I Analyzed 1.9 Million Parler Posts Made During the Capitol Riot
There were more chicken soup recipes than posts about fallen Capitol officers
On January 11, Amazon pulled the plug on hosting services for social app Parler, stating violations of their terms of service. But just before that happened, @donk_enby, the Twitter alias of a security researcher, culled and published all available public images, public videos, and public posts made on the platform from January 6 through January 10.
Before being taken off-line, Parler had achieved over 11 million downloads, with most installs (9 million) coming from the United States, reports CNET. According to analytics company Sensor Tower, the app was downloaded 997,000 times from Apple’s App Store and Google Play during that final week before going dark. And in that week, 70 terabytes of user-generated content was generated.
As a result of the data dump project, the videos have been published on ProPublica so the public can see firsthand what Parler users saw. In contrast, the cache of unstructured HTML pages, though stored on the Internet Archive, is not easily searchable.
So this week, I ran those public HTML pages, all 1,854,330 of them, through a full-text database and indexed them to make search easier and glean some preliminary insights from the platform. Those indexed pages contain public profiles, messages, threaded comments, engagement metrics, metadata, and links to corresponding media if images or video were attached.
There are several caveats in this text analysis. The Internet Archive describes the dump as partial, and it is unclear how many HTML pages might be missing from the set. The provenance of the files is also unclear. For this research, I will assume, but cannot validate, that each page was downloaded from the public Parler website and was unaltered in the process.
“A text analysis of nearly 2 million records culled by donk_enby’s archiving project revealed zero pages regarding the unprovoked beating, punching, crushing, tasing, dragging, or death of police officers by rioters. Not one.”
While many pages include links to images and video, only user-generated text was analyzed. The engagement metrics stored in the HTML were captured at the time of the download, which may vary from page to page. Specifically, most files are timestamped on January 10, but there may be some slight deviation by a day, depending upon when the sample page was archived.
Also, an embedded media asset like a meme might provide additional context for a post. For example, it’s always possible that text such as “I’m going to kill you!” can be said in jest, if it’s included with a picture of someone stealing a chocolate brownie.
Some pages contain no user-generated text at all. Like Twitter, a Parler user can publish an image or video, or simply upvote (that’s a like on Parler) or echo (the same as a Twitter retweet) another user’s post. So analysis in aggregate to determine overall tone or sentiment could be flawed. Last, it must not be assumed that upvotes or echoes equate to a user’s endorsement of a post.
After several hours, the index was complete, and I was able to quickly find, filter, and rank pages by keywords. Helpers, like an intelligent autofill, similar to a Google search query, allowed for the easy identification of top keywords in the mounds of content.
For example, a search for conspiracy-related terms such as “deep state,” “QAnon,” “the great awakening,” or “WWG1WGA” was found on 45,370 pages.
That said, it wasn’t all storming the Capitol and conspiracy theories. The Parler dump also includes a whopping 10 pages that discuss chicken soup. User @DKirkendall shared a recipe for the dish with homemade egg noodles on YouTube “to warm your bones at the peaceful protests.”
(Hang or shoot or kill) and Pelosi
Next, I tried some specific keyword searches. According to an FBI witness, Newsweek reported that Proud Boys intended to kill Vice President Mike Pence and Nancy Pelosi. My attention shifted to a review of posts expressing a sentiment to harm Nancy Pelosi. While I don’t believe Parler was used as the platform to conspire against the speaker of the House (that’s what Telegram is for, apparently), I was interested in reviewing the Parler community’s perceptions.
“In 1.9 million Parler pages currently available in the Internet Archive, there isn’t a single mention of Officer Brian Sicknick, who was killed by rioters after being bludgeoned with a fire extinguisher.”
Some posts were as simple as “Hang Pelosi!” Others were much more descriptive and achieved a bit of online love. “This is the tunnel system they will try and sneak away in, catch Pelosi and hang her ass,” received 119 echoes and 139 upvotes on 11,917 impressions. Rioters erected a noose at the Capitol building.
Some variation of the word “kill” or “harm” and Pelosi was found in 577 pages. I know what you’re thinking. Isn’t is possible that in the mix was a reference to “I’d kill for Nancy Pelosi’s recipe for homemade apple pie?” Surprisingly, no. A search of keywords satisfying that query returns zero results.
How about the number of posts that simply include the phrase “love Pelosi?” There’s just one example in the entire sample set. That message reads, “We love Pelosi’s office, it’s great,” in a presumed reference to being inside the Capitol building.
Patriots or rioters?
That first search revealed I needed to filter false positives if there was a content match based upon a referenced news report like “Patriots now occupy the Senate Chamber and invaded Pelosi’s office. A Trump supporter has been shot.” These pages were excluded from the query results, but the term “patriot” struck a new chord of curiosity.
That’s because two weeks ago, I revealed the findings of a poll conducted the day after the riot showed a stark difference between the way self-identified Democrats and Republicans labeled the participants. Overwhelmingly, up to 78% of Democrats used terms like rioter or extremist, while only 26% of Republicans used the word. Thirty percent of Republicans used the term “patriot,” while only 2% of Democrats did so.
So how did Parler users describe the rioters and agitators?
To answer that question, I first removed words like “BLM” or “Antifa” because search results returned an abundance of content related to those key pairs in an attempt to justify the actions in D.C. An example is a response to a thread started by Dinesh D’Souza, where user @mrforUSA claims that Trump supporters “do not riot” (in all caps for maximum effect) and that “Antifa and BLM are rioters, destroy property, set buildings on fire.” Results like these were eliminated from the search output since it did not directly assign the label to agitators.
What I found instead was that the majority of pages on Parler, and those with the highest engagement metrics, used terms like “protesters,” “Trump supporters,” or “peaceful protesters.” Over 116,000 pages contain those terms, compared to less than 2,700 pages that cite the word rioter when both BLM and Antifa are excluded.
One exchange sums up the typical sentiment on the platform as one user posts, “Go figure, 100’s of thousands of Trump supporters in D.C.” with “zero looting, arson, violence, assault, murders, theft, riots, destruction of property.” To which @Notoriousvpe agreed, “That’s exactly why we don’t get results, because we’re peacefully protesting.”
One of those posts, which linked to a Washington Post article titled “Trump Tells Rioters to Stop and Respect Law and Order in Video,” was “corrected” by another user, @Res18, who replied, “what rioters… you mean peaceful protesters.”
Trying to find a direct reference to the phrase “rioter” or similar variant which did not include a link to a news item or video using the word in the headline was rare but not nonexistent. In fewer than 400 of the 1.9 million posts, terms like “domestic terrorists” could be found. But these posts also had some of the lowest engagement metrics on Parler.
Consider the lone parley using the hashtag #riotersarecriminals. In it, user @barefootb shares the opinion that the Proud Boys are domestic terrorists and “in case you haven’t heard, domestic terrorists have stormed the Capitol.” This unpopular post received no echoes and not a single upvote on Parler.
A comment by another brave soul said the Capitol was “invaded by anti-American domestic terrorists” and that the “attempt to overthrow the government and destroy our democracy is the most unpatriotic act I’ve seen from our people.” This post also received no echoes but did manage to claim one upvote. @SavingAmerica’s post that declared the “Proud Boys need to be labeled as a domestic terror group!” received a whopping 18 upvotes and three echoes, the highest I came across of posts critical of the riot.
What wasn’t in the Parler dump?
Speaking of “patriots,” this was a common definition used to describe Ashli Babbitt (and the many misspelled keyword variants) on Parler. Babbitt was shot and killed by Capitol police as she attempted to climb through a window near the Speaker’s Lobby. One thread said she was “a war hero” whose “life was senselessly taken when she was gunned down, defending our nation’s freedom.”
That post received 2,760 upvotes and 543 comments on 278,273 impressions. But there were plenty more. Over 11,100 pages reference Babbitt, which doesn’t include posts with a general classification of “woman shot at Capitol,” — at least another 3,000.
By comparison, a text analysis of the nearly 2 million records culled by donk_enby’s archiving project revealed zero pages regarding the unprovoked beating, punching, crushing, tasing, dragging, or death of police officers by rioters. Not one.
For example, a search of messages about police officer Michael Fanone, who was tased in the neck at the tunnel’s mouth on the Capitol building’s west terrace, returns no results. And in the set of 1.9 million Parler pages currently available in the Internet Archive dump, there isn’t a single mention of Capitol Police Officer Brian Sicknick by name, who was killed by rioters after being bludgeoned with a fire extinguisher.
Sicknick’s name was known to the public per a statement from the U.S. Capitol Police Department. It was widely circulated across all media outlets on January 7, well before Parler’s shutdown on January 11.
More Parler pages are devoted to chicken noodle soup than posts referencing police officers harmed or killed at the Capitol.
Also included in the cache of pages are posts from users who claim to have been in D.C. with firsthand eyewitness accounts. One user, who in a parley describes himself as a decorated veteran and former intelligence officer in the USAF, said he was “front and center at the Capitol” and loved it despite “being attacked and pepper-sprayed by Capitol Police.” According to his account, they antagonized “peaceful protestors.”
Still another user, commenting on Dinesh D’Souza’s feed, said those “peaceful protesters” at the Capitol building were simply law-abiding, First Amendment assemblers. “My family was there and saw what happened. The D.C. police started to pepper-spray patriots unprovoked,” says @IMAPATRIOT4EVR. “Biden’s secret service was there to stir up trouble,“ they added.
“Well, there you have it,” is the conclusion of the post. Indeed, there you have it.
Another user claimed he found piles of bricks “everywhere,” which he says were used to arm BLM and Antifa. “I’m in D.C. and found multiple stacks of bricks on M Street on my walk back to my hotel,” he shouts in an expletive-laden post.
Misinformation and misdirection to cast blame on others were rampant among high-profile users as well. @SupportLinWood’s account posted, “BLM & Antifa dressed up as Trump supporters to cause violence at the Capitol,” receiving 1.45 million impressions and 12,873 upvotes.
Minority voices of reason
There were also glimmers of sanity within the echo chamber. “I’m embarrassed by the actions of so many misinformed, ignorant, and easily manipulated people in my party,” one user said. “No credible evidence was entered into the courts. President Trump simply lost.” Still another user, @Phugmotley shared, “These are thugs and domestic terrorists. This is not a peaceful protest. Breaking windows and breaking down doors is not peaceful. Trump is responsible for this.”
But engagement metrics for these opinions were dismal, even when shared by the most high-profile users, like a nugget in the posting dump from John Matze (alias @John), founder of Parler.
Far from a full-throated condemnation of the rioters, Matze said, “I’m not happy with today’s actions.” His post, receiving over a half-million impressions, continued, “I didn’t think conservatives would do that. Let’s condemn all looting and violence no matter what cause is proposing it. Admit we have dehumanized each other for the last year. Admit there is too much hate.”
Most did not agree with the Parler founder’s commentary, based upon engagement metrics. Matze’s post earned just 1,313 echoes and 3,200 upvotes.
In contrast, @LLinWood’s post (Lin Wood is a prominent pro-Trump attorney) that “Pence, Pelosi, and McConnell” were in the midst of a “planned coup to overthrow the duly elected president, Donald Trump,” saw much higher engagement. That page was viewed 7.5 million times, liked by 79,995 people, and echoed 31,132 times.
On the day of the riot, @LLinWood also said that Vice President “Mike Pence is a traitor & must be charged with treason this morning.” The penalties for treason are death or a minimum of five years imprisonment and a hefty fine. That post also eclipsed Matze’s; it was viewed 4.5 million times, received 27,984 likes, and 110,035 echoes.