In the wake of the 2016 election, Facebook had a problem. A lot of people thought the site had done too little to combat fake news. At the same time, Facebook was desperate to maintain its reputation as a politically innocuous technology company. If it started directly declaring certain news stories fake, it would inevitably be drawn into emotionally charged political fights, damaging its credibility with partisans in the process.
So instead of trying to identify fake news itself, Facebook is outsourcing the work to third parties. Initially, five fact-checking organizations — Snopes, PolitiFact, Factcheck.org, ABC News, and the Associated Press — will be asked to review questionable articles and render a verdict about their veracity. If an article is rated false by these organizations, it will be marked as “disputed” and be pushed lower in a user’s newsfeed.
That puts some distance between Facebook and the politically charged question of what counts as fake news. But it’s not clear if this will be sufficient to mollify critics — predominantly on the right — who object to the very concept of trying to weed fake news out of the newsfeed.
At the same time, Facebook is giving fact-checking organizations a kind of power they’ve never had before: the power to publicly brand other websites’ stories as “disputed” and push them down in Facebook users’ newsfeeds. Facebook’s new fact-checking system is going to subject these organizations — some of them quite small — to an unprecedented amount of public scrutiny.
Some conservatives distrust fact-checking sites
The problem is that — especially in the middle of a political campaign — what’s a “fact” is often hotly disputed.
“PolitiFact consistently calls Republicans liars at two or three times the rate of Democrats,” wrote Mark Hemingway in the conservative Weekly Standard in 2013. “Its individual judgments are regularly erroneous in ways that make it hard not to suspect the organization has a serious problem with political bias.” In a Friday piece reacting to Facebook’s announcement, Hemingway wrote that PolitiFact “might well be the most biased news organization in America.”
Of course, fact-checking organizations deny that they’re biased. “Our challenge to folks who say PolitiFact is biased is: ‘Give me examples,’” says Aaron Sharockman of PolitiFact, a fact-checking group associated with the Tampa Bay Times. “Often we get a lot of silence back.”
But PolitiFact’s conservative critics say they’re happy to supply examples of apparent bias. In a recent critique of fact-checking websites at Current Affairs, Nathan Robinson pointed out a case where PolitiFact had rated as “mostly false” Donald Trump’s claim that “windmills are killing hundreds and hundreds of eagles.” PolitiFact interpreted this as a claim that windmills are killing hundreds of eagles per year, which is probably false.
But as Robinson points out, Trump didn’t say “per year,” and over multiple years windmills almost certainly have killed hundreds of eagles. And PolitiFact’s nitpicking of Trump’s statements contrasts with the more generous attitude it sometimes takes toward the statements of other politicians.
For example, PolitiFact rated as “mostly false” Donald Trump’s claim that the real unemployment rate among black people ages 18 to 24 was 59 percent. Trump was likely referring to the fact that the employment-to-population ratio for black youths was 41 percent, leaving 59 percent outside the workforce. But that’s a different figure from the unemployment rate.
Yet when Bernie Sanders made a very similar mistake a year earlier, claiming that the “real unemployment rate” among teens ages 17 to 20 was 51 percent, PolitiFact rated this “mostly true.” True, Sanders was using the wrong terminology, it said, but the larger point Sanders was making was basically true: Young black people were having more trouble finding work than young white people. This, of course, was also the point Trump tried to make a year later, but PolitiFact was not in such a forgiving mood when he made this particular gaffe.
“Critics point to one or two or three or even 10 statements over a nine-year span,” Sharockman says, but he points out that his organization has conducted more than 13,000 fact-checks in its history. He argues that the isolated examples conservatives like to cite don’t reflect the organization’s larger body of work.
Sharockman also says that PolitiFact’s critics are not limited to the right. Liberal cable TV host Rachel Maddow has devoted several segments to blasting the organization for what she views as mistakes. For example, PolitiFact initially rated Marco Rubio’s claim that most Americans are conservative “mostly true,” even though most polls show that a mere plurality — but not a majority — of voters describe themselves that way. Maddow thinks the claim should have been rated false — PolitiFact eventually modified its rating to “half true.”
“People on the edges of a particular ideology — I think this applies in both directions — they see the world in much starker terms than PolitiFact does,” Sharockman says. “For a lot of really partisan conservatives, and for progressive liberals, they see things as true and false and don't believe there are shades of gray in between.”
Fact-checkers are taking on a new role
Of the fact-checking groups selected by Facebook, PolitiFact is the one that has drawn the most ire from conservatives. This may be because its mission of vetting the statements of candidates for office regularly puts it in the middle of partisan political disputes. Factcheck.org has a mission similar to PolitiFact’s but has not drawn as much conservative criticism.
Another Facebook-selected group, Snopes, is more focused on debunking urban legends and bogus email forwards — sometimes, but not always, about politics. The final two groups, the AP and ABC News, are conventional news organizations.
The optimistic case for Facebook’s fact-checking project is that identifying fake news is an easier and less politically fraught problem than determining when politicians are lying. A common problem for fact-checkers is that their calls are often tied up with the merits of the underlying issues candidates are arguing about.
In a 2012 piece, for example, Hemingway faulted PolitiFact for giving Mitt Romney a “pants on fire” rating for claiming that Obama was planning to gut the work requirements of the 1996 welfare reform law. Unsurprisingly, a conservative expert thought Romney’s characterization of Obama’s proposal was fair game, while more liberal experts thought Romney’s criticism was unfair.
It might seem like identifying fake news — at least the kind that’s made up out of whole cloth — would be more straightforward.
But if you dig a little deeper into real fake-news controversies, you quickly find that distinguishing real stories from fake ones can require the same kinds of subjective judgment calls required to distinguish between politicians’ truthful statements and their lies.
Consider, for example, BuzzFeed’s list of the five biggest fake news stories in the three months before the election. The top story, Pope Francis endorsing Donald Trump, is obviously fake. But the second and third stories claim that WikiLeaks emails show Hillary Clinton selling weapons to ISIS, and here the verdict isn’t so clear.
Snopes examined this claim and found that it contained a “mixture” of true and false claims. There is zero evidence that Clinton directly and knowingly sold weapons to ISIS, Snopes concluded. But WikiLeaks founder Julian Assange has claimed that leaked documents show Clinton’s State Department selling weapons to militants in Libya, some of which ultimately ended up in the hands of ISIS fighters.
Assange’s evidence for this claim seems flimsy. But this doesn’t seem like “fake news” in the sense of being deliberately made up. Whoever wrote that story may have really believed that Clinton’s policies led to ISIS fighters getting US weapons.
Another story that was widely branded as fake news claimed that Pepsi CEO Indra Nooyi told Trump supporters to “take their business elsewhere” a few days after the election. Nooyi never actually said those words. But she did make comments expressing dismay at Trump’s election — comments that conservative sites quickly reported on. In a newsgathering game of telephone, one publication paraphrased Nooyi as telling people to take their business elsewhere, which others then misinterpreted as a direct quotation.
Snopes rated this “mostly false” but acknowledged it wasn’t completely false — Pepsi’s CEO really had made negative comments about Trump’s election. And it’s not obvious that anyone was trying to be deceptive here; it was just extremely sloppy journalism.
When they evaluate articles for Facebook, these organizations won’t have a “half true” or “mostly false” option. They’re going to have to decide which news stories to rate as fake — thereby branding them as “disputed” in the Facebook newsfeed — and which ones to leave alone.
Sharockman says PolitiFact plans to be very conservative about this.
“We're looking for clear calls,” he says. “We're going to look for things that we think are not close.” If an article has shades of gray, he says, PolitiFact will skip it.
But of course, what looks like a shade of gray to one person might look like a clear falsehood to others. No matter what PolitiFact and other fact-checking organizations do, they’re likely to make someone unhappy.