Synthetic crises: When AI turns attitude into reputational damage

Generative AI is quietly and systematically changing how corporate reputation is created. Not through scandals, not through mistakes in crisis communication - but through a mechanism that has hardly been on the radar of communication and HR managers: AI models misinterpret brand content, link it to the wrong contexts and produce narratives that no one has authorized. The result can be described as synthetic crisis communication.

The new terrain: reputation is calculated

People searching for a company today are increasingly no longer using a search engine, but an AI. Applicants ask for the «truth» behind the career page. Buyers ask about known problems with a product. Journalists ask for critical voices about a company. In all these moments, it is no longer just what a company has published about itself that is decisive, but how a language model statistically interprets and weights this content.

This is a fundamental shift. Traditional reputation management strategies rely on visibility, sentiment and reach. AI models operate according to a different logic: vectors, semantic proximity, accumulation effects. What is written on a website is located mathematically.

How synthetic crises arise

AIs work with semantic proximity. Terms that appear in the same context are statistically linked to each other - regardless of their relationship to each other in terms of content. The nuance between «we fight X» and «we have X» is not a semantic category for a language model. What is relevant is the co-presence of the terms.

This results in a paradoxical risk: the more actively a company communicates, the more it takes a stance and identifies problem areas, the more material it provides for erroneous links.

An example from the field of sustainability: A consumer goods manufacturer publishes extensive sustainability reports over the years, openly identifying challenges in the supply chain and formulating ambitious reduction targets. An AI that is asked about «problems» with this company accesses precisely these reports - and lists the challenges as the current situation without taking the target dimension into account. Transparency becomes a target.

An example from the pharmaceutical sector: a pharmaceutical company proactively communicates about the side effects of a drug - in detail, because it is required by regulations and is ethically correct. Anyone who asks an AI about this drug may receive a summary that makes the communication of side effects appear to be a central brand feature of the product - not its effectiveness.

An example from the HR sector: a medium-sized company publicly advocates mental health in the workplace, talks about toxic patterns in the industry in podcasts and positions itself as a counter-design. Applicants who ask an AI for employer ratings run the risk of receiving a summary that links the company to the very terms it is competing against.

The pattern is identical in all three cases: content that was intended for positioning becomes the basis for a false attribution.

Why classic crisis management does not work here

Traditional crisis communication reacts to events: a product recall, an interview that goes wrong, a viral post. The antidote is clarification, context, chronology. All of this assumes that there is a public that listens and evaluates.

Synthetic crises arise without an event. There is no headline that can be reacted to. There is no actor who is wrong and can be corrected. The faulty link is created in secret, in billions of inferences every day, and it only becomes visible when a talent cancels, a lead doesn't call back, a journalist goes into an interview with a strange assumption. This makes this form of reputational risk structurally more difficult to grasp than any classic crisis.

The bitter irony of commitment

There is a specific group of companies that is particularly exposed: those that consistently take a stand. NGO-affiliated brands that write about social grievances. Purpose-driven employers who name industry problems. Companies that address customer pain points in their communications in order to position themselves as a solution. All these companies do exactly what good communication should do: They talk about real problems. And that is precisely why they accumulate semantic links that potentially turn AI models against them.

The bitter irony is that those who stand up for the good the most and name grievances the most clearly have the biggest target in a world that calculates rather than reads context.

What communication professionals can do now

The obvious reaction - formulate critical issues more softly in future, avoid problem areas, only communicate positive things - would be wrong. If you write for algorithms instead of people, you lose credibility. And if you stop taking a stance, you give up what makes brands different in the age of AI intermediation.

The answer lies elsewhere: in actively populating the digital space with content that creates positive links - and in a communication architecture that makes it easier for AI models to interpret.

A clear sender logic is key here: criticism of industry standards must be structurally decoupled from the brand itself and at the same time be recognizable as its counter-proposal. This requires a consistent message architecture that appears not just once in a campaign, but as a common thread running through all formats. In addition to this, it is important to create so-called love content as a counterbalance: User-generated content - ratings, employee statements, customer reports - generates data sets that give a positive weighting to AI models. A static «About us» page cannot achieve this.

Generative Engine Optimization (GEO) must also be anchored as a strategic discipline. Content must be structured in such a way - with clear facts, verifiable statements, unambiguous attribution - that AI models can interpret it correctly. This means less mass, more precision. Finally, Schwellnus recommends regular prompt monitoring: communication managers should test on a quarterly basis how AI models respond to critical questions about their own brand - not to become paranoid, but to recognize early on which narratives are emerging and to take countermeasures before they become entrenched.

Conclusion

Crisis communication is entering new territory. It is no longer just about reacting to events, but about shaping the data basis on which AI models make judgments. Companies that understand this and align their communication strategy accordingly have a structural advantage. Those who continue to play by the old rules leave the definition of their brand to an algorithm that does not know context - and does not understand attitude.

About the author: Miriam Schwellnus is the founder and managing director of the Berlin agency Mashup Communications. As the author of the bestseller «Storytelling for Companies», she has been combining modern public relations with sound content marketing and narrative strategies for almost 20 years. Her goal: to help organizations gain more visibility, trust and relevance - especially in the age of artificial intelligence and Generative Engine Optimization (GEO).

More information: https://www.mashup-communications.de