What is an AI hallucination?
A hallucination is an answer a language model produces that is fluent, confident and false — invented rather than drawn from any source.
Language models predict plausible text, not true text. Asked a question with no supporting material, a model does not fall silent; it produces the most likely-sounding answer, which is often a wrong price, a policy that does not exist, or a citation to a page that was never written. The behaviour is not a bug to be patched out of the model. It is what the model does when nothing constrains it.
Why it matters
For a business, a hallucinated answer is worse than no answer. A customer told the wrong return window will act on it, and the business is then in the position of contradicting its own website. The fix is not a better model — it is a system that retrieves real passages first and is built to refuse when retrieval comes back weak.
Examples
- Asked about a discount code that does not exist, an ungrounded bot invents one and a customer tries to use it at checkout.
- Asked which integrations a product supports, a model lists three competitors' integrations because they were common in its training data.
- Asked for a phone number, a model produces a plausible-looking one with the right area code and the wrong digits.
Not the same as
Retrieval failure
Retrieval failure means the right passage was not found. Hallucination is what happens next if the system answers anyway instead of admitting the gap.
Outdated answer
An outdated answer was true when the page was written and is quoted correctly. A hallucination was never true anywhere.
Related
term
What is grounding in AI?
A grounded system retrieves passages before it writes anything, puts them in front of the model, and instructs it to answer only from what it was given. The test of whether a system is grounded is simple and does not require access to its internals: ask it something, then ask it where the answer came from. A grounded system names a page. An ungrounded one describes its reasoning.
term
What is RAG (retrieval-augmented generation)?
RAG is what separates a chatbot that knows your business from one that guesses. Documents are split into passages and indexed by meaning; each question retrieves the closest passages, and the model is instructed to answer only from them.
term
What is a confidence score in AI answers?
Two very different things get called confidence. Retrieval confidence measures how close the best-matching passages were to the question — a property of the search step, computed before any answer is written. Model confidence is the model's own assessment of its answer, which it will happily report as high while inventing a price.
term
What is a citation in an AI answer?
A cited answer tells the reader not just what the answer is but where it came from. In a customer-facing system, that link is usually a page on the business's own website, so following it is both a verification step and a way further into the site.
tool
Website support score
Scores a website on how well it could support customers — and how well an AI representative trained on it would perform. Five dimensions, a total out of 100, and the specific gaps behind the score.
term
What is grounding in AI?
A grounded system retrieves passages before it writes anything, puts them in front of the model, and instructs it to answer only from what it was given. The test of whether a system is grounded is simple and does not require access to its internals: ask it something, then ask it where the answer came from. A grounded system names a page. An ungrounded one describes its reasoning.
RubyRep is this, applied to your website.
Give it your URL and it reads your pages, then answers your customers from them.
Train My RubyRep