Real-Time Crisis Communication: Lessons from 15 Years of Social

My first shitstorm caught me on an evening when all I wanted to do was approve comments. I was an intern doing community management for GZSZ.de, the website of a big German TV soap. A storyline had landed badly with the audience, and suddenly comments were coming in faster than I could read them. There was no playbook, no escalation matrix, no monitoring tool. There was me, a keyboard, and a very angry audience. I learned more that night than in most seminars since.

That internship turned into almost seven years at Mediengruppe RTL, where I eventually became Germany's first official social media manager, responsible for more than 100 profiles and over 20 million fans. At that scale, something is always burning somewhere. Crisis communication was never a special event for me. It was a craft. In this article I want to share what has held up over 15 years, and where AI is changing that craft today.

Preparation: the plan is written before the storm

From 2017 to 2019 I ran digital communications at Phantasialand, a German theme park with around two million visitors a year. Every guest carries a smartphone. Every queue, every ride outage, every rain shower can be online within minutes. If you only start thinking about who is allowed to reply once it happens, you have already lost. Preparation means knowing your likely scenarios, having first responses drafted in advance, and knowing who signs off when things get serious.

Three things belong in every playbook. First, an honest list of your own weak spots, because most crises are foreseeable. Second, text modules for the first hour, written calmly in advance instead of with shaking hands. Third, clear responsibilities with backups, because crises do not respect office hours. It sounds trivial. It is still missing in a surprising number of companies.

Escalation levels: not every storm is one

The most common mistake I have seen in 15 years: companies treat three angry comments like the end of the world, and a genuine wildfire like a routine community topic. Both go wrong. That is why, since my RTL years, I have worked with clear levels that determine the response and who owns it.

What matters most is the transition between levels. Sleep through level 2 and you will face level 3 unprepared. Send your CEO out at level 1 and you have nothing left for level 4. The levels are not a form to fill in. They are an agreement that gives everyone involved something rare in a crisis: calm.

• Level 1: individual critical comments. Business as usual. The community team replies calmly and personally, no alarm bells.
• Level 2: a topic gains density, comments start referencing each other. Inform the team, agree on wording, watch the frequency.
• Level 3: outside reach arrives, large accounts or first media inquiries. Crisis mode: one voice, one channel, leadership involved.
• Level 4: traditional media report, the business is affected. Now top management speaks, visibly and in person.

Tone: how you sound when things are burning

At RTL, working with Vidpresso, I produced the first professional Facebook livestream of a Klitschko boxing match. Live means no edit, no second chance, no delete button. That is exactly how you should think about crisis communication, because everything you write becomes a screenshot immediately. So write like a human, stay concrete, and name mistakes when there were mistakes. Legal boilerplate in a comment section is like wearing a suit at the pool. It always looks wrong.

My Rhineland ground rule: stay calm, be honest, and do not make the issue smaller than it is. Apologize when an apology is deserved, and without the classic 'if anyone felt offended'. And do not feed the trolls. Some comments do not want an answer, they want a stage. After a short time in community management, you learn to tell the difference very reliably.

Where AI genuinely helps today

Today, as an AI consultant at BE BRAVE AG and co-founder of ORION AI, I build exactly the tools I wished for back then, staring at a screen at night. More than 30 AI and digital products are live in production, including live AI translation for streams and voice agents on the phone. For a crisis, this means AI takes over monitoring, notices when a topic starts to condense, and clusters thousands of comments by content and sentiment while humans are still reading the first ten.

AI also helps enormously with drafts. A first version of a response, variants in different tones, translations in minutes instead of hours: all valuable when the clock is running. Infrastructure matters here. At BE BRAVE we work on sovereign Swiss AI infrastructure, because a crisis is exactly the moment when sensitive internal information should not travel through foreign clouds. If you want to use AI when it counts, settle that question beforehand, not in the middle of it.

Where only humans are credible

And now the line I consider the most important one: AI may prepare, but it must not take responsibility. An apology that visibly comes from a machine makes the crisis worse, not better. Deciding whether to trigger the next level, whether to admit the mistake, whether the CEO steps in front of the camera: these are questions of judgment, not text generation tasks.

As CDO of a Swiss livestreaming scale-up, I worked on large productions where, live on air, things occasionally went wrong. In those moments, nobody wants to hear a bot. People forgive mistakes with surprising generosity when a human explains them and stands behind them. What they do not forgive is simulated concern. My rule of thumb: AI writes the draft, a human decides, signs, and carries the consequences.

Conclusion: write your playbook before you need it

Real-time crisis communication is not a talent, it is preparation plus attitude. Define your escalation levels, write your first responses in calm times, decide who speaks and who approves. Let AI do what it does best: watch, sort, and deliver drafts. And keep the moments where credibility is built firmly in human hands.

My question for you: if a topic boils over tonight at 10 pm, do you know who replies for your company, and would that person know what they are allowed to write? If you hesitate on that, let's talk. Building exactly these playbooks together with teams is some of my favorite work, and you will find my contact details right here on this page.