How DCE Uses AI — Part 1: Reading the Conversation
At any given moment, millions of conversations about vaccines are unfolding across social media, search engines, messaging apps, radio and the press — far more than any human team could ever read, let alone interpret in time to act.
DCE built its AI ecosystem to address both sides of this changing environment. AI helps teams detect, understand, and respond to emerging narratives faster than any manual process alone could. At the same time, DCE has made AI an object of real-time scrutiny, because what the machines tell people about vaccines matters as much as what people tell each other.
Curious what happens when the conversation is with a machine — and what DCE does about it? That is the companion piece, “How DCE Uses AI — Part 2: From the Machines to the Field.” This part stays with the human conversation: how DCE listens, understands and stays ahead.
Listening at a scale no human team could match
DCE uses AI-supported processes to build and refine complex monitoring queries across more than 200 social, search, and content platforms.
The system goes beyond written posts. Where source access allows, it transcribes and categorizes images, videos, radio broadcasts, podcasts, and memes, making visual and audiovisual content searchable and analyzable for the first time at this scale.
Content is analyzed in 187 languages, with dual indexing in both the original language and an automated English translation. This matters enormously for low-resource languages where keyword-based tools fail. A rumour spreading in Hausa or Pashto shows up in the system as quickly as one in English or French.
And it’s not just online. DCE is integrating offline data streams — national newspaper monitoring, radio tracking, community feedback systems, quality assurance surveys — through automated pipelines. Once a data source is connected, every subsequent update flows in automatically. No repeated uploads. No manual matching. The system keeps learning.
Understanding what people actually mean
Most social listening tools rely heavily on positive, negative or neutral sentiment. Sentiment alone reveals little about whether someone is asking a genuine question, expressing uncertainty, or is likely to refuse a vaccine or hesitate at the door.
Through the support of Large Language Models, DCE’s listening engines analyse emotional signals, narrative framing, and expressed intent toward vaccination. Is someone hesitant, willing, refusing, or actively advocating? This distinction is the difference between a community that needs reassurance and one that needs a completely different engagement strategy.
DCE’s listening platform classifies signals of willingness, hesitation, refusal and advocacy, helping analysts distinguish between audiences that may need reassurance, practical information, dialogue or a different engagement strategy. Of course, these classifications are probabilistic and context-dependent, and they support human analysis rather than replace it. But they are crucial to the rapid assessment of the response that is needed to prevent crisis communication events and respond to outbreaks.
Millions of individual posts are grouped into coherent narratives and tracked in a derived database. This allows DCE to examine recurring belief systems and storylines rather than treating each post as an isolated event. Analysts can map how a single narrative travels across countries, platforms, and languages over time, see exactly when it starts to accelerate, and what are the dynamics behind its propagation.
Flagging threats before they spread
Potential misinformation is identified through a combination of machine-learning models and rule-based analysis.
AI models trained on historical misinformation patterns read and classify millions of individual pieces of content. In parallel, algorithmic rules assess a set of risk indicators — linguistic, semantic, and metadata attributes that recur in disinformation — refined through two years of DCE monitoring data.
The resulting signals are combined into a risk-scoring process that helps teams to:
- identify streams of potential misinformation and assess their reach and virality;
- prioritise analyst attention where potential harm is highest;
- trigger risk communication alerts when thresholds are crossed.
DCE is currently developing an enhanced automated risk matrix. When complete, medium- and high-risk events will automatically trigger alerts to leadership and accountable teams, significantly shortening response times during emerging immunization crises.
Anticipating what’s coming
Historical data can reveal recurrent misinformation cycles, seasonal narratives and patterns linked to outbreaks, campaigns or public controversies.
DCE is exploring how these patterns can support forecasting and scenario planning. The system can analyse previous content and engagement curves, identify narratives with a history of resurfacing, and highlight conditions associated with increased amplification.
These outputs are probabilistic rather than predictive certainties, but informed estimates that help teams plan. They’re particularly valuable for preparedness planning and for defining the urgency and timing of prebunking interventions: should we act now, or wait and watch?
DCE does not only listen to people — it has also begun to listen to the machines people now ask for advice, and to turn all of this listening into action in minutes. Continue with “How DCE Uses AI — Part 2: From the Machines to the Field.”