Digital Library
Bad Gateway How Deplatforming Affects Extremist Websites
Topic:
Antisemitism & Antizionism, Israel & Regional Politics
Principal Investigators:
Center for Technology and Society, Megan Squire
Study Date:
2023
Source:
Anti-Defamation League (ADL)
Key Findings:
This report shows that deplatforming can decrease the popularity of extremist websites, especially when done without warning. ADL presents four case studies of English-language, U.S.-based extremist websites that were deplatformed: the Daily Stormer, 8chan/8kun, TheDonald.win/Patriots.win, and Nicholas Fuentes/America First. In all of these cases, the infrastructure service providers considered deplatforming only after highly publicized or violent events, indicating that at the infrastructure level, the bar to deplatforming is high. All of the site administrators in these four cases also elected to take measures to remain online after they were deplatformed. To understand how deplatforming affected these sites, ADL collected and analyzed publicly available data that measures website-popularity rankings over time.
Four important lessons about how deplatforming affects extremist websites:
(1) It can cause popularity rankings to decrease immediately.
(2) It may take users a long time to return to the website. Sometimes, the website never regains its previous popularity.
(3) Unexpected deplatforming makes it take longer for the website to regain its previous popularity levels.
(4) Replicating deplatformed services such as discussion forums or live-streaming video products on a stand-alone website presents significant challenges, including higher costs and smaller audiences.
Findings show that fighting extremism online requires not only better content moderation and more transparency from social media companies but also cooperation from infrastructure providers like Cloudflare, GoDaddy, and Google, which have avoided attention and critique.
Methodology:
“Deplatforming” means removing infrastructure services they need to operate, such as website hosting.
One way to understand the effect of deplatforming is by measuring a website’s usage statistics before and after its critical infrastructure services are removed. But since ADL do not have access to the internal logs of each website, the researchers sought an alternative source of data about site traffic and usage.
The effects of deplatforming extremist websites were measured by using two free, publicly available domain-ranking lists: the Alexa Top 1 Million and the Cisco Umbrella Top Million. Researchers use such lists as indicators of site popularity because they are free and available for multiple years.
Cisco Umbrella: The first data set comes from Cisco’s Umbrella OpenDNS service. OpenDNS is used by millions of computers around the world to translate human-friendly domain names (e.g., adl.org) into the numeric addresses computers need to find each other on the internet. The Umbrella Top Million list, first released in January 2017, ranks websites by how many OpenDNS users have requested a domain name. The Umbrella data set is compiled by counting these passive DNS requests and ranking the top million websites each day.
Alexa: Now discontinued, Alexa was a top-list service that ranked websites using a proprietary methodology based largely on the activities of its users, who installed its browser extension and had their web traffic compiled into the rankings. Unlike the Umbrella data, Alexa users were active participants in the data-collection effort, and only web traffic was included. The advantage of Alexa data is that it is long-running. Since the Alexa archive included files beginning in 2009, the researchers knew that it would include data from the early days of dailystormer.com and 8chan (predating Cisco Umbrella’s coverage).
ADL retrieved the entire history of Alexa Top Million lists from the archive server associated with the top lists comparison paper from Naab et al. ADL retrieved the Cisco Umbrella top million lists from Cisco’s archive server. ADL downloaded each of the daily files of the top 1 million sites, selected any domain matching their case studies, and loaded each day’s rank for that domain into a database. In total, there were 4,871 Alexa daily files and 2,113 Umbrella daily files.
The primary method of analysis was a day-by-day ranking comparison over time for each domain. The domains tended to be used linearly, with one being deplatformed and the next taking its place, which lent to a longitudinal comparison. ADL then produced day-by-day ranking comparisons for each data source and each case study. However, given the limitations of top lists in general and Alexa and Umbrella top lists specifically, the researchers focused on finding broad patterns that persist across multiple weeks or months rather than looking at single-day peaks or valleys.