Have you ever asked a chatbot for a source, a fact, or a citation, and later found out it simply didn’t exist? That’s not an isolated bug — it’s a phenomenon with a specific name, and understanding how it works helps you use these tools more safely.
What it is, in detail
A hallucination is when a generative AI model produces false, made-up, or unverifiable information, but presents it with the same degree of confidence it would use for correct information. It could be a nonexistent bibliographic citation, a wrong date, a distorted historical fact, or even a link to a web page that doesn’t exist.
Why it happens
Language models don’t “know” things the way we do: they generate text by predicting, word by word, which sequence is statistically most plausible based on what they learned during training. When the requested information is rare, ambiguous, or absent from the training data, the model can still generate an answer that’s plausible in form but false in content, instead of simply saying “I don’t know”.
How to protect yourself
There’s no way to eliminate this risk entirely, but you can meaningfully reduce it:
- Always verify data, figures, citations, and references before using them in an important context.
- For topics where you need up-to-date, verifiable information, prefer tools that cite sources in real time (like Perplexity) over general-purpose chatbots.
- Be wary of overly detailed answers on very specific or niche topics: that’s exactly where the risk of hallucination is highest.
- Explicitly ask the model to flag when it’s not sure about a piece of information, even though this doesn’t fix the problem at its root.