Hallucination
When an AI states something false with full confidence: a made-up API, a token that does not exist, a citation that was never written. Fluent, plausible, and wrong.
What it is
A hallucination is when an AI produces something that sounds right but is not true. It invents a component prop that does not exist, references a design token you never defined, or cites a source that was never written. The output is fluent and confident, which is exactly what makes it dangerous.
Why it happens
An LLM predicts plausible text. It does not look facts up by default. When it does not know, it does not stop. It generates the most likely-sounding answer, and likely-sounding is not the same as correct.
Why this matters for designers
If you ask Claude to use your design system and it has not actually seen your tokens, it may confidently invent token names that look real. The fix is context and verification: give it the real tokens to work from, and check generated code against the source rather than trusting the names it produced.
How to reduce it
- Give the model the real source: tokens, docs, files, not a description.
- Ask it to cite where each value came from.
- Verify anything specific (names, props, values) against the actual system.