Internet Research Agency (Russian operation aimed at US audiences, 2014 to 2018)

Mechanism

The IRA registered as a Russian company in July 2013 and worked from 55 Savushkina Street, St Petersburg. From June 2014 it operated through other Russian entities, and Concord paid it through about fourteen affiliate bank accounts, labelled as software support (indictment ¶¶ 10 to 11). By July 2016 more than eighty staff were assigned to its US section. The indictment says the IRA sought to conduct what it called "information warfare against the United States of America," and the Special Counsel's report quotes a May 2014 planning document that set the goal "to spread distrust towards the candidates and the political system in general."

17 April 2018. The building at 55 Savushkina Street that the indictment names as an IRA operating hub, photographed for Voice of America. Public domain, via Wikimedia Commons.

The method was persona building. Staff created individual accounts and then themed pages that claimed to be run by Americans, including a Twitter account presented as the Tennessee Republican Party. Pages mostly posted material that suited their audience with occasional "payload" posts. The Senate Intelligence Committee cites the "Army of Jesus" page (over 216,000 followers), which posted devotional lines in late October 2016 and on 1 November posted "HILLARY APPROVES REMOVAL OF GOD FROM THE PLEDGE OF ALLEGIANCE." Paid ads were a small part: roughly 3,500 ads costing about $100,000, against about 61,500 organic Facebook posts, 116,000 Instagram posts and 10.4 million tweets. An IRA manual also tasked Twitter staff, who ran a network of automated accounts, with preparing messages "to get a hashtag to the top."

Recruitment moved the operation offline. The IRA promoted rallies through its pages, found an American to coordinate, in most cases told that person it could not attend, and paid some participants, in some cases over $1,000. Facebook counted at least 130 events promoted through 13 IRA pages.

'Russian Ads on Facebook' (House Intelligence Committee exhibit). 2018 report page reproducing IRA Facebook pages built for different American identities. House Permanent Select Committee on Intelligence; public domain.

Audiences documented

The record shows the same staff addressing opposed audiences at once.

• Black Americans. The Senate committee found that "no single group of Americans was targeted by IRA information operatives more than African-Americans." Pages included "Blacktivist" (about 11.2 million engagements) and "Don't Shoot Us." In late 2016 the indictment records posts urging Black voters to stay home or vote third-party.
• Conservatives and the right. "Being Patriotic," "Secured Borders," "Stop A.I.," "Heart of Texas," "South United" and "Tea Party News." From mid-2016 "Being Patriotic" and IRA Twitter accounts promoted pro-Trump rallies, including 'Miners for Trump' rally poster (Internet Research Agency).
• Veterans and police supporters. Instagram accounts "@american.veterans" (215,680 followers), "@pray4police" (127,853) and "@cop_block_us" (109,648); Oxford researchers found ads keyed to "Veterans" interests and to support for police against Black Lives Matter.
• Left and progressive audiences. "LGBT United" and an Instagram account on sexual and gender identity rights (156,465 followers), plus left-leaning accounts that criticised Clinton and promoted Bernie Sanders or Jill Stein (Senate Vol. 2).
• Religious and other audiences. "Army of Jesus" (conservative Christian themes), "United Muslims of America" (over 300,000 followers when removed) and, mostly after the election, "Brown Power" for Latino audiences.

'Miners for Trump' rally poster (Internet Research Agency). Poster for 2 October 2016 rallies in Pittsburgh and Philadelphia, reproduced in the Special Counsel's report, Vol. I, p. 31. US Department of Justice; public domain.

Techniques are tagged where the record supports them. Astroturfing is the core method. Outrage bait: Oxford researchers describe appealing to group narratives, then posting content "intended to elicit outrage from these groups." Narrative laundering: the false Tennessee Republican account, American outlets that quoted IRA tweets as public opinion, and what DiResta et al. call a "media mirage" around Black community media. Manufactured consensus: the bot network and many personas repeating one line about Clinton. Scapegoating, one stream only: Oxford researchers found "prejudiced and bigoted" statements about minorities, particularly Muslims, aimed at right-leaning audiences, while another IRA page posted positive material about Islam. Identity fusion was considered and not tagged: the record shows identity used to build trust, not to make checking a betrayal.

Evidence

• June 4, 2014. Two IRA employees enter the US on visas after stating a false purpose for the trip. Special Counsel Vol. I, p. 21; indictment ¶¶ 15, 17.
• May 21, 2016. "Heart of Texas" and "United Muslims of America" promote opposing events at the same Houston Islamic centre at the same time, at a stated cost of about $200. Senate Vol. 2, p. 47.
• May 29, 2016. A US person is recruited to hold a sign at the White House reading "Happy 55th Birthday Dear Boss," for Prigozhin. Indictment ¶ 12(b).

16 February 2018. Page 7 of the indictment, United States v. Internet Research Agency LLC et al., No. 1:18-cr-00032 (D.D.C.), Document 1. US Department of Justice filing; public domain.

• 2016-09. The indictment puts the Project Lakhta monthly budget submitted to Concord above 73 million rubles (over $1,250,000). Project Lakhta also had components aimed at Russian audiences, so the figure is not a budget for the US work alone. Indictment ¶ 11(b).
• November 12, 2016. The same operation organises a rally to "show your support for President-Elect Donald Trump" and another named "Trump is NOT my President," both in New York. Indictment ¶ 57.
• 2017-08 to 2018-01. Facebook removes 470 IRA accounts; Facebook told the Senate on November 1, 2017 that content from the IRA "may have been served" to about 126 million people. Twitter identifies 3,814 accounts and notifies about 1.4 million users. Special Counsel Vol. I, pp. 14 to 15, 28.
• February 16, 2018. A grand jury indicts 13 individuals and three companies: conspiracy to defraud the United States (all), wire and bank fraud (three defendants), aggravated identity theft (five). Indictment; Special Counsel Vol. I, Section II and p. 174.
• August 13, 2018 and November 15, 2018. The court denies Concord's motions to dismiss on the Special Counsel's appointment and on Count One. Doc. 55; Doc. 74.
• October 19, 2018. The Justice Department unseals a complaint against Elena Khusyaynova, described as chief accountant of Project Lakhta, alleging interference including in the 2018 elections and a proposed budget of about $35 million from 2016. Nextgov; Lawfare.
• November 6, 2018. US Cyber Command cuts the IRA's internet access on Election Day and for a few days after, according to unnamed officials quoted by the Washington Post on February 26, 2019. Details are not public. Lawfare; MIT Technology Review.
• 2018-12. Two research teams working for the Senate publish analyses of platform data: Howard et al. (Oxford, Graphika) and DiResta et al. (New Knowledge). Both report activity continuing after Election Day 2016, with Instagram rising most.
• 2019. The Senate Intelligence Committee publishes Vol. 2 of its report. On August 9, 2019 the court rules on a discovery protective order: the government had given Concord's counsel about 4 million documents and sought to restrict about 3.2 million (Doc. 169).
• March 16, 2020. The Justice Department moves to dismiss the charges against Concord, and Judge Dabney Friedrich grants the motion. NPR; CNN.
• September 1, 2020. Facebook removes accounts it attributes to "individuals associated with past activity by the Russian Internet Research Agency" (Peace Data, a site that used real freelance writers). Graphika.

What the record does and does not show about effect

Scale is documented. Effect on belief and votes is not established and is contested.

• Reach figures are platform estimates of exposure. Facebook said 29 million people were served IRA content directly and 126 million "may have been." DiResta et al. record about 187 million Instagram engagements and note they may include click farms. Engagement is not persuasion.
• Bail et al. (PNAS, 2020) linked 1,239 US Twitter users to IRA data from late 2017 and found "no evidence" that interaction with IRA accounts substantially changed six measures of attitudes and behaviour over one month. They state that they cannot say whether the accounts influenced the 2016 election.
• Eady et al. (Nature Communications, 2023) linked survey respondents to their 2016 Twitter feeds. One percent of users accounted for 70% of exposures, exposure was concentrated among strong Republicans, and they found "no evidence of a meaningful relationship" with changes in attitudes, polarisation or voting. Both studies cover Twitter only.
• Kathleen Hall Jamieson (Cyberwar, Oxford, 2018) argues it is probable that the Russian effort, which included hacked material as well as trolls, helped elect Trump. That is an inference about the whole effort, not a measurement of IRA exposure.
• The Special Counsel found no evidence that any US person knowingly or intentionally coordinated with the IRA. Trump campaign affiliates promoted dozens of IRA posts, and the report says no campaign official was shown to understand the requests came from foreign nationals. It found no similar connections with the Clinton campaign.
• The Senate committee found IRA content "overtly and almost invariably supportive" of Trump before the election, but also that most content concerned social divisions rather than the election. In the ten weeks before the election, Twitter said about 8.4% of 175,993 IRA tweets were election-related. IRA accounts also targeted Republican primary candidates, and after the election the operation ran anti-Trump events.

This dossier does not say the operation explains the 2016 result. The record shows a large, deceptive, well-funded operation and gives no agreed measure of what it changed.

Sides on the record

• Prigozhin and Concord. On the indictment in February 2018 Prigozhin was quoted by RIA Novosti: "Americans are very impressionable people; they see what they want to see." In November 2022 a statement posted by his company's press service said: "Gentlemen, we have interfered, are interfering and will interfere." Concord alone hired US counsel. It argued the Special Counsel was unlawfully appointed and that Count One failed to state an offence, and its lawyer said in September 2019, "Even 18 months in we still don't know what we're charged with" (Courthouse News). The court rejected both motions.
• The dismissal. The government said Concord had been "eager and aggressive in using the judicial system to gather information about how the United States detects and prevents foreign election interference," and that a trial risked exposing law enforcement tools. CNN reported that some Justice Department lawyers had doubted the decision to charge the companies. No court reached the facts.
• Kremlin. The Senate committee acknowledges "Moscow's denials" and concludes that Prigozhin's role and ties "point to significant Kremlin support, authorization, and direction." The indictment charges Prigozhin and Concord as funders and does not allege state direction; the Special Counsel found evidence of his relationship with the Ministry of Defense and high-ranking officials.
• Same definition, other flags. Astroturfing lists persona cases from the US military and US party-aligned groups scored with the same test; see also CENTCOM / Pentagon “gc_” newsroom network and Russia Social Design Agency — Projects 2026. Each case page states its own scale and disclosure.

What would falsify this read

Evidence that the pages and accounts were run by real, independent Americans; that the funding chain ran to someone other than Concord; or that the indictment's paragraphs on personas, stolen identities and hidden origin were wrong. Evidence of a measured change in votes from IRA exposure would change the effect section, not the classification.

Literacy counter

Treat a fast-growing "community" page as unverified until it names a real organiser. Check who runs an event before joining and whether the same operator promotes opposing events. Notice when a single account is the only source for a claim about "what people think." Prebunking and lateral reading are described in Inoculation and prebunking and Media literacy; platform takedown reports and their limits are in Platform interventions: labels, nudges, notes, friction and takedowns. Related cases: Black Lives Matter and Blue Lives Matter: two slogans, their campaigns, and a foreign operation that posed as both, MAGA and the Resistance: grassroots organising for and against Trump, 2016-2021, 'For whites only': the forged Klan letters sent to Olympic committees in 1984. The technique is anatomised for recognition in Anatomy of deception: how documented operations worked, and how to spot them.