A hallucination occurs when a generative AI model generates incorrect, invented, or unsubstantiated information, but presents it with a confident and plausible tone. It is not a one-off calculation error, but rather an inherent limitation of how language models work: they predict the most likely text based on learned patterns, without verifying whether what they state is true.
This matters because the fluency of the output can lead to trusting false data. Some common examples are:
- Non-existent citations or references, with invented authors and titles.
- Erroneous numerical data or dates presented as facts.
- False attributions, such as quotes that a real person never said.
The risk increases in sensitive contexts such as medicine, law, or journalism, where an invented piece of data can have serious consequences. Therefore, it is advisable to always verify critical information with reliable sources and treat AI responses as a starting point, not as a confirmed truth.