Over a year ago, Twitter launched a pilot of an ambitious project that was meant to harness the wisdom of crowds to answer just these sorts of questions on its platform, potentially across countries and languages, in near real time. Called Birdwatch, it lets volunteer fact-checkers add notes to tweets that are going viral, flagging them as potentially misleading and adding context and reliable sources that address their claims. By crowdsourcing the fact-checking process, Twitter hoped to facilitate debunkings at a greater speed and scale than would be feasible by professional fact-checkers alone.
Yet after 13 months, Birdwatch remains a small pilot project, its fact checks invisible to ordinary Twitter users — even as its volunteer contributors dutifully continue to flag false or contested tweets for an audience of only each other. That suggests that either Twitter hasn’t prioritized the project amid internal upheaval and pressure from investors to grow faster, or that it has proved thornier than the company hoped.
A Washington Post analysis of data that Twitter publishes on Birdwatch found that contributors were flagging about 43 tweets per day in 2022 before Russia’s invasion of Ukraine, a microscopic fraction of the total number of tweets on the service and probably a tiny sliver of the potentially misleading ones. That’s down from about 57 tweets per day in 2021, though the number ticked upward on the day Russia’s invasion began last week, when Birdwatch users flagged 156 tweets. (Data after Thursday wasn’t available.)
Twitter said it has about 10,000 contributors enrolled in the pilot, which is limited to the United States. (One of the authors of this article, Will Oremus, joined the Birdwatch pilot so that he could report on how the project operated.) But its data indicates that just 359 contributors had flagged tweets in 2022, as of Thursday. For perspective, Twitter reports that it is used by 217 million people worldwide each day.
Asked why it hasn’t launched Birdwatch publicly, and whether it has a timetable for doing so, Twitter spokeswoman Tatiana Britt did not answer directly.
“We plan to scale up as we’re able to do so safely, and when it can help improve learning,” she said in an emailed statement. “Our focus is on ensuring that Birdwatch is something people find helpful and can help inform understanding.”
That would seem to imply the company has not yet figured out how to scale up Birdwatch safely or how to ensure it’s helpful. Twitter indicated it will have more information to share about it soon.
Twitter itself sometimes appends fact-checking labels to a few limited categories of misleading tweets, including misinformation about the coronavirus and voting in elections. Its curation team, a small editorial division within the company, occasionally highlights debunkings of viral rumors within Twitter’s “trending” features. Last week, during Russia’s invasion of Ukraine, that team created a “moment” — a curated collection of tweets — focused on correcting or contextualizing misleading tweets about the conflict, such as tweets purporting to show an ace Ukrainian fighter pilot nicknamed the “Ghost of Kyiv.” (Some of the videos were actually taken from a simulation game.) In August, the company announced its first partnerships with professional fact-checking organizations, belatedly following the approach rival Facebook pioneered in 2016.
The Birdwatch project, which launched as a pilot in January 2021, was hailed by some as a bold and creative approach to the problem of addressing misinformation on a vast public platform that serves as a critical news conduit for many in the media and politics.
Others raised the concern that delegating fact-checking to the public would create new problems, such as groups of activists working together to flag tweets they simply disagree with. Without professional oversight, crowdsourced fact-checking is “far too easy for bad entities to hijack,” said Brooke Binkowski, managing editor of the fact-checking site Truth or Fiction.
In the Birdwatch pilot, contributors must register with a verified email address and receive approval from the company to join. As of November, they can hide their identity from one another and from the public by using an alias. Any contributor in the program can append a fact-checking “note” to any tweet. Other contributors are then asked to rate that note’s helpfulness, using criteria such as whether it cites reliable sources, uses neutral language, provides important context and directly addresses the tweet’s claims.
Those notes and ratings are available to the public in spreadsheet form, and they’re visible to Birdwatch contributors on Twitter itself. For the vast majority of Twitter users who are not part of the Birdwatch pilot, however, it might as well not exist: The notes are not visible in the main Twitter feed, and they have no effect on the algorithm that decides what tweets each user sees.
Crowdsourcing fact checks can be dicey if not done carefully, said Joshua Tucker, co-director for the NYU Center for Social Media and Politics. He co-authored a recent study, published in the Journal of Online Trust and Safety, which found that people struggled to identify false news stories, performing no better than random guessing in many contexts. The study did not attempt to replicate Birdwatch’s approach, which relies on self-selecting volunteers, but it did indicate that certain more sophisticated approaches to crowdsourcing might have some potential as part of a larger fact-checking project — especially if that project includes professional fact-checkers, which Birdwatch so far does not.
A review of some of the tweets flagged on Thursday, the first day of the invasion, turned up a mix of dry factual corrections, helpful debunkings of tweets that misleadingly presented old images or videos as new, and a few notes that focused more on ideological disagreements than factual accuracy.
For the most part, the fact-checking notes rated “helpful” actually did seem potentially helpful — that is, if they were incorporated into Twitter in any meaningful way, which they aren’t.
A video of a dramatic explosion, tweeted with the text “Mariupol” — the name of a Ukrainian border city — had been flagged by two Birdwatch users who correctly pointed out that the same video had been posted to TikTok months earlier. Another viral tweet, which showed the flight path of a lone Air India aircraft headed straight for the conflict zone, had been flagged by a user who cited a reputable source showing that it actually flew around Ukrainian airspace, like all other commercial air traffic.
Some other notes visible under the “new” tab of the Birdwatch feature seemed, let’s say, less helpful. One note appended to a tweet on Monday read simply, “baba booie.”
