Twitter appears to have relaxed the effects of its “quality” search filter function associated with so-called “shadow banning,” after evidence indicated that the content of some accounts on the social media platform was obscured to followers, while many others appeared to be unaffected.
The Epoch Times discovered the change in the filter functionality while researching the shadow banning phenomenon. Some search results changed overnight on July 28 after The Epoch Times alerted Twitter to some of its findings.
On July 26, President Donald Trump addressed shadow banning, which appeared to affect certain Republicans, and promised to “look into this discriminatory and illegal practice at once.” On July 27, Rep. Matt Gaetz (R-Fla.), whose Twitter account was affected, said he filed a complaint against the company with the Federal Election Commission (FEC).
For several years, Twitter users have accused the company of shadow banning—hiding a user’s content from other users without informing them.
While the full extent and history of the practice remain unclear, one identified method was to remove the tweets of targeted accounts from search results, unless a user turned off the “quality” filter function, found in the “search filters” box. The filter is on by default and snaps back on after just one search.
Even with the filter off, some tweets from the targeted accounts remained invisible in the default “top” search results, at least on the afternoon of July 27.
As of the afternoon of July 28, tweets from targeted accounts appeared to show up in the “top” results even with the “quality” filter on. They still remain hidden by the filter in the “latest” results.
In August 2016, Twitter announced a filter function that removes “lower-quality content, like duplicate Tweets, or content that appears to be automated.” But that filter only applies to user notifications. There appears to be no announcement regarding the “quality” filter affecting search results.
Users who hold conservative and libertarian political views have been accusing Twitter for years of handicapping their accounts.
Since January, such complaints have been substantiated by multiple experiments, as well as current and former Twitter employees’ acknowledgments to undercover investigators.
On July 26, Twitter policy lead Vijaya Gadde and product lead Kayvon Beykpour attempted to explain the situation in a blog post.
They denied shadow banning users, arguing it’s only shadow banning if the user content is made “undiscoverable.” They did say that tweets and search results are ranked, making some content harder to find.
That explanation has been rejected by many of the users affected by the practice.
“A rose, by any other name, is still a rose,” said an independent investigator, whose account had been effectively shadow banned. As a conservative, he wishes to remain anonymous because of safety concerns.
Aside from the exclusion from search results, he’s noticed other effects, which, he says, could be more accurately described as account throttling.
“Your tweets may not show up on other followers’ accounts, they may not see your replies, and your tweet responses will end up behind a wall that non-followers will have to check to see your tweet,” he said, adding that he and others have confirmed these effects by running tests through alternative accounts.
Gadde and Beykpour wrote that tweets and search results are “ranked lower” if posted by “bad-faith actors who intend to manipulate or divide the conversation.”
“Bad-faith actors” are identified by whether they lack a confirmed email address, by how recently their accounts were created, and whether they uploaded a profile image.
Gadde and Beykpour said that Twitter also determines “bad faith” by “actions you take on Twitter,” such as who one follows or retweets, and even by factors beyond one’s control, such as “who mutes you, who follows you, who retweets you, who blocks you.”
That’s why, they said, several prominent Republican congressmen had their Twitter accounts recently scrapped from the platform’s search suggestion function.
“There are communities that try to boost each other’s presence on the platform through coordinated engagement. We believe these types of actors engaged with the representatives’ accounts,” Gadde and Beykpour wrote, also saying that “as of yesterday afternoon, this issue was resolved.”
Gaetz is convinced Twitter targeted him intentionally.
“I am certain there were only four members of Congress who had their voices suppressed on Twitter,” he told Fox News’ Tucker Carlson on July 27. “Matt Gaetz, Jim Jordan, Mark Meadows, and Devin Nunes. That’d be one hell of a coincidence.”
All four lawmakers are Republicans. All four also have been vocal in seeking more documents regarding alleged bias within the FBI and the Foreign Intelligence Surveillance Court warrants that spurred the Russia investigation, alleging that the Obama administration spied on the Trump campaign before and after the 2016 election. Official documents have shown the Justice Department and FBI under former President Barack Obama used unverified opposition research paid for by Democrats to obtain a warrant to spy on the Trump campaign.
Gaetz said he filed a complaint to the FEC, accusing Twitter of an illegal campaign contribution to his opponents.
“Here, the corporate donation is allowing Democrats, and the people running against me specifically, to have access to elements of the search feature that I didn’t have access to,” he said.
The Epoch Times reviewed more than 50 accounts identified by hashtags commonly associated with Trump supporters (such as #MAGA) and Trump opponents (such as #TheResistance). Many of the accounts on both sides were prolifically cross-promoting their tweets, garnering thousands or even tens of thousands of followers.
Almost all accounts of the Trump supporters were targeted by the “quality” filter.
None of the accounts used by Trump opponents were apparently targeted by the filter, according to the confirmation function on the website Shadowban.eu, which The Epoch Times repeatedly tested to verify it can indeed identify accounts that have been affected.
Twitter didn’t respond to requests for comment.
To disable the “quality” filter, a user would first need to know it exists, but neither the July 26 blog post nor Twitter’s earlier May 15 post discussing its new content policies contained any mention of the function.
Moreover, the July post only acknowledged that Twitter does “rank” tweets and search results, omitting the fact that the filter does hide content altogether.
Many of the Trump supporters posted comments, links, and memes expressing faith in God, love for America, support for Trump, and criticism (at times speculative) of prominent Democrats such as Obama and former Secretary of State Hillary Clinton.
Even Scott Presler, the vice chairman of the Virginia Beach Young Republicans who’s known for his upbeat attitude, has been shadow banned, despite having a verified account with almost a quarter-million followers.
“My Twitter account is being severely shadow banned. Despite this, I’m not going to stop tweeting,” Presler wrote in a July 25 tweet. “I just need help getting out messages about voter registration, trainings, & electing Republicans this November. Thank you all for the constant love and support.”
He used the popular recurring hashtag #WednesdayWisdom, but nobody clicking the hashtag and browsing the latest results on the default settings would see his tweet.
Many of the Trump opponents posted comments, links, and memes in support of Democrats and progressive policies and criticizing (at times, speculatively) Trump. Some of them regularly used foul or abusive language, without any apparent effects.
“SHUT THE [expletive] UP!” stated user THE[expletive][expletive]™ in a July 27 tweet, using the popular recurring hashtag #FridayFeeling as well as hashtags #DearDonald and #[expletive]Traitor, with an obscene emoji and a picture of the message “[expletive] TRAITOR” in large red letters. No shadowban.
The alleged slant in Twitter filtering matches how one of the company’s engineers, Pranay Singh, described algorithms that are supposed to identify automated “bot” accounts.
In reality, as Singh told undercover reporters from Project Veritas, the majority of the algorithms appeared to target Republicans.
“You look for Trump, or America, or any of 5,000 keywords to describe a ‘redneck’,” he said.
Twitter responded to the Project Veritas report by saying it was “committed to enforcing our rules without bias and empowering every voice on our platform, in accordance with the Twitter Rules.”
“We want a vibrant and healthy public conversation inclusive of all perspectives, and one that’s immediately relevant and valuable,” Twitter CEO and co-founder Jack Dorsey said in a July 27 tweet. “We’re always listening to the conversation around this, and we commit to participating more fully in it. It’s important to us we get this right.”
Twitter shares fell 21 percent during regular trading on July 27, after the company reported a decline in monthly active users.
If Gaetz’s complaint to the FEC succeeds, it could open the floodgates to others.
Gaetz said Twitter and other social media companies have successfully averted lawsuits by declaring themselves neutral public forums that can’t be held responsible for what people post.
“Twitter … can’t say on one hand, ‘We’re neutral and thus we shouldn’t have to respond to lawsuits,’ and then, on the other hand, tell me and other outspoken conservatives that our behavior results in suppression on their platform,’” he said. “They can’t have it both ways.”