How to Use AI for Crisis Communications: A Practical Guide for Senior Leaders

In a fast-moving crisis, AI is most useful for:

First-draft speed: getting a structured first draft in front of decision-makers quickly. The most experienced communicators in the room can then improve a draft faster than they can write from scratch.

Multi-audience adaptation: taking a core approved message and adapting it for different audiences (staff, customers, regulators, media, board) simultaneously. This is time-consuming without AI and prone to inconsistency.

Sense-checking: using AI to identify what a message might be missing, what a sceptical audience might focus on, or what a hostile journalist might ask based on the draft.

Preparing for questions: generating likely questions from journalists, regulators, or shareholder groups based on the situation and the drafted response.

AI is not useful for the decision-making about what to say. That is a human responsibility. AI helps execute the communications once the strategic decisions are made.

For an initial crisis statement, give the AI as much context as you can provide safely:

'We need a holding statement for a situation where [describe the situation without confidential specifics]. The audience is [describe]. Our key message is [what we want to convey]. Constraints: [things we cannot say, legal sensitivities, what is not yet confirmed]. Tone: [direct, empathetic, measured]. Length: under 200 words.'

Review the draft against these criteria: Does it acknowledge the situation without admitting to things not yet established? Does it express appropriate concern for those affected? Does it indicate the next steps we are taking? Does it avoid language that will read badly under scrutiny?

For multi-audience adaptation: once the core statement is approved, 'Adapt this statement for [audience]. The core message must remain the same but adjust the language, level of detail, and tone for this audience. Maintain the same factual claims exactly.'

Maintaining factual consistency across audience versions is critical; any divergence will be identified and weaponised.

Before issuing a crisis communication, use AI to stress-test it:

'Read this statement from the perspective of a sceptical journalist covering this story. What questions would you immediately ask? What does the statement leave unexplained? What phrases or words might attract criticism?'

'Read this from the perspective of an affected employee/customer. What are they likely to feel on reading this? What questions will they want answered that this statement does not address?'

'Is there anything in this statement that contradicts publicly available information about this situation, or that could be shown to be misleading by reference to external sources?'

This AI-driven stress-testing replaces or supplements the communications team review in time-compressed situations. It will not catch everything a skilled communicator would catch, but it surfaces obvious problems quickly.

Before media appearances, regulator meetings, or all-staff sessions following a crisis, use AI to prepare for likely questions:

'Based on this situation [description] and our public statement [paste statement], generate the 20 most likely questions a hostile journalist would ask at a press conference.'

'Generate likely questions from the FCA [or relevant regulator] based on [situation description] and our current public position.'

'Generate the questions our employees are most likely to be asking that our communications have not yet addressed.'

Then use AI to help draft answers: 'Suggest a response to this question that is consistent with our public statement, does not admit to things not yet confirmed, and addresses the concern genuinely rather than deflecting.'

This preparation is valuable precisely because crisis situations leave little time for thorough preparation; AI compresses the time needed to cover the obvious ground.

Crisis situations create pressure to move fast, which creates pressure to cut corners on data security. Do not use consumer AI tools (free ChatGPT, consumer Claude.ai) to process confidential crisis information. The data handling terms are not appropriate for commercially sensitive material.

For crisis communications work, use: Microsoft 365 Copilot (if your organisation has it), Claude.ai Teams or Enterprise, or ChatGPT Enterprise. These have data handling terms suitable for confidential business information.

If none of these is available and you are using a consumer tool, work with anonymised or generic descriptions rather than specific confidential details. The communications principles AI helps with are applicable even without the specific facts.

Crisis situations are precisely when poor data security decisions are made because of time pressure. Brief the crisis team in advance on which AI tools are approved for crisis communications use, so there is no ambiguity in the moment.