As the public understanding of social media’s role in spreading harmful misinformation grows, Facebook has scrambled to improve its image. Rather than truly addressing its problem with misinformation and removing it, Facebook keeps touting its labels as a proactive response -- ignoring both internal and external data that shows they are ineffective and too inconsistently applied to help.

During the 2020 presidential election, Facebook used labels to warn that a post contained false information or to provide authoritative information about a topic discussed and reportedly down-ranked such posts. Facebook executives touted these labels after the election, claiming that they labeled 180 million posts as “false” between March and November. But they refused to address how effective the labels were at deterring users to like or share the posts. In fact, Facebook’s data scientists reportedly found that labels on Trump’s posts decreased reshares by only about 8%, noting that “given that Trump has SO many shares on any given post, the decrease is not going to change shares by orders of magnitude.” Media Matters and others repeatedly reported on how election misinformation, particularly from Trump, was still performing well and spreading on Facebook, despite the labels. 

Even some right-wing media personalities have admitted that labels actually bring attention to Facebook posts, with the Daily Wire’s Candace Owens saying, “It makes me want to click through and see what it says.”

In a previous study, Media Matters found that Trump posted COVID-19 misinformation, election lies, or extreme rhetoric about his critics in roughly 1,443 of his Facebook posts between January 1, 2020, and January 6, 2021, when his account was suspended. 

As The Washington Post noted, our research showed “in stark numbers just how many times Trump came up to the line of Facebook’s rules — if not crossed them — but was given a pass by the company.” The platform labeled at least 506 of these posts that contain misinformation or extreme rhetoric, mostly with a generic label such as “See Election Results,” which provided links to authoritative information but did not provide users with any indication of whether the post was false or misleading. Of these 506 posts, just one was labeled false and another was labeled partly false, while the rest received generic labels, such as: “The US has laws, procedures, and established institutions to ensure the integrity of our elections.”

Facebook’s seemingly haphazard labeling system missed some of Trump’s posts promoting election misinformation and amplified irrelevant posts with generic labels.

Media Matters’ latest analysis finds that Trump’s posts with labels actually earned roughly 2.6 times more average interactions per post than his posts overall. The 506 posts that were labeled earned over 205.8 million interactions, or an average of roughly 407,000 interactions per post. Comparatively, Trump’s 6,081 Facebook posts between January 1, 2020, and January 6, 2021, earned over 927 million interactions, or an average of roughly 152,000 interactions per post.

In another study, Media Matters found that Trump cited right-wing media outlets in roughly 868 posts between January 1, 2020, and January 6, 2021, earning roughly 110 million interactions or an average of roughly 126,000 interactions per post.

Media Matters has now found that Facebook labeled 147 of these 868 posts that cited right-wing media outlets. These 147 posts earned over 42 million interactions, or an average of roughly 291,000 interactions per post, which is more than double the average for all Trump’s posts citing right-wing media outlets.

Notably, 127 -- or over 86% -- of Trump’s labeled posts citing right-wing media were related to election integrity, with an additional five labeled posts specifically claiming that the election was “stolen” or calling on people to “stop the steal,” even though “Stop the Steal” content was supposedly banned by Facebook after the Capitol riot on January 6. Trump’s labeled posts that cite right-wing media and are related to election integrity earned more average interactions per post than his posts citing right-wing media and his overall posts in 2020 and 2021. Posts mentioning a “stolen” election or “stop the steal” earned even more average interactions per post, with roughly 640,000.

In a recent press call, Facebook executives touted the platform’s fact-checking and labeling process, which includes reducing the distribution of labeled posts, providing users with authoritative information, and using automation to identify similar content to “amplify the impact of any review decision.”

But Media Matters and others have long documented Facebook’s inability to consistently apply its policies, including its failure to label and/or remove misinformation. The latest example of Facebook’s failure is Trump’s statements from his new blog that contain misinformation, which are spreading on the platform without labels, even though he is suspended.

On May 13, Trump posted a statement on his blog containing a now-debunked claim about Maricopa County’s election database being deleted. The statement has been shared on Facebook, with posts including text from, images of, or links to the statement. 

At least two posts with images of Trump’s May 13 statement have since been labeled “false” by Facebook, with a link to a fact check of the claim.

However, as with other types of content, Facebook has been inconsistently labeling these posts sharing Trump’s statement. Media Matters found dozens of other posts with images or text from the statement that remain on Facebook and are not labeled as “false.”

Notable posts include:

Using Crowdtangle, Media Matters compiled a list of 6,081 posts created by Trump from January 1, 2020, to January 6, 2021, when Facebook suspended his account. 

We compiled posts that were related to COVID-19, the presidential election, or Trump’s critics, a process which is described here. The final lists included 516 posts with the COVID-19 related keywords, 363 posts with keywords related to the election, and 683 posts with harmful or dehumanizing rhetoric. Overall, 1,443 posts contained COVID-19 keywords, election-related keywords, and/or harmful rhetoric, and 506 of these posts were labeled by Facebook. We reviewed the resulting posts, including the total number of interactions -- reactions, comments, or shares -- they earned using CrowdTangle.

We also compiled posts that cited right-wing media outlets and personalities, a process which is described here. We reviewed the resulting 868 posts and the 147 of them that were labeled by Facebook, including the total number of interactions -- reactions, comments, or shares -- they earned using CrowdTangle.