Imagine you run communications at a fintech and your chatbot replies "I'm glad" to a customer asking for urgent banking support. That single message can trigger a full chatbot PR crisis, and the way you respond in the first hours decides whether you protect the brand or lose customer trust for good.

Here you'll walk through the five steps I'd follow to contain a crisis like this, the narrative choices that matter, and the KPIs that tell you if your recovery plan is actually working.

What is the first thing you should do during a chatbot crisis?

Before you draft anything, you need a fast read of the situation. A rushed statement without context can make the fire bigger.

How do you run a quick diagnosis?

Start by asking three blunt questions: what kind of crisis is this, what technical failure caused it, and how far has it spread already. The goal is to know if you still have time to act before the story explodes on social media or in the press.

This is where you separate a contained incident from a reputational emergency. If only a handful of users saw the message, your playbook looks very different from a case where screenshots are already going viral.

What is a chatbot PR crisis? It's a reputational incident triggered when an automated bot gives an inappropriate, insensitive or wrong answer to a user, damaging trust in the brand and forcing the company to respond publicly.

How do you contain the crisis in the first hours?

Once you understand the scope, you move into containment. Speed matters, but coordination matters more.

Why do you need a war room and clear spokespeople?

Activate the war room immediately and define who speaks on behalf of the company. You need designated voceros, a first narrative drafted, and a clear list of official channels you'll use.

If the crisis had come from a hack, your official channels might be compromised and you'd need alternative ones. In a chatbot failure, your channels are still available, so use them, but adapt the tone to each platform. The format changes between LinkedIn, X or a press release; the core message does not.

How should the official statement sound?

The public message is where most companies fail. They sound corporate, defensive, or cold, and that only fuels the crisis.

What makes a crisis statement work?

A strong communiqué is empathetic, transparent and owns the mistake. It also offers a concrete fix, like opening a human communication channel so affected users can talk to a real person instead of the bot.

List the measures you're taking and commit to follow up. The audience needs to feel that the company understands the harm and is doing something about it, not just issuing legal language.

• Acknowledge the error openly.
• Explain what failed in plain words.
• Offer a human alternative to the chatbot.
• Promise an update within a clear timeframe.

After the statement goes out, monitor whether the measures are actually reducing the noise or if you need to escalate.

What does a recovery plan look like after a chatbot failure?

The first 48 hours are about containment. The next phase is rebuilding trust, and that takes a structured plan.

How do you plan the first seven days?

Map the actions for the first week and decide the narrative you'll hold throughout. Track how user interactions are evolving and document everything, because you'll need it for the post mortem.

You should also commission an external audit of what happened and a technical review of the chatbot itself. Find out exactly what failed in the logic or the training data, and what you can adjust so the same answer never appears again.

Why do you need an external audit after a crisis? Because internal teams have blind spots. An external review brings credibility to your findings and shows stakeholders, regulators and customers that the investigation was independent.

Which KPIs tell you the crisis is under control?

Numbers turn perception into evidence. Without them, you only have opinions about whether the response worked.

What metrics should you track?

Look at whether negative comments are dropping, whether the human phone line you opened is resolving more cases, and whether the chatbot, if you kept it live, is now performing better.

• Volume of negative mentions before and after the statement.
• Resolution rate through the new human channel.
• Number of cases closed per day.
• Chatbot accuracy on sensitive queries.

These indicators show if the recovery is real or just cosmetic.

Why does AI crisis management go beyond a good prompt?

Artificial intelligence can help you draft, simulate and analyze, but only when you use it ethically and double check every output. The technology accelerates work; it doesn't replace judgment.

Managing a crisis is about seeing the full picture, deciding with empathy and rebuilding the relationship of trust that the incident put at risk. A clever prompt won't save you if you don't understand the human stakes behind the screen.

In the next module you'll dive into the crisis protocol itself: the actors involved, the structure of the war room, and the exact moment when it should be activated. Tell me in the comments how you would have handled the "I'm glad" reply if it had landed in your inbox.