For decades, companies in crisis relied on a simple truth: the news cycle moves on. Keep your head down, wait out the storm and in a few weeks the headlines fade. According to South Africa’s top ranked PR agency PR Worx, artificial intelligence (AI) has quietly killed that strategy.
“AI doesn’t have a news cycle, it has a memory,” says Madelain Roscher, CEO of PR Worx. “When a customer, investor or journalist asks an AI platform about your company, it draws on everything ever published about you. If the last credible word on your crisis is the accusation, then the accusation becomes the answer. Permanently.”
The risk is compounded by how AI systems handle information. A landmark study by the European Broadcasting Union and the BBC, which evaluated more than 3,000 responses from leading AI assistants, found that 45% of answers contained at least one significant issue, including outdated information presented as current fact. For a company that weathered a crisis two years ago, that can mean AI still telling the story as if it were unfolding today, with no mention of the resolution, the court outcome or the corrective action.
“We’ve seen businesses do everything right during a crisis, fix the problem, compensate customers, change their processes and then go silent,” says Roscher. “The trouble is that machines can’t read what you’ve never published. If your response only lives in boardroom minutes or a legal settlement, as far as AI is concerned, it never happened.”
With consumer use of generative AI surging from 45% to 73% in two years, according to Prophet’s 2026 research, Roscher says the outcome of a reputational crisis is increasingly determined not in the week the story breaks, but in the record that exists long after it.
She offers four principles for managing a crisis in the AI era:
Respond on the record, not just behind closed doors.
Apologies, corrections and remedies must appear on credible, indexed platforms: earned media, official statements and recognised industry publications. AI weighs independent sources most heavily, so a resolution reported by a journalist counts for more than one claimed on your own website.
Build the counter-record deliberately.
One statement does not outweigh fifty negative headlines. Companies emerging from a crisis need a sustained programme of credible coverage, executive interviews, third party commentary and documented outcomes, so the balance of evidence changes over time.
Close the loop publicly.
When the investigation concludes, the recall completes or the court rules, say so where it can be found. The single most common AI distortion in a crisis context is an old story with no ending.
Monitor what the machines are saying.
Ask the major AI platforms what they say about your company, your executives and your crisis, and treat the answers as a reputational audit. If the response is outdated, incomplete or wrong, that is the gap your communications programme must close.
“Reputation management used to be about surviving the storm,” says Roscher. “Now it’s about what the machines remember once the storm has passed. You cannot delete a crisis from AI’s memory, but you can determine whether it’s remembered as the scandal, or as the recovery.
“The companies that stay silent are letting machine learning write the final chapter of their story. The smart ones are picking up the pen.”