A engineer's plan to "ban a way of talking" didn't refer to conservatives.

James O'Keefe is a conservative activist who has made a name for himself with hidden camera investigations of supposedly liberal organizations. This week, he turned his attention to Twitter, publishing a series of secretly recorded videos of Twitter employees (and former employees) discussing Twitter's content moderation policies and political culture.

O'Keefe claims to have uncovered smoking-gun evidence of a far-reaching conspiracy to suppress conservative speech on the Twitter platform. Conservative media outlets have taken that frame and run with it.

But there's a lot less to the two videos Project Veritas released this week than meets the eye. For example, O'Keefe has repeatedly highlighted Twitter engineer Steven Pierre's comment that Twitter was working on software to "ban a way of talking." The strong implication is that the "way of talking" Pierre wants to ban is conservative political speech. But if you actually watch the full video, that's clearly not what Pierre meant.

"Whether it's positive or negative doesn't look for content," he said. "It's more like if somebody's being aggressive or not. Somebody's just cursing at somebody."

In other words, Pierre was describing a project to filter out trolling and harassment. O'Keefe could have made this clear—or just left that clip on the cutting room floor. But O'Keefe is a political activist who often casts the targets of his investigations in the worst possible light—even if he has to use smoke and mirrors to do it.

Why it’s worth taking O'Keefe's investigations with a grain of salt

James O'Keefe has used the same basic playbook since 2009, when he made a name for himself catching representatives of the now-defunct liberal community group ACORN advising a client (actually an associate of O'Keefe's posing as a prostitute) on how to conceal her illegal prostitution business.

In 2010, O'Keefe plead guilty to misdemeanor charges after breaking into the offices of Sen. Mary Landrieu (D-LA) as part of an attempted sting operation.

As O'Keefe's fame has grown, he's been able to raise money and recruit a growing army of staffers to carry out wide-ranging stings against a variety of targets. In recent months, his organization, Project Veritas, has focused on elite media institutions, with recent exposés focusing on the Washington Post, the New York Times, and CNN.

The group's basic approach is to talk to a wide range of people connected to an organization in hopes that a few of them will say things that sound bad. Project Veritas isn't very picky about who it targets for its sting operation. For example, in the group's exposé of the New York Times, O'Keefe quotes a Times IT consultant saying that people in the Times newsroom hate Trump.

If you talk to enough people at an organization with thousands of employees, it's inevitable that you'll catch some of them saying stuff that at least sounds bad. We don't know how many Twitter employees O'Keefe's organization talked to who didn't say anything embarrassing—or even directly contradicted O'Keefe's thesis that Twitter is systematically censoring conservatives.

What the Twitter videos showed

In total, Project Veritas has published comments by eight Twitter employees—all apparently without their knowledge or consent. While O'Keefe portrays all of these clips as evidence that Twitter is pushing a secret anti-Trump and anti-conservative agenda, most of the clips fall apart upon close examination.

Here's a full list of the eight people featured in Project Veritas video so far—and what they actually said.

Clay Haynes: "There's a reason why we have a subpoena process"

The first Project Veritas video focused on Clay Haynes, a senior network engineer at Twitter. The headline on that video focuses on Haynes saying that "we're more than happy to help the Department of Justice in their little investigation" against Donald Trump by turning over direct messages and other private information.

But the tape shows that when Haynes is urged to look at the private messages of Donald Trump and his son Donald Jr., Haynes laughs nervously and says, "We have a subpoena process for that very reason." In other words, Haynes's comments are entirely consistent with Twitter's official policy: that it only discloses private information about its users if it receives a court order to do so.

Olinda Hassan: "You need to also have control of your timeline"

As a Twitter policy manager, Hassan helps develop Twitter's regulations for issues like hate speech. In the video, a Project Veritas person says to her, "I've tried to, like, block people like Cernovich and stuff like that and mute and stuff like that, but they still show up." Hassan says that Twitter is "working on" the problem, with a goal to "get the shitty people to not show up." Hassan doesn't specify which "shitty people" she's talking about, and O'Keefe naturally insinuates that she's talking about an official Twitter blacklist of conservative pundits.

But in the same conversation, Hassan stressed that "you need to also have control of your timeline." In other words, there was no indication that this "shitty people" feature was limited to any one ideology or partisan affiliation. If you're a conservative user who wants to block a bunch of annoying liberal pundits, the same technology should allow that. It was the Project Veritas staffer, not Hassan, who asked about banning conservatives—and nothing Hassan said implied that the feature was limited to conservatives.

Conrado Miranda: "That's a thing"

A woman associated with Project Veritas said to Miranda, a former Twitter engineer, "I've heard talk that it's a good thing because they'll use it to ban, like, Trump supporters or conservatives, so I don't know if, like, that's just a rumor or if that's true."

"That's a thing," Miranda replied.

The problem is that Project Veritas edited the segment so that we don't know what what the woman said just before this exchange. That means we don't know what "it" is or who "they" are. For example, this could just be referring to technology that gives users greater ability to filter out tweets they find annoying—whether they're conservative or liberal. In his response, Miranda doesn't say anything more about political filtering, instead giving a high-level technical description of how content filtering works in general.

Abhinav Vadrevu: "I don't know if Twitter does this any more"

The phrase O'Keefe quotes most frequently from the second video is "shadow banning," a technique described by former Twitter engineer Abhinav Vadrevu where "you ban someone but they don't know they've been banned, because they keep posting, but no one sees their content." Vadrevu goes on to admit that the practice seems "unethical in some way."

O'Keefe likes to conflate this interview with others, implying that multiple sources talked about shadow banning conservatives. But in reality, Vadrevu is the only one who discusses the practice. And he doesn't mention conservatives or Trump supporters at all. So there's no evidence that Twitter is shadow banning conservatives—as opposed to garden-variety spammers and trolls.

Oh, one other important fact that O'Keefe downplays: Vadrevu says, "I don't know if Twitter does this anymore." ("Twitter does not shadow ban accounts," a Twitter spokeswoman told Ars by email.)

Mo Norai: "We're in California, very liberal"

Norai was a former content review agent—someone on the front lines enforcing Twitter's content policy. Norai said that rank-and-file moderators like him had a lot of discretion, and that San Francisco's predominantly liberal culture meant that conservative tweets got more aggressive scrutiny than liberal ones.

Norai didn't have a lot of concrete details about Twitter's policies, though. After mentioning that San Francisco is liberal, he added, "As a company you can't really say it because it would make you look bad, but behind closed doors are lots of rules." But he didn't elaborate on what those rules are.

Steven Pierre: "Whether it's positive or negative doesn't look for the content"

O'Keefe has repeatedly quoted software engineer Steven Pierre's statement that Twitter was planning to "ban a way of talking." O'Keefe wants viewers to believe that the "way of talking" here is conservative ideology. But it's clear from the video that this isn't what Pierre meant at all. Rather, Pierre said that the algorithm he was working on would look at "if somebody is being aggressive or not. Somebody's just cursing at somebody."

"Whether it's positive or negative doesn't look for the content" of a tweet, Pierre adds. In other words, the "way of talking" he has in mind has nothing to do with ideology.

Mihai Florea: "It's really hard to decide what to do about Donald Trump."

Florea is a software engineer at Twitter, and is featured only briefly in the video. "Half the people want to ban him," he says. "Half the people want to keep him."

Pranay Singh: "Who talks like that? It's for sure a bot."

Project Veritas's videos with Singh, a direct messaging engineer at Twitter, is probably the most the most awkward for Twitter.

"Just go to a random [Trump] tweet, and just look at the followers," Singh says. "They'll be like guns, God, America, like, and with the American flag and like the cross. Who says that? Who talks like that? It's for sure a bot."

What's not clear is whether Singh is describing an actual Twitter feature or just speculating on how Twitter might combat the problem of Russian bots supporting Donald Trump. Notably, Singh is on the direct messaging team—so he might not even know how Twitter's anti-bot software works.

"Are you sure you’re not a programmer?"

O'Keefe's investigation breaks one of the cardinal rules of the journalistic profession: that you shouldn't quote a source unless you've first identified yourself as a reporter. That's not just a matter of fairness to the people being quoted. Talking to a source under false pretenses can distort what the source says.

The segments with Singh illustrates why this principle is so important. The undercover video shows him talking to two women.

One of the women—a project Veritas employee—asked Singh to explain how machine learning works: "So if there's like, American guns, can you write an algorithm to just take all those people out?"

Singh tells her: "It's actually how we do it, yeah. You should be an engineer. You're really good at this shit."

The woman responds: "So can you just write them against, like, so like say someone has Trump in the name, or like, conservative, or like... can you just, like, write algorithms for that?"

"Are you sure you're not a programmer?" Singh asks "That's exactly how you do it."

It seems pretty clear that Singh's main goal in this conversation is to impress the woman he's talking to. Could that cause him to exaggerate his knowledge of Twitter's filtering algorithms—as well as his own liberal politics? It's the kind of question a serious reporter would ask—after all, as a direct messaging engineer, Singh may not even know very much about Twitter's algorithms for deleting bot accounts.

Project Veritas hasn’t produced a smoking gun

It's important to zoom out here and sum up what Project Veritas has and hasn't proven with their exposé. The group definitely established that Twitter's workforce is predominantly liberal—something that will be unsurprising to anyone familiar with the politics of the Bay Area. Given the power of Twitter's content moderators and engineers, there's an obvious danger that the company's liberal biases will cause them to treat conservative content more harshly than liberal content.

What Project Veritas hasn't uncovered, however, is any evidence that Twitter is systematically using its platform to silence conservative voices. When an engineer talked about "banning a way of talking," he was clearly not referring to conservatives. The same is true of another former engineer's discussion of "shadow banning"—and it's not even clear if Twitter still engages in shadow banning at all.

Another Twitter employee's comment about "getting the shitty people to not show up" seems to refer to efforts to enhancing Twitter users' ability to decide for themselves what kind of tweets they want to see. A liberal Twitter user could use it to avoid seeing conservative tweets, but a conservative could just as easily use it to avoid seeing tweets from liberals they find annoying.

Project Veritas did find one person involved in Twitter's content moderation policy who said he saw the policy being used more strictly against conservatives than liberals. But that's about it. O'Keefe didn't find any evidence of systematic anti-conservative bias in Twitter's policies or filtering algorithms.

Any large organization in America would look bad if a lavishly funded group like Project Veritas took dozens of employees out for drinks on false pretenses and tried to get them to dish dirt about their employers. The fact that Project Veritas had to rely so heavily on selective editing and innuendo to make its case suggests that Twitter might actually be doing a pretty good job of managing its platform in a fair, ideologically neutral fashion.