Large Language Model
AI & SearchA type of AI trained on huge volumes of text that can read, write, and answer questions in plain language — the engine behind ChatGPT, Claude, Gemini, and most…
AI & Search
Hallucination
AI Hallucination | LLM Hallucination
Robert KlimantCo-founder
What is Hallucination?
An AI hallucination is a confident but factually wrong output from a large language model. The model generates fluent, plausible-sounding text that contains invented facts, fabricated citations, or details that simply do not exist. Hallucinations happen because LLMs predict likely word sequences, not verified truth. They are the single biggest reliability issue with modern AI tools and the reason every serious deployment includes retrieval, citations, and human review.
Why it matters
An AI hallucination in a customer-facing channel is a brand risk dressed up as a productivity win. A chatbot inventing a refund policy is a lawsuit. A sales agent making up a feature is a churned account. The fix is not to ban AI — that ship sailed — but to design around the limitation. Retrieval over generation. Citations on every claim. Human review for anything customer-facing. Teams that treat hallucination as a known engineering constraint ship reliable AI products. Teams that pretend the model just needs a better prompt ship demos that embarrass them in production.
How it works
LLMs do not look up answers. They predict the most likely next token based on patterns from training. When the model has seen a topic many times in training data, it is usually right. When it has seen it rarely, or when the context is ambiguous, it fills the gap with whatever sounds most likely — which is sometimes correct and sometimes invention. Modern systems reduce hallucination with retrieval-augmented generation, tool use, and prompts that tell the model to say 'I do not know' when uncertain. None of these eliminate the problem. They lower the rate from 'frequent' to 'occasional and catchable.'
Related terms
Retrieval-Augmented Generation
AI & SearchA technique that lets an AI model look up fresh, specific information from a database or the web before answering, so it does not have to rely only on what it…
AI Content Detection
AI & SearchTools that try to guess whether a piece of writing was generated by an AI model — used by publishers, schools, and clients, with accuracy that ranges from…
Prompt Engineering
AI & SearchThe practice of writing instructions to an AI model in a way that gets a reliable, useful result — part technical writing, part specification, part figuring…
Claude
AI & SearchAnthropic's AI assistant, known for careful reasoning, long context windows, and being a favorite among developers and writers who want a model that does not…
AI Citation
AI & SearchWhen an AI model credits your website as a source in its answer — the new equivalent of ranking on page one, and increasingly the most important metric to…