Hallucination
Updated June 10, 2026
When an AI model states false information with perfect fluency — the failure mode that no amount of style fixing touches.
Definition
A hallucination is model output that's confidently wrong: invented citations, fabricated statistics, plausible-sounding events that never happened. It's not a bug being patched out so much as the nature of the machine: an LLM generates the most plausible next text, and plausible and true are different properties. When the likeliest continuation is a reference, the model produces one — existence not required.
Where it bites
Citations are the classic catastrophe: fabricated references in legal filings have drawn sanctions, and a hallucinated source in an essay is the one AI tell that gets checked and ends badly — far stronger evidence than any detector score. Statistics, quotes and dates fail the same way: fluently.
Why humanizers don't (and shouldn't) touch this
Humanizing changes how text reads, not whether it's true — a rewrite that "fixed" your facts would be a liability, which is why Humanize Studio preserves numbers, names and quotations verbatim and flags potential drift instead. Verification of facts stays a human job: check every citation and number AI gives you, every time. Style tools polish prose; nothing polishes a fabricated source.