Crisis communications teams use AI tools under time pressure that most other PR work does not have, which means the tool that helps most is not the fastest one, it is the one that hedges carefully and holds the most context about what has already been said. This guide covers where AI tools genuinely help during an active crisis, and the specific workflow discipline that keeps speed from turning into a mistake.
What Makes Crisis Communications Different for AI Tool Use?
Two pressures shape AI use during a crisis that do not apply to routine PR work. The first is speed: a holding statement is often needed within the hour, which tempts a team to ship AI output with less review than it would get on a normal day, exactly when the stakes of an error are highest. The second is context accumulation: a fast-moving crisis generates a growing pile of prior statements, reporter questions, and internal decisions, and a new statement that contradicts something said three hours earlier compounds the crisis rather than resolving it.
Where Do AI Tools Genuinely Help During a Crisis?
Four tasks come up repeatedly once a crisis is active.
Drafting a Holding Statement in Multiple Registers at Once
Asking for a cautious, a direct, and an accountable version of the same statement surfaces the tradeoffs faster than a team debating tone from scratch, and Claude's ability to hold the full incident timeline in context keeps each version consistent with what has already been confirmed publicly. Starting from a pre-approved structure helps, and these fill-in crisis statement templates give a team the fixed elements to hand the model before it drafts.
Checking a New Statement Against Everything Said So Far
Pasting in every prior statement and asking whether a new draft contradicts any of them catches an inconsistency before it becomes the next story, which matters more here than in any other PR workflow.
Monitoring How the Story Is Developing in Real Time
Perplexity's own web index responds faster to breaking coverage than a cached search index, making it the strongest tool for same-day tracking of how a crisis is spreading across outlets.
What Should Never Happen During a Crisis, AI Tool or Not?
No AI-drafted statement ships without a human reading it against the actual facts of the moment, since a fast-moving crisis is exactly when a model's fluent, confident output is most likely to state something as settled that is still being confirmed internally. Legal review before publication is not optional during a crisis with any liability exposure, and that review step should not be skipped to save the few minutes an AI tool saved on the draft. Material that has not been cleared for public release, an internal cause-of-incident theory, an unconfirmed casualty count, a legal exposure assessment, should not go into any AI tool at all, since the speed of a leak matches the speed of the crisis itself.
What Is the Fastest Safe Workflow During an Active Crisis?
Step
Tool
Why
1. Set up a dedicated Claude Project for the incident
Claude
Holds the timeline, prior statements, and counsel notes in one place for every subsequent draft
2. Draft the holding statement in three registers
Claude
Surfaces the tone tradeoff fast, using the full incident context already loaded
3. Check the draft against every prior statement
Claude
Catches contradictions before they compound the story
4. Prepare hostile-question rehearsal
Claude or ChatGPT
Compresses spokesperson prep time
5. Monitor how coverage is spreading
Perplexity
Fastest sourced tracking of a breaking story across outlets
Legal and, where relevant, executive sign-off happens after step 3 and before anything goes external, regardless of how much time step 5 shows the story has already spread.
What Should a Crisis-Specific AI Policy Add to the General Template?
Beyond the standard confidentiality rules any PR team should follow, a crisis-specific addition names the exact tool and account tier pre-approved for crisis use before a crisis happens, since deciding which tool to trust in the middle of an active incident wastes the time a crisis does not have. It also names who has authority to approve an AI-drafted statement for release under time pressure, so that authority is not being figured out for the first time during the crisis itself.
What Are the Best AI Prompts for an Active Crisis?
Two examples show the pattern used most often once a crisis begins.
Holding statement in three registers.
"Draft a holding statement for [company] responding to [situation in one paragraph]. We can confirm: [list]. We cannot confirm: [list]. Produce three versions: cautious, direct, and accountable. For each, flag the sentence a hostile reporter would quote against us."
Consistency check against prior statements.
"Here is every statement we have made publicly about this situation so far, in order: [paste all]. Here is a new draft: [paste]. Flag any place the new draft contradicts or adds unconfirmed information beyond what we have already said."
Which AI tool is best for drafting a crisis statement?
Claude, because it can hold the full incident timeline, prior statements, and counsel notes in context at once, which produces a more internally consistent statement than starting a fresh conversation each time.
Should a team skip legal review to move faster during a crisis?
No. Legal review before publication matters more during a crisis, not less, since the stakes of an error are highest exactly when the pressure to move fast is greatest.
Can unconfirmed internal information go into an AI tool during a crisis?
No. An unconfirmed cause, casualty count, or liability assessment should not go into any AI tool, since a leak during a crisis spreads exactly as fast as the crisis itself.
How should a team prepare for AI use in a crisis before one happens?
Pre-approve the specific tool and account tier for crisis use, and name who has authority to approve an AI-drafted statement under time pressure, so neither decision is being made for the first time during an actual incident.
What is the single biggest AI risk during a crisis?
Shipping a fluent, confident AI draft with less scrutiny than normal because of time pressure, at the exact moment an error is most costly.
Written by
Sarel Doglu
Sarel Doglu is a technology and product executive with more than 25 years of experience building software and leading technology teams. He is a technology and marketing expert.
Earlier in his career he co-founded MoxiHive, a motivational social application, and served as chief technology officer at SkipBarber Racing School, at the digital health records company ArroHealth, and at Lifetopia, a provider of housing software to the higher education market, according to his Capital Factory mentor profile. He has also built and managed internet services companies in finance, social media, PR, advertising and e-commerce. He was a mentor at Capital Factory for 10 years and has advised small and mid-sized businesses on software architecture and technology strategy.
