Hallucination
When an AI model generates factually incorrect information confidently. Monitoring hallucinations is critical for brand safety.
What is Hallucination?
A hallucination is output from a language model that is fluent and confident but wrong: a product feature that does not exist, a price that was never charged, a founder who never worked there, a quote nobody said. It happens because the model generates the most plausible continuation of text rather than looking facts up, so gaps in its knowledge are filled with statistically likely but unverified detail.
Retrieval reduces hallucination but does not eliminate it. When an engine fetches live pages before answering, it has real material to ground the answer in, which is why search-enabled modes are generally more accurate than plain chat. But the model can still misread a source, merge two sources, cite a page that says something different, or fall back on training memory when retrieval returns nothing useful. Ambiguous, outdated or inconsistent information on the web makes all of these more likely.
For brands, hallucinations show up as wrong pricing, invented integrations, confused product names, outdated leadership and misattributed competitor features. They are usually discovered by running realistic prompts across engines and reading the answers, then tracing the error back to whatever source (or absence of a source) caused it.
Why it matters for AI search
An assistant that confidently misstates what a product does or costs is talking to buyers on the brand's behalf. The most reliable defence is to leave the engines nothing to guess about: clear, current, consistent facts on the brand's own site and on the third-party sources engines retrieve from. Monitoring AI answers for a fixed prompt set catches new errors early and shows whether corrections to source content are actually being picked up.
Related terms
Source Attestation
The process by which an AI model verifies the authority and truthfulness of a source before citing it in an answer.
RAG (Retrieval-Augmented Generation)
A technique that allows LLMs to pull in real-time information from external sources (like a website) before generating an answer.
Knowledge Graph
A network of entities and their relationships that AI models use to provide structured, factual answers.
LLM (Large Language Model)
A type of AI trained on vast amounts of text data to understand and generate human-like language (e.g., GPT-4, Claude 3).
Frequently asked questions
Why do AI models hallucinate?+
They generate text by predicting likely next tokens, not by consulting a database of facts. When the model lacks reliable information it still produces a plausible-sounding answer. Retrieval, better grounding and clearer sources reduce the rate but do not remove it.
How can a brand reduce hallucinations about itself?+
Publish accurate, specific facts on crawlable pages (pricing, features, integrations, leadership), keep them consistent across your site, directories and profiles, use structured data, and update or remove stale content. Then monitor answers across engines to confirm the fix landed.