AI hallucinations

AI hallucinations are no longer a theoretical niche issue, but a real risk for the media, businesses and organisations. It has become clear, not least following several high-profile AI scandals, that when AI hallucinates, it can cause massive damage to trust, credibility and reputation.

What are AI hallucinations?

AI hallucinations are false, fabricated or misleading content generated by AI models because they are based on probabilities rather than fact-checking.

A recent example from the media: The ZDF AI scandal

In February 2026, ZDF came under heavy criticism after AI-generated video footage was broadcast on heute journal as real news footage. The video purportedly showed scenes of deportations from the USA – in reality, it originated from an AI video tool (Sora) and was not authentic. The AI-generated nature of the footage was even recognisable by a visible watermark, yet it was not properly verified or labelled by the editorial team.

News programmes enjoy a presumption of trust, particularly public service broadcasts. If AI material is used without labelling, this trust is automatically transferred to something artificially generated. Media experts warn that this blurs the line between documented reality and simulation.

What do AI hallucinations look like?

True AI hallucinations occur when an AI system generates content that is objectively false, fabricated or logically impossible, even though it appears realistic, without any such error having been explicitly specified.

Typical hallucinations include:

• an AI invents scientific studies, statistics or sources
• an AI cites incorrect historical events or dates
• a language model explains relationships that are physically or logically incorrect
• an AI makes legal or medical statements for which there is no legal or factual basis
• AI-generated images featuring people with too many fingers, clocks with the wrong number of hands, characters, logos or numbers that look like language but make no sense, or physically impossible representations of light, shadow or perspective
• etc.

It’s not just language AI models that hallucinate; image and video models do too.

How does an AI hallucinate?

To understand why an AI hallucinates, you need to know how language models work. They analyse large amounts of text and calculate which word is most likely to follow the previous one, rather than whether the statement is true.

Why do AI hallucinations occur?

The causes of AI hallucinations are structural in nature and can be easily explained. Hallucinations occur particularly frequently when AI generates summaries or invents sources that have never existed.

👉 The more open-ended your question, the higher the risk of hallucinations.

AI tends to hallucinate particularly frequently when:

• studies, statistics or sources need to be consulted
• current events need to be analysed
• specialist knowledge beyond the scope of the training is required

AI is constantly improving in terms of quality. However, you should still always check your sources before using any data.

How can you reduce AI hallucinations?

AI without hallucinations does not (yet) exist. However, you can significantly reduce the risk.

Best practices for reducing AI hallucinations

• Use precise prompts
• Explicitly request facts and sources
• Use uncertainty queries
• Verify results manually

Particularly when it comes to YMYL topics (money, health, law), human review is mandatory.

What does this mean for businesses and SEO?

For businesses, agencies and marketing teams, AI hallucinations are not a technical detail, but a real business and reputational risk. Particularly in the SEO landscape, where content is increasingly AI-assisted or entirely AI-generated, hallucinated information can have a direct negative impact on visibility, trust and rankings. Search engines evaluate content not only based on keywords, but increasingly on reliability, factual accuracy and trustworthiness. If AI-generated texts:

• contain false facts
• invent sources
• provide outdated or contradictory statements

this can lead to ranking losses in the long term, even if the content is linguistically high-quality.

Why AI hallucinations remain a persistent issue

AI hallucinations are an inherent characteristic of modern AI models; they are neither a coincidence nor a fault in the strict sense. If you understand what AI hallucinations are and why an AI hallucinates, you can use AI in a targeted, safe and efficient manner.

Frequently asked questions about AI hallucinations

What does hallucinating mean in the context of AI?

It describes the generation of false but plausible content without the AI recognising its own uncertainty.

Why do AI models hallucinate so convincingly?

Because they are optimised for language quality rather than truth.

Which AI has the fewest hallucinations?

No AI is free from hallucinations – systems with source integration reduce the risk, but do not eliminate it.