II. What We Know about Parler and Publicly Available Parler Data

Parler is part of a rising wave of so-called “alt-tech” sites targeting conservatives that sprang up after major social media companies like Facebook and Twitter began taking a more aggressive approach to content moderation in the wake of deadly clashes between protesters at the 2017 “Unite the Right” rally in Charlottesville, Va.¹ Launched in the summer of 2018 with the support of financial backing from Rebekah Mercer, the wealthy daughter of hedge fund manager and Cambridge Analytica investor Robert Mercer, it immediately generated buzz for attracting pro-Trump influencers.² Founded by University of Denver graduates John Matze and Jared Thompson, the platform billed itself as the “the world’s town square” and the “premier free-speech” alternative to Twitter where users can “speak freely” and “without fear of being deplatformed.”³

Initially, Parler’s user base showed low but steady growth at the outset with an estimated 130,000 users signing up in June 2018 in addition to the company’s 30 employees.⁴ The platform was unique in that it purportedly relied on big, marquee names to forge connections between its user base and advertisers. Matze told CNBC that the company would deploy a “decentralized ad model where people can advertise with influencers specifically and those influencers’ brands rather than our company as a whole.”⁵ “We’re a town square,” Matze said in that interview, “but we’re not a publication.”

Early on, several right-wing personalities and conservative politicians signed up, including Trump campaign insiders Sidney Powell, Lin Wood, Sen. Ted Cruz (R-Texas) of, Rep. Devin Nunes (R-Calif.), and Trump’s former national security adviser, retired Army General Michael Flynn.⁶ With few guidelines on what could be posted beyond a prohibition against illegal activity, low requirements for identity verification, and light content moderation, the platform appealed to social media users caught crosswise by mainstream platform governance standards.

In fact, as noted by the Stanford Internet Observatory, before Parler was deplatformed, moderation was driven by users themselves with only a little more than 800 users designated as content moderators for upwards of 13 million users in January 2021.⁷ Researchers have noted that the platform’s loose governance standards during the period leading up to January 2021 made it a hothouse for networks of fake accounts, parody accounts targeting Democrats and perceived leftist elites, and accounts that augmented their engagement by using integrations that would ingest news article RSS feeds from media sites known for promoting disinformation, such as the Epoch Times.⁸

Prior to the temporary halt of service, public profiles on Parler were largely accessible with no login to the site required. Scholar Jacqueline Otala, et al. notes that on the early version of the platform users were scored by their level of engagement that they and their content drew.⁹ The 1.0 version of the platform gave specific designations to users (Verified Influencers, Parler Affiliate, Verified Real Member, Parler Partner, Private Account, Early Parler-er, and Parody Account).¹⁰ A plethora of account holders also appeared to come from outside the United States, and researchers have documented evidence of Russian influence operations conducted by the Internet Research Agency that targeted far-right groups on the platform.¹¹

Trump campaign insiders and prominent proponents of his brand of politics were among the most actively engaged on the site, which began to slowly increase its user base in mid-2019 as the 2020 election primaries grew nearer. Activity on the 1.0 version of the platform reached new peaks after the November 3, 2020 election, a wave that crested on January 6, 2021 and then was halted after major tech companies suspended or ended web hosting and mobile app services during the January 7-10 time frame.

Parler’s unique business model and user-driven content moderation drew scrutiny from the media and researchers early on. Empirically based research on the platform, however, only began to surface in academic journals after the January 6 attack. Among the most significant studies to date is one published in February 2021 by Maxwell Aliapoulios and Emmi Bevensee, et al.¹² Based on data collected from a little over 4 million Parler user accounts containing a little over 183 million posts made from August 2018 to January 2021, the detailed study provides a useful snapshot of Parler’s user base during a period when it experienced high levels of growth. The Aliopoulios, et al., dataset also contains metadata about 13.25 million users and provides important insights into how the platform’s governance structure and design shaped user behavior for a substantial slice for the roughly 15 million users who were on Parler before January 10, 2021.¹³

Researchers on the Aliapoulios study note, for instance, that Parler experienced growth spurts that showed close proximity to real world events involving polarizing public figures like Candace Owens, or polarizing events like the police killing of George Floyd and subsequent protests, and Election Day itself. They also note that only two percent of users in their dataset opted to upload a photo of themselves and a photo identification under Parler’s opt-in verification rules at the time.¹⁴ Aliapoulios, et al., suggest that most of the content moderation on the site is done manually as opposed to by automated means.

One curious feature also noted is that the early version of Parler allowed users to monetize their online activity by permitting them to “tip” users for content creation. The tipping scheme also includes an Ad Network and Influence Network which allows for users “to pay for or earn money for hosting ad campaigns.”¹⁵ Given the number of designated influencers on the site, including those with close ties to Trump or conservative personalities like Proud Boys leader Enrique Tarrio, the monetization scheme raises interesting questions about the degree to which election disinformation on the platform contributed to profit making schemes. There were 596 prolific platform influencers with verified accounts and large followings designated as “Gold Badge,” users at the time of the study and almost 40 percent of those highly active users had a following of more than 10,000 users.¹⁶

Since so much of the platform’s activity clearly tracks with real world events related to the content moderation decisions of other mainstream platforms like Twitter and Facebook, this suggests, at minimum, that Parler’s “tipping” model and that adopted by other fringe platforms is worth greater scrutiny. This is especially true since so much of the content on Parler appears to skew in favor of Trump and far-right movements like QAnon and to pull in and push out content from Twitter and Facebook, making Parler an effective campaign vehicle for spreading false information about elections on multiple platforms.

It is perhaps because of those features that Parler also drew interest from online hacktivists concerned about the spread of hate speech and disinformation on the Internet. One such hacktivist, the Twitter user @donk_enby, in fact, made this clear in social media posts when releasing a huge cache of Parler data five days after the Capitol breach. The hacktivist scraped and archived much of the website’s content after it became clear that hundreds of Trump supporters had stormed the Capitol and simultaneously uploaded potentially incriminating photos and videos of their activities.¹⁷

When @donk_enby released publicly archived data from Parler on January 11 via Distributed Denial of Secrets (DDOSecrets), a hub for anonymously leaked data, it opened the floodgates for researchers.¹⁸ In all, the total haul consisted of 1,030,523 video posts consisting of 35 terabytes of data. In addition, the hacktivist later made available a tranche of 1,071,975 publicly available image files.¹⁹

Publicly released Parler data from the hacktivist @donk_enby also included 13,254,084 user profiles, with personally identifying information redacted. While @donk_enby reportedly archived all Parler posts, this dataset has not been made publicly available as far as we know.²⁰ Although the tranche of data is huge, it is worth noting that, much like the large cache of data uploaded by Aliapoulious, et al. to Zenodo, it is difficult to ascertain to what extent the data archived by @donk_enby represents the entirety of the content on the website. Moreover, the dataset is really only a snapshot of the platform. It would be difficult to validate missing values.

Both publicly available Parler datasets have clear limitations that have implications for analysis. But there is sufficient evidence to show substantial overlap between both the Zenodo dataset and Parler image and video data archived by @donk_enby. Yet these datasets, which are to a certain degree independent, have relationships and these relationships can serve to evaluate inter-dataset consistency. The 183 million Parler post and comment data from Aliapoulios, et al. contain hyperlinks to embedded media; we found statistically significant overlaps with data contained in the tranche of image and video data released by the hacktivist @donk_enby.²¹ Since some of the attributes of both datasets also seem to align with Stanford’s assessment of the platform, it seems reasonable to conclude that what has been made publicly available to date represents a robust sampling of platform content before it was taken offline.

Much of the Parler post data uploaded to Zenodo contains links to embedded media, including 6,932,087 links to a total of 6,832,016 images with distinct URLs hosted on Parler. Of these links to Parler images in the Zenodo set, 383,115 were to images in the @donk_enby tranche of images. There were also many more links to images hosted on a variety of other services, including more than a million links to Giphy, as well as various news sources, and these are presumably publicly available subject to the terms of robots.txt. These post data also contain 270,831 links to video hosted on Parler, for a total of 261,859 distinct video URLs. Of these links, 242,617 are contained in the 1,030,523 @donk_enby videos.

The sheer volume of publicly available data from Parler is impressive but it is important to note it does have limitations. It is impossible to collect complete data from most social media platforms and the deplatforming of Parler means that the millions of posts we and other researchers have collected represent only a snapshot in time, and likely a highly pixelated one at that. As Aliapoulios and other researchers have noted, none of the publicly available Parler data was collected following the protocols of any statistically valid experimental design and none of it is demonstrably complete by us as a record of what was present or available on the Parler 1.0 platform before it shut down. Rather, these millions of Parler user posts are best categorized as “found data,” and because of that, it is difficult to ascertain whether the publicly available data has been altered in some way either by accident or design.

With the above caveats in mind, there is no question that the publicly available Parler data we collected still holds value. To examine the broad trends within Parler and other social media platforms, we combined that data with curated datasets of our own to gain insights into how and when the online activities of individuals and groups on Parler and other platforms intersected with real world events in the lead up to the attacks of January 6.