The Computational Propaganda Record

I. The Mechanism

The human brain uses a heuristic: if many people appear to believe something, it is probably true. This heuristic — social proof — is among the most robust cognitive shortcuts documented in the social psychology literature. It is efficient. It is usually accurate. And it is exploitable at scale by anyone who can manufacture the appearance of distributed agreement without the substance of it.

Computational propaganda is the systematic deployment of coordinated inauthentic accounts — bots, sock puppets, troll farms, astroturf organisations — to create the appearance of broad organic agreement where no such agreement exists. The target is not the individual's reasoning capacity. It is the individual's social proof heuristic. The individual does not need to be persuaded by an argument. They need to perceive that many others have already been persuaded.

II. The Evidentiary Record

The documentation is now substantial and multi-jurisdictional.

The Internet Research Agency (IRA). The Mueller Report's indictment of thirteen Russian nationals and three Russian entities (February 2018) documented the IRA's operations in granular detail: a St. Petersburg-based organisation with a monthly budget of approximately $1.25 million, employing hundreds of operatives who created and operated thousands of fictitious American personas across Facebook, Twitter, Instagram, and YouTube. The operatives worked in shifts, posting in American English during American time zones, with assigned ideological personas (some left-wing, some right-wing, some single-issue) designed to amplify existing divisions in American political discourse.

The IRA's technique was not persuasion. It was volume flooding: overwhelming organic conversation on target topics with coordinated inauthentic content at sufficient volume that the inauthentic content became indistinguishable from the organic conversation. The objective was not to install a specific belief but to create the perception that a specific belief was widely held.

Meta's Coordinated Inauthentic Behaviour Reports. Beginning in 2018, Meta (Facebook) began publishing quarterly reports documenting CIB networks it had identified and removed. As of 2025, Meta had removed over 200 influence operations originating from more than 70 countries. The reports document a consistent pattern across operations: account creation (using stolen photos, fabricated biographies, and AI-generated profile images), network building (cross-following, mutual engagement, group creation), and coordinated amplification (simultaneous posting on target topics, coordinated reactions and shares, hashtag flooding).

The Oxford Internet Institute's Computational Propaganda Project. Samantha Bradshaw and Philip Howard's research programme documented that by 2020, organised social media manipulation campaigns had been identified in 81 countries — up from 28 countries in 2017. The research distinguished between government-run operations (state propaganda), private contractor operations (hired by governments or political campaigns), and civil society operations (partisan groups using CIB tactics). The finding: computational propaganda is not a technique used by a specific type of actor. It is a capability available to any actor with sufficient resources and motivation.

III. The Three Techniques

Volume Flooding

The first and simplest technique. Coordinated inauthentic accounts post on a target topic at sufficient volume to dominate the conversation — flooding organic content below the fold, trending the topic artificially, and creating the perception that a massive number of independent voices are all saying the same thing simultaneously.

Volume flooding does not require sophisticated content. It requires volume. A thousand bot accounts posting a single hashtag simultaneously produces a trending topic. A trending topic triggers media coverage. Media coverage triggers organic conversation about the topic. The organic conversation, stimulated by the artificial trend, produces genuine engagement — at which point the inauthentic origin of the trend is invisible because genuine users are now participating.

Origin Diversity

The second technique addresses the most obvious weakness of volume flooding: if all the accounts appear to originate from the same source, the social proof heuristic does not activate. Origin diversity creates the appearance of geographically, demographically, and ideologically distributed agreement — accounts that appear to be from different cities, different age groups, different professions, and different political orientations, all independently arriving at the same conclusion.

Origin diversity requires more sophisticated account creation: varied profile photos (often AI-generated or stolen from real users), varied biographical details, varied posting histories (accounts are "aged" by posting innocuous content for weeks or months before being deployed in a coordinated operation), and varied linguistic patterns (different vocabularies, different levels of formality, different regional idioms).

Timing Coordination

The third technique exploits the platform's trending algorithm. Coordinated simultaneous posting — all accounts posting within a narrow time window — produces a spike in engagement velocity that the trending algorithm interprets as organic virality. The algorithm then amplifies the content to a broader audience, which produces genuine organic engagement, which further accelerates the trend.

The timing coordination must be precise enough to trigger the trending algorithm but imprecise enough to appear organic. Sophisticated operations use staggered posting windows (not simultaneous but within a 15-30 minute band) and varied initial engagement patterns (some accounts post first; others respond; others share; others react) to simulate the cascade pattern of genuine organic virality.

IV. The Detection Problem

The fundamental detection problem is asymmetric. Creating a CIB operation requires resources — accounts, personas, coordination infrastructure — but the resources are modest relative to the impact. Detecting a CIB operation requires analysing billions of accounts and interactions across multiple platforms, identifying statistical anomalies that distinguish coordinated behaviour from correlated but independent behaviour, and doing so in near-real-time before the operation achieves its objective.

The detection methodologies that exist — network analysis (identifying clusters of accounts with unusually high mutual engagement), temporal analysis (identifying posting patterns inconsistent with organic behaviour), linguistic analysis (identifying accounts with unusually similar language patterns), and account-age analysis (identifying newly created accounts deployed in coordinated bursts) — are effective for retrospective identification of operations that have already been disclosed. They are less effective for real-time identification of operations that have been designed to evade the specific detection methodologies the platform is known to use.

This is the computational propaganda equivalent of the Audit Capture Cycle (AOA-005): the detection methodology is published; the operators design their next operation to evade it; the methodology is updated; the operators adapt. The cycle favours the attacker because the attacker needs only to evade the specific methodologies in use, while the defender needs to detect every evasion across an effectively infinite attack surface.

V. The Democratic Consequence

The Deliberative Problem series (DP-001 through DP-005) specified that democratic deliberation requires shared epistemic standards — a common evidentiary basis from which different perspectives can evaluate the same facts. Consensus engineering attacks this prerequisite directly.

When the apparent distribution of belief is manufactured rather than organic, the epistemic signal that "many people believe X" — which is normally evidence that X has survived many independent evaluations — becomes meaningless. The social proof heuristic, which evolved to aggregate distributed intelligence, is converted into a vulnerability. The population cannot distinguish between a belief that many people independently arrived at (evidence of likely accuracy) and a belief that many accounts were coordinated to appear to hold (evidence of nothing except coordination).

The consequence is not merely that people believe false things. It is that the mechanism by which populations correct false beliefs — the aggregation of many independent evaluations — is itself compromised. The immune system has been captured.

VI. Connection to the Programme

The Semantic Record (SR-001 through SR-006) documented how the content of priors is installed through language. The Computational Propaganda Record documents how the installed priors are reinforced through manufactured apparent consensus. The two mechanisms operate in sequence: semantic capture provides the vocabulary; consensus engineering provides the social proof that the vocabulary is universally accepted.

The Neural Complexity sciences page documented that algorithmic engagement loops tighten prior precision-weighting — the brain becomes more confident in its existing model with each reinforcing exposure. Consensus engineering is the social-scale version of this mechanism: manufactured apparent consensus provides the social reinforcement that tightens priors across the population simultaneously.

Named Condition

Consensus Engineering — the systematic deployment of coordinated inauthentic behaviour to create the appearance of distributed organic agreement, exploiting the social proof heuristic to substitute manufactured apparent consensus for genuine epistemic evaluation. Identifiable through three detection methodologies: network analysis (account clustering), temporal analysis (coordinated posting patterns), and origin analysis (account-age and diversity anomalies).