The Troll Farm: The Internet Research Agency’s Identity Theatre

The Internet Research Agency occupied a four-story building at 55 Savushkina Street in St Petersburg. It had a human resources department, shifts, quotas and key performance indicators. Employees were assigned personas and required to produce a set number of posts and comments per shift, with targets for engagement.

It was financed by Yevgeny Prigozhin, and on 16 February 2018 Special Counsel Robert Mueller indicted thirteen Russians and three entities in connection with its operations against the 2016 United States election.

The indictment is the best available primary source on the method, because it is specific.

The IRA operated accounts posing as American individuals and organizations across Facebook, Instagram, Twitter and YouTube. Some had hundreds of thousands of followers. They posed as Black activists, as Christian conservatives, as Texan secessionists, as Muslim American groups, as animal-rights organizations. Employees studied American political vocabulary and were corrected internally for insufficient authenticity.

They bought political advertising in dollars. They organized real events on American soil, arranging rallies through local coordinators who did not know who they were talking to, and in at least one documented instance produced simultaneous opposing rallies at the same location in Houston.

The mechanism is impersonation of in-group identity, and it is the reason this is different from ordinary propaganda.

A message identified as foreign propaganda triggers resistance. A message from an apparent member of your own group does not, because shared identity is one of the strongest credibility cues available — chapter 12's unity principle, weaponized at scale by people who were not members of any of the groups they claimed.

The second mechanism is manufactured consensus. Many coordinated accounts produce the appearance of a distribution of opinion, and chapter 28 establishes that people have no reliable internal check against a fabricated majority. Nobody can sample public opinion directly; everyone infers it from what they see.

And the platforms did the distribution. Brady and Van Bavel's research found that moral-emotional language substantially increases sharing within ideological networks, and the IRA's content was selected for outrage precisely because the recommendation systems reward it. The operation did not need to reach an audience; it needed to produce material the audience's own infrastructure would spread.

The measurable effect on the election outcome is genuinely contested and probably small. The measurable effect on trust — the residue that any American political content might be foreign, and that any opponent might be a bot — is not small, and it is the outcome chapter 231 identifies as the point.