AI Search

What Is Hallucination (AI)?

Definition

When a large language model generates confident-sounding but factually false or fabricated information — a known and unsolved limitation of current AI systems.

Why It Matters

AI hallucinations matter for marketers because AI systems occasionally generate incorrect information about brands, products, or facts and present it confidently. When a business is misrepresented in AI responses, there's limited recourse beyond publishing accurate, crawlable, authoritative content that the AI can retrieve and prefer.

How It Works

Hallucinations happen because LLMs predict the most probable next token based on patterns in training data — they don't verify truth. When the model encounters a query without strong supporting patterns, it may generate plausible-sounding but invented details. Retrieval-augmented generation reduces (but doesn't eliminate) hallucinations by grounding answers in retrieved sources.

Real-World Example

A user asks ChatGPT about an Australian business, and ChatGPT fabricates a non-existent phone number because the real number was not in training data. Publishing clear, authoritative contact information on the business website — and ensuring AI crawlers can access it — helps future retrieval-based AI responses get the answer right.

Quick Facts

Hallucination rates have declined from ~20% in early LLMs to 2–5% in current models
Retrieval-augmented generation (RAG) cuts hallucinations substantially
Hallucinations are most common on niche, recent, or local information
No LLM is hallucination-free — output verification is still essential for high-stakes use