“I must be honest — I never imagined crisis PR could be handled by AI,” says crisis communications consultant Winner Wang.
At first glance, crisis management relies heavily on written communications. In reality, it demands rapid decision-making, high-stakes negotiation, relationship management, and emotional support — all deeply human skills. Yet Winner has identified that the most catastrophic aspect of crisis management is rarely an imperfect resolution; it is getting the initial response wrong. If a company veers off track in the first 30 to 90 minutes of a public relations disaster, subsequent damage control often comes too late.
This stark realization led her to transform years of crisis management protocols into a first-response artificial intelligence tool dubbed "SOS AI."
The concept emerged from Winner's experience as an internal trainer for senior executives at the state-owned oil company CPC Corp. After delivering a three-hour session on crisis response to 50 division heads, she was flooded with private messages from executives who called the strategies "life-saving." It became immediately clear that while not every organization has the resources to keep high-priced advisors or lawyers on retainer, virtually every business faces the looming threat of sudden public backlash.
An Overnight Digital Twin
The tool's genesis was remarkably rapid. Winner handed over her training videos, proprietary materials, and compiled case studies to an engineer friend: Wang Chih-ching, co-founder and chief digital officer of LeadBest Consulting Group. During a single sleepless night, Wang experimented with building an AI model using the provided content. By morning, he had created a working prototype.
When a crisis hits, corporate leaders typically experience a cascade of panicked questions: Who should I contact? What should I do? What actions do I need to take right now? Having an immediate, structured outline of what to do in the first 30 minutes can pull individuals back from the brink of collapse and into an actionable framework.
SOS AI fulfills this triage function precisely. The system first prompts users to clearly describe the incident. It then guides them through Winner's established methodology, assessing the source and trajectory of the crisis before organizing potential response strategies and backup requirements.
Generic AI chatbots, which lack media timing awareness and crisis logic, often provide advice that looks professionally polished but is fundamentally flawed in practice. SOS AI, however, codifies proven methodologies into structured steps, significantly reducing the risk of fatal early missteps.
According to Wang, the initial modeling approach aimed to create a “digital twin” of crisis management expertise. By deconstructing and structuring existing presentations and case studies, developers shaped the AI's role, tone, and response consistency. Crucially, they implemented strict guardrails to prevent the AI from generating responses that might worsen PR disasters or cause secondary harm to affected parties in sensitive situations.
The system relies on a boundary-driven approach to human-machine collaboration: It offers front-end self-service to stabilize the situation but is programmed to recognize when a crisis escalates beyond its scope. At this point, it advises users to seek human counsel.
The Unreplaceable Human Element
Winner identifies SOS AI's optimal window of effectiveness as the first 60 to 90 minutes of a crisis.
“If you follow the steps, you basically won't go off track,” she says. During this golden hour, the AI doesn't make final decisions. Instead, it organizes chaotic information into interpretable formats, enabling human executives to make faster, more accurate judgments. The AI can draft two or three different tones of messaging, but the final call on settlements, legal actions, apology videos, or press conferences remains entirely human.
What can't AI accomplish? Winner's answer is unequivocal: trust and emotions.
Crisis management fundamentally involves managing human feelings. Winner frequently fields calls from clients who break down in tears — a level of psychological distress that code cannot soothe.
“AI probably cannot provide that continuous sense of reliability and security,” she notes.
Furthermore, media relations require highly nuanced judgment. The same statement sent to different reporters, at different times, with different tones, can produce vastly different outcomes. These subtleties remain far beyond the capabilities of reliable automation.
Ultimately, SOS AI is not designed to turn machines into standalone consultants. Its true value lies in helping organizations maintain their footing during their most panicked moments, ensuring that in the chaos of the first hour, they don't say or do the one wrong thing that pushes them past the point of no return. Human empathy, accountability, and negotiation remain the only way to see a crisis through to its end. （Related： Ko Wen-je’s Prosecutorial Complaint Exposes Systemic Flaws in Taiwan’s Justice System ｜ Latest ）
