As social media platforms face a wave of political misinformation in the run-up to the US presidential election, many of them are leaning on one tool to slow the spread of misleading content: warning labels.
These labels, when affixed to a piece of potential misinformation, alert users that fact-checkers dispute the claims made in the post and point them toward better information. The hope is that users who see flagged content will be less likely to believe and share it.
Warning labels are a central part of the biggest social media platforms’ strategies for tamping down misinformation. Twitter uses them when a tweet mischaracterizes the truth to incite violence, delegitimize election results, or prematurely claim election victory. If a tweet contains a specific falsehood with the potential to cause harm, the company will go a step further and delete the tweet.
Facebook and Instagram rely on warning labels even more. They won’t delete any misleading posts, and will instead leave them up with a warning to protect freedom of expression. “We’ll allow people to share this content to condemn it, just like we do with other problematic content, because this is an important part of how we discuss what’s acceptable in our society,” CEO Mark Zuckerberg wrote in a Facebook post.
Warning labels do appear to work—at least when it comes to the story they’re attached to. Though an influential 2015 paper presented evidence for the so-called “backfire effect,” in which social media users who saw fact checks were more likely to believe false claims, subsequent studies have struggled to replicate the effect. Other studies show that labeling a story as false dramatically decreases engagement with that story.
But there’s disagreement over whether warning labels can change the overall rate of misinformation spread on a platform.
In March, researchers from Harvard, Yale, MIT, and the University of Regina published a study suggesting that warning labels could backfire by making other, unlabeled content seem more reliable. They found that when some false content is flagged, people judge all other content—true or false—to be more accurate. They called it the “implied truth” effect.
“I think most people working in this area agree that if you put a warning label on something, that will make people believe and share it less,” said David Rand, an associate professor of management and cognitive science at MIT’s Sloan School of Business and one of the paper’s co-authors. “But most stuff doesn’t get labeled, so that’s a major practical limitation of this approach.”
The researchers note that it’s much easier to produce false content than to debunk it, and fake news often floats around unnoticed for days before it’s finally flagged. As a result, the benefit of labeling a small subset of misinformation could be outweighed by the harm of lending an aura of legitimacy to the larger set of unlabeled misinformation. “The net effect of the warnings may emerge as an increase in misperceptions,” the academics wrote.
They suggest social media companies address that concern by adding “verified” labels on reliable sources of information, which in their study eliminated the “implied truth” effect. That makes the problem an economic one: It’s not so much a question of whether labels work once they’re applied, but how fast and at what scale you can apply them. And fact-checking is expensive.
Facebook and Twitter didn’t respond to emails seeking comment on Friday. In the past, Facebook has touted the fact that between March and May 2020, it labeled 50 million pieces of misinformation, and 95% of users didn’t click through the warning to see the flagged content. But Rand points out that click rates on Facebook feeds are generally very low, so if 5% of people did click to see a piece of misinformation, it’s not necessarily a win.
“The platforms need to be doing evaluations of the impacts of these interventions, and they need to be making those assessments public in a credible way,” Rand said.
To be sure, warning labels aren’t the only strategy social media companies are using in the run-up to the election. Facebook, Twitter, TikTok, and Snapchat are promoting election information from reputable sources in prominent spots on their apps and websites. And many platforms are making tweaks to their algorithms to show false content to fewer people and keep it from going viral. In the coming days, we’ll find out how effective this combination of interventions proves to be.
“I think social media platforms will continue to serve as one of the dominant places where political discourse in the United States happens,” said Dipayan Ghosh, who heads Harvard’s digital platforms and democracy project, “and they’ll continue to serve as tremendous conduits for misinformation.”