Not just the fake news. The whole machine.

Context

A state institution needed a real tool to counter the wave of disinformation flooding social media around politically sensitive events. Conventional media monitoring was not enough: it surfaced individual pieces of false content but missed the structure behind them.

Problem

Coordinated botnets and troll farms are the harder problem. They distribute manipulative content in ways designed to look organic. Manual verification cannot keep up, and text-only analysis ignores half the signal: the images, the timing, and the network of accounts pushing the message.

Solution

We deployed a multimodal AI monitoring system. Text and image are analysed jointly. Network analysis identifies clusters of accounts behaving in coordinated ways. The output is not a list of suspicious posts. It is a map of the infrastructure behind a campaign.

Implementation

The system tracks keywords, narratives and media in real time. Multimodal models classify content by topic, sentiment and propaganda markers. Graph analysis identifies coordination patterns: synchronised posting, shared content with subtle variation, network-amplification topologies, account-creation bursts. Alerts surface campaigns at the level of structure, not individual posts.

Outcome

We mapped the account networks and distribution patterns behind specific disinformation campaigns, disrupting the infrastructure rather than chasing individual posts. The same multimodal stack extended to adjacent intelligence and security workflows.