Stop AI from wasting time in your next crisis

Stephanie Nivinskus is principal at Ragan’s Center for AI Strategy.

Last month, hackers used stolen passwords to break into Chick-fil-A’s customer loyalty accounts, potentially exposing sensitive customer information like names, email addresses, phone numbers and payment information.

If your company stores customer data, it’s just one password away from needing to send a breach notice and face a reporter’s questions just like Chick-fil-A. AI can’t help you if your carefully tailored response plan is invisible to it.

Your organization likely has a message map for crisis events like the one Atlanta-based Chick-fil-A faced. But if you haven’t shared it with your AI tool, it might as well not exist. Without that documentation and context, your AI tool will default to the generic answer, relying on learned patterns rather than the decisions your organization has already made.

Four inputs stop the loop

These four inputs help AI draft usable material when the next incident begins:

1. Log rejected crisis language and why it was turned down. Any framing, position or talking point legal or leadership rejected during a past incident belongs in the record, not just one example recalled from memory.
2. Attach an expiration date or review trigger to approvals tied to a live situation. Guidance approved for a specific breach, lawsuit or spokesperson rarely stays accurate once circumstances shift, even though nothing prompts anyone to revisit it.
3. Name the person who authorized any exception to standing guidance. An override used once under pressure, with no owner attached, can be mistaken for new house policy the next time a crisis hits.
4. Tag crisis questions your organization has discussed but never resolved. An unresolved question is not the same as a rejected one. It was neither rejected nor approved.

Make the log usable under pressure

Store the record somewhere your comms team and your approved AI tool can access it during an incident. Assign one owner to update it after every crisis exercise or live event, and include the date, decision, rationale, approver and conditions under which the guidance applies.

A folder full of old documents will not solve the problem. The value comes from giving AI a current, structured record it can retrieve before drafting begins.

Use the log to check the message map before it goes public

The same record that stops AI from resurfacing rejected language also lets it review a message map before an executive faces reporters. Feed it the log, the current message map and a list of likely reporter questions. It can flag any answer that contradicts a rejected position or relies on an expired approval.

That step takes minutes and can expose contradictions before they reach a spokesperson.

A crisis does not wait while your team reconstructs decisions that live only in someone’s memory. Build the log before the next incident, and help your AI tool stop burning minutes you do not have by relearning what your organization already knows,