How COVID-19 pushed Twitter to fact-check Trump’s tweets

It was the pandemic, Twitter says, that freed the company to attach fact-check warnings to a pair of U.S. Present Donald Trump’s tweets this week.

Critics have complained for years that Twitter lets Trump run wild on the platform. But the company had generally taken a hands-off approach to the president, partly because of a company policy that considers it in the public’s interest to know what world leaders are thinking, and partly because Twitter judged many of Trump’s tweets to fall into a gray area not covered by its rules banning specific behaviors like abuse or posting hateful content.

Twitter says the coronavirus outbreak prompted it to re-evaluate its approach to these gray-area tweets and treat some of them as potentially dangerous.

“Covid was a game changer,” said Twitter spokesperson Katie Rosborough.

“We now have the tools in place to label content that may contain misleading claims that could cause offline harm,” said Rosborough. Those tools include the warning labels attached to Trump’s tweets that link to a Twitter “Moment,” or collection of content, explaining objections to his post.

Twitter has in recent months been quietly laying the groundwork for further questioning of Trump’s tweets.

Twitter has not yet defined what generally counts as offline harm, but in the case of coronavirus-related content, it has identified tweets like those that advocate protecting yourself from the virus using methods public health authorities say are ineffective.

Now Twitter is applying a similar metric to non-coronavirus-related tweets by the president of the United States.

On Tuesday, Twitter added warning labels to two Trump tweets claiming, without evidence, that mail-in ballots, like those being used in California amid the pandemic, are likely to be “substantially fraudulent.” The labels directed users to a link to “get the facts about mail-in ballots.”

That decision has triggered a high-pitched confrontation with Trump, who vowed Wednesday to take “big action” against social media companies that he accuses of silencing conservatives. White House Press Secretary Kayleigh McEnany later said Trump will sign an executive order Thursday aimed at social media companies, amid a flurry of calls from GOP allies for reducing the companies’ protections from lawsuits.

But Twitter has in recent months been quietly laying the groundwork for further questioning of Trump’s tweets — possibly setting the stage for even harsher collisions with a president and his supporters already raging against the San Francisco company.

That said, Twitter has at least one reason to not be as shaken by such threats as bigger players in social might be. Unlike Facebook and Google, the company is too small to trigger the scrutiny of federal antitrust authorities who could be leaned on by an angry president.

The company has made other recent moves in the political arena meant to inoculate it against complaints that it’s destabilizing democracy in the United States but that have also served to anger some on the right, including banning political advertising altogether.

The coronavirus lesson for Twitter was that the company’s trust and safety team, which is charged with taking action on inappropriate posts, could selectively challenge tweets that don’t violate its stated rules — the black-and-white laws of Twitter that gets tweets deleted for specific behaviors like encouraging violence — but that wade into grayer territory, where by misleading or confusing people they can cause real-world harms. In those cases, the supposed worst-of-the-worst get tagged with additional context from around the web, summarized by a specialized Twitter content team.

Under the new approach, for example, Twitter added warning labels to tweets alleging that the novel coronavirus was spread not by human contact but the wireless technology 5G.

“Our research has shown people don’t want us to decide what’s true for them, but they want us to provide further context when possible,” Rosborough said.

In the case of the Trump tweets on mail-in ballots, part of the harm calculation was their timing: Trump was calling mail-in ballots “fraudulent” right at the moment some U.S. states are using them to administer elections amid the pandemic, the company said.

On the left, many argue that Twitter is only taking the weakest possible action in response to rampant misinformation.

But while Twitter may see the decision as a natural evolution, this week’s tagging of the Trump tweets about mail-in ballots marks a change in the company’s posture toward the president, and could be the beginning of a new escalation of tensions. It was the very first application of Twitter’s new approach on misinformation to tweets not having to do with COVID-19 — much as, back in February, the company first applied a three-day-old policy on so-called synthetic media to a tweet from Trump aide Dan Scavino featuring a clipped video of Joe Biden.

And the Trump warning labels have significantly ratcheted up tensions between Trump and Twitter.

Trump denounced Twitter over its flagging of his mail-in-ballot tweets, and POLITICO’s Playbook reported Wednesday afternoon that the White House is working on a threatening letter to the company. Twitter declined to comment on both that would-be missive and the reports of an imminent executive order.

On the left, many argue that Twitter is only taking the weakest possible action in response to rampant misinformation from the White House and ignoring many tweets that are just as egregious. The same day that Twitter decided Trump’s voting tweets went too far, it declined a request by the husband of a deceased Florida woman whom Trump had suggested with no evidence had been murdered by MSNBC host Joe Scarborough. The company said the decision to leave Trump’s Scarborough quotes alone was made quickly, as they were judged to not be in violation of any of Twitter’s existing policies.

But Twitter left a hint this week that it may continue to expand the categories of tweets it evaluates with this new approach to tweets like the one Trump levied at Scarborough.

In expressing sympathy to the widower, Twitter gave a bit of insight into its thinking in a line that looked at first like a throwaway: “We’ve been working to expand existing product features and policies so we can more effectively address things like this going forward, and we hope to have those changes in place shortly.”

Skeptics will say that Twitter remains, as it has been throughout its 14-year existence, haphazard about how it creates, articulates and enforces its policies. And its new handling of misleading tweets isn’t likely to do much to quiet those concerns. For one thing, the policy isn’t articulated anywhere in the site’s stated rules.

Rep. Ro Khanna, a Democrat representing a considerable slice of Silicon Valley, gave voice to some of that argument in a tweet of his own Wednesday: “This disagreement is exactly why we need thoughtful policies to consistently limit misinformation, instead of ad hoc fact checks whenever the headlines push Twitter hard enough.”

Twitter first announced its approach to coronavirus misinformation in March, just as the COVID-19 crisis was heating up, in a blog post from its head of policy and legal, Vijaya Gadde, and customers lead Matt Derella.

Trump speaks during a press briefing with members of the White House Coronavirus Task Force team | Drew Angerer/Getty Images

Twitter said then it would expand its definition of the kind of “harm” that could get a tweet knocked off the service to include those that conflict with public health authorities’ stated guidance on the coronavirus. By way of example, Twitter said that could encompass tweets suggesting that social distancing doesn’t work to slow the spread of the virus.

Twitter quickly tested out that new policy on at least one world leader: At the tail end of March it deleted coronavirus-related tweets from Brazilian President Jair Bolsonaro, judging them to be so harmful that they didn’t even merit protection under Twitter’s policy of placing a label on rule-breaking tweets by world leaders. (Twitter’s announcement of that policy last June was immediately assumed by many observers to be targeted at Trump.)

On May 11, Twitter announced that in addition to its flagging of questionable coronavirus content, “we may use these labels and warning messages to provide additional explanations or clarifications in situations where the risks of harm associated with a Tweet are less severe but where people may still be confused or misled by the content.”

That post sets up three categories of tweets where it might step in: “misleading information,” “disputed claims” and “unverified claims.” The Trump tweets on mail-in ballots, the company said, were the first instance since the implementation of the new approach in which it became aware of a piece of misleading information posted by the president concerning enough to take on under the new framework.