possible Trump supporters saw the tags as a clear attempt to restrict their autonomy, prompting them to double down on misinformation.”

• This article was amended on 23 September 2024 to correct the spelling of John Blanchar’s name. It was initially written as Blanchard.