What are AI hallucinations?
An AI hallucination is an output that seems plausible but is factually incorrect, invented or unsupported by the available information. For example, a language model may cite sources or product features that do not exist. Convincing wording is therefore no proof of accuracy.
AN EXAMPLE IN PRACTICE
Where you might encounter it
An assistant names a purported product feature. That feature is not included in the approved product data. Checking against the actual information prevents the draft from being passed on to a customer unchanged.
Why does this matter for your business?
By planning for this behaviour, you can add checks at the right points. Uncertainty and missing information should be made visible rather than concealed with invented details.
What measures help in day-to-day business?
RAG can provide relevant sources. Clear prompts set boundaries. Human review remains useful for important results. No single measure eliminates all errors.
Frequently asked questions
Can I spot a false answer from its writing style?
Not reliably. A false answer can also sound fluent and confident. Check the claims and the sources actually available.
Are citations always proof?
No. A citation can be invented or may not support a claim. A good application links to the sources actually used and lets you review them.
What to keep in mind
- Check important claims against sources that can be reviewed.
- Do not let missing figures or product features be filled in.
- Test with unclear questions and questions that cannot be answered.
Further sources
From NetzTurm Decode. Clear explanations for your business.