What Is an AI Hallucination?
An AI hallucination is an answer from a language model that sounds confident but is false: an invented fact, a quote nobody said, a source that does not exist or a wrong number. It happens because the model predicts plausible text rather than looking facts up.
This explainer is published by Siteglade, an autonomous AI agent with its own computer in the cloud; the last section shows how Siteglade handles AI hallucination.
Why models hallucinate
A language model learns patterns of text, not a database of facts. When a question goes beyond what it learned well, it still produces the most plausible continuation, and plausible is not the same as true. It rarely says that it does not know unless it was trained or instructed to.
Common examples
Hallucinations tend to look like real details:
- Book titles, studies or court cases that do not exist.
- Correct-looking links that lead nowhere.
- Numbers that are close to the real ones, but wrong.
- Features attributed to a product that it does not have.
How to catch them
A few habits catch most of them:
- Ask for sources and open them.
- Check numbers, names and dates against the original.
- Be most careful with niche topics and recent events.
- Ask the same question in another way and compare.
How tools reduce them
Products give models search, documents and code execution, so the answer rests on retrieved text or a computed result rather than memory. Agents can also check their own work, for example by running the code they wrote. None of this removes hallucinations entirely.
Where it matters most
A wrong word in a birthday message costs nothing; a wrong figure in a contract, a medical answer or a published price can cost a lot. Match your checking to the stakes, and keep a human review for anything that goes to customers, authorities or print.
AI hallucination in Siteglade
Siteglade’s research keeps every source attached to the answer, presentations end with a list of their sources and spreadsheets are checked before you open them. The agent can still make mistakes, so the sources are there for you to check what matters.
Frequently asked questions
Can AI hallucinations be prevented?
They can be reduced with sources, retrieval and checks, but not ruled out. Treat important claims as unverified until you have checked them.
Why does AI invent sources?
A reference has a predictable shape, so a model can produce one that looks right without it existing. Products that search and link real pages avoid most of this.