AI can find gaps in crisis plans, but teams still have to act

Where plans break

Jason Mudd, CEO of Axia Public Relations, used AI to review a client’s crisis plan. A simulation raised a question the team had not considered: what if the company’s spokesperson became the main subject of the crisis?

Other tests surfaced more practical gaps. A plan had a complete list of crisis-team contacts, but no one had checked whether those people would answer a call after midnight with Do Not Disturb enabled. Another plan had not been updated since social media and AI became distinct channels for crises.

The examples point to a basic problem: a plan can look complete on paper while failing under pressure. AI can help expose that gap, but people still have to test whether the response is workable.

Test more than one scenario

A company should not assume it will face only one problem at a time. One scenario might combine a cyberattack that disables computers and communications with allegations of misconduct against company officials.

The way a prompt is framed also matters. Instead of asking a model to confirm that a plan is good, ask it to act as a tough critic and identify what could break the plan or make it useless.

Scott Kiver, CEO of Reputation Pros, suggests removing confidential and sensitive information before asking questions such as:

Kiver also recommends reviewing the plan from the perspectives of five groups: customers, employees, journalists, regulators and critics. If it cannot withstand a simulation involving those audiences, he says, it may not withstand a real crisis.

AI is not the rehearsal

There is a limit to what a model can establish. Its answers may be inaccurate or inappropriate, and specialists should review any recommendation before a company acts on it. Public AI platforms also present a data risk when crisis plans contain sensitive information about the company or its employees.

Katherine Montgomery, a reputation and crisis-management expert and founder and CEO of Better Together Agency, advises against uploading a plan to a public tool or treating the model’s answer as fact. AI can help teams ask more questions, she says, but experienced people must decide what is accurate, appropriate and right for the situation.

That distinction matters in practice. Kevin Dinino, president of KCD PR, has seen plans pass an AI review and then fail a tabletop exercise. In one case, three executives had to approve a response, and none had ever responded in less than four hours. In the worst part of a crisis, that could leave the company silent for 12 hours.

I think that is the more useful test of AI-assisted planning: not whether a model can find a flaw, but whether a team can act when the flaw becomes real. Companies still need tabletop exercises and should update their plans after tests, and when leaders, communication channels or external events change.

The risk runs both ways. Hacks of companies’ systems using Gemini are a reminder that AI itself can create threats a crisis plan must account for. A model can help a team prepare, but it cannot make a plan reliable—or make the people responsible for carrying it out respond in time.