AI and situational awareness: from social listening to early warning

What situational awareness is, and why time matters

In the first phase of a crisis a crisis team has to do one thing well: know what is going on. In the BOB model — Beeldvorming, Oordeelsvorming, Besluitvorming, or building the picture, forming a judgement and taking a decision — this is the picture-building phase. Only once there is a reliable, current picture of facts, reports and signals can a team judge sensibly and then decide. Some crisis teams additionally use the FRE model: Facts, Rumours, Emotions, plus open Questions. That distinction helps to avoid treating everything that comes in as hard truth.

Traditionally an information manager builds this picture by hand: phone calls, control room messages, news reports and increasingly social media posts are each reviewed and weighed. During a large crisis that volume rises quickly. And precisely in the first hours, when decisions have the greatest effect, there is the least time to work through everything manually.

How AI speeds the process up

What began as a pilot by four Dutch safety regions — Fryslân, Groningen, North and East Gelderland and Utrecht — together with the Netherlands Institute for Public Safety (NIPV, formerly IFV) and iNowit, around a virtual assistant that automatically searches sources and proposes material for the situational picture, is now available to every Dutch safety region. Through DATA4OOV, regions can surface the functionality in their own geo-viewers and in LCMS, the existing crisis management system in which Dutch safety regions maintain their information; the assistant is intended as an addition to it, not a replacement.

This development fits a broader international trend. According to market research by SNS Insider, the global market for AI-driven detection of crisis signals on social media is growing from 1.62 billion US dollars in 2024 to a projected 8.39 billion by 2032, an annual growth rate of 22.8 per cent. That growth shows that more and more organisations are deploying software to pick up early signals from the enormous volume of messages posted during a crisis.

In practice, AI-supported situational awareness rests on four building blocks, as set out by the Public Relations Society of America. First, early signal detection: automatically linking complaints, claims and legal signals so that a pattern becomes visible before a problem becomes public. Second, risk prioritisation: distinguishing risks that are likely from risks that are mainly hypothetical, so a team does not devote equal capacity to every signal. Third, fast and targeted response: automatically generated draft messages in several languages, combined with sentiment analysis showing in real time how an audience is reacting. Fourth, scenario simulation: using AI personas, for instance a virtual "critical journalist" or "activist", a team can test a proposed message for weak points before it goes out.

What AI does and does not automate

It matters to place this development correctly. AI speeds up the classic picture-building phase; it does not replace it. The technology helps to reach a shared, current picture of the facts faster: which reports are coming in, which sentiments dominate, which rumours are circulating and which questions remain open. That is exactly the FRE distinction: facts, rumours, emotions and questions are mapped faster with AI and presented to the crisis team more clearly.

What AI does not do is take over judgement and decision-making. Weighing signals, determining urgency and the eventual choice of a strategy or message remain human work. A virtual assistant may point out that a particular message is being shared unusually often, but whether that is reason to escalate the crisis organisation stays with the crisis team. AI supplies material faster and more completely; the team stays responsible for the conclusion.

Points of attention with AI-supported monitoring

This way of working requires a few practical safeguards. First, the use of sources must stay transparent and traceable. A crisis team must always be able to check where a signal came from and on the basis of which messages a risk assessment was reached. A black-box judgement, where nobody can reconstruct how the AI arrived at a conclusion, is unworkable in a crisis: accountability is inseparable from crisis management, including afterwards during an evaluation.

Second, a human stays needed in the loop for interpretation. Automated monitoring flags patterns, but context, local sensitivities and the question of whether a signal matters in this specific situation call for human judgement. In sensitive or fast-changing situations in particular, the risk of relying too heavily on an automated assessment is real: one missed nuance can lead to a wrong reading of the situation, with consequences for the communication that follows.

AI-supported situational awareness is therefore not a matter of replacing the human, but of letting the human start faster and with a broader overview on the phase that ultimately matters: judging and deciding.

Sources

• Virtual assistant available to every safety region (NIPV, 31 October 2025), in Dutch.
• How AI contributes to effective crisis management at IFV (M&I/Partners), in Dutch.
• AI and the New Era of Crisis Comms (Public Relations Society of America).
• AI-Powered Social Media Crisis Detection Market (SNS Insider, 2025) — 1.62 billion US dollars in 2024 to a projected 8.39 billion by 2032, growing 22.8 per cent a year.