attentionmachine.org

The Attention Machine

← Symptoms

Elections & Democratic Harm

The connection between platform design and democratic harm became impossible to ignore in 2016. It has been documented in detail since.

How to read this page. Tap any underlined word to see the precise academic term and a short definition. Expand any "Deeper" box for the evidence and contested points. The main text works on its own — you can skip both and still get the whole argument.

2016: The first documented large-scale operations

Russia's Internet Research Agency ran systematic operations to influence the 2016 US presidential election. Fake American personas, inflammatory content targeting racial and political divisions, advertising purchased in roubles. This is documented in the Mueller Report (2019) and the US Senate Intelligence Committee's five-volume report — the most comprehensive public account. The platforms' own Congressional testimony confirmed the reach: IRA content reached 126 million Facebook users, 1.4 million Twitter users, and millions more on Instagram and YouTube.

Cambridge Analytica harvested data from 87 million Facebook profiles without consent and used it to build psychological profiles for political micro-targeting in the 2016 US presidential election. The claim that the same data swung the Brexit referendum is a separate and weaker one: the UK Information Commissioner's Office investigated and found the harvested data — relating to US-registered voters — could not have been used in the referendum. Whether the US targeting changed the result is contested by political scientists; that the infrastructure existed and was used is confirmed.

×

Micro-targeting

Aiming a tailored message at a very narrow audience — sometimes individuals — defined by inferred traits rather than broad demographics. In politics, it means different voters see different, individually-optimised appeals that no one else sees, which makes the overall campaign hard to scrutinise or fact-check. Its real-world persuasive power is genuinely debated; its opacity is not.

Sources

How the algorithm amplifies political polarisation

Facebook's 2018 algorithm update, intended to promote "Meaningful Social Interactions," was studied empirically using data from both the United States and Italy. Peer-reviewed research published in 2025 found the update increased ideological extremism and affective polarisation in both countries. A commercial product decision had measurable effects on political opinion at national scale.

×

Affective polarisation

Not disagreeing more about policy, but disliking and distrusting the other side more — coming to see opponents as enemies rather than fellow citizens who are wrong. It is the kind of division that engagement-ranked feeds reward most, because contempt and us-versus-them content travels further than agreement. It corrodes the shared ground democracy needs even when people's actual policy views barely move.

Sources

  • Iyengar, S. et al. (2019), The Origins and Consequences of Affective Polarization in the United States, Annual Review of Political Science 22.

Internal Facebook data found that recommendation features were responsible for 64% of extremist group joins. The mechanism: users are walked incrementally toward more provocative content because outrage drives engagement. Political content is among the most outrage-generating. The algorithm has no concept of democratic health.

Romania 2024: a presidential election annulled

In November 2024, previously unknown ultranationalist candidate Călin Georgescu won the first round of Romania's presidential election with 23% of the vote, having polled in single digits weeks before. Romanian intelligence services declassified documents showing coordinated algorithmic amplification and paid promotion of Georgescu on TikTok. The Constitutional Court annulled the election — the first EU presidential election annulled on grounds of digital interference. The electoral process was restarted entirely.

The Romanian authorities asked the European Commission to investigate TikTok's role under the Digital Services Act. TikTok disclosed it had disrupted small clusters of coordinated accounts promoting Georgescu. The intelligence documents cited algorithmic amplification beyond what these accounts alone could explain.

What allied intelligence agencies say

Four days before the 2024 US presidential election, the ODNI, FBI, and CISA issued a joint public statement confirming Russia was conducting influence operations targeting the election. The Dutch AIVD's 2025 annual report described the threat picture as unprecedented in 80 years of the organisation's existence. The German BfV recorded extremism crimes at an all-time high, with online platforms identified as primary radicalisation infrastructure.

The Dutch media regulator, the Commissariaat voor de Media, published a report in May 2026 concluding that algorithmic feeds are "demonstrably risky for democracy" because they influence what citizens see and who they follow, and called for better enforcement of existing European legislation.

The structural argument

Democratic elections require that citizens can form genuine opinions based on accurate information about the world. A system that systematically amplifies false, inflammatory, and polarising content — because it drives engagement — is in structural tension with that requirement. This is not primarily an argument about any one election or any one platform. It is an argument about what information environments do to the conditions that democracy requires.

How we know — what is confirmed, and the discipline on 'it changed the result'

This page deliberately separates three things that are often blurred together. Confirmed: the IRA operations and their reach (Mueller, Senate Intelligence Committee, platform testimony); the Cambridge Analytica harvesting; the Facebook 2018 algorithm study's findings; the Romanian annulment (a court ruling). Contested: whether micro-targeting or influence operations actually change election outcomes — political scientists are sceptical that persuasion at the ballot box is as powerful as the marketing of these tools implies. Refuted: that Cambridge Analytica's harvested data swung Brexit (UK ICO).

Why the distinction matters. The site's argument does not need any single election to have been decided by a platform. It is the structural claim — that engagement-optimised feeds systematically degrade the information conditions democracy depends on — that the evidence supports. Overclaiming a decided election would weaken exactly the case we are making by making it refutable on a technicality.

Sources

What you can do →
Health Misinformation →
State Actors & Information War →