Which one would you send?
Two messages — one AI-fabricated and confident, one hand-written and accurate. Pick the one you'd send to a client. Reveal: which one was wrong, why, and the real-world receipt of someone who fell for it. Multi-round. Score persists.
Both messages were drafted to send to a client. One was written by an AI — fluent and confident, but it invented a detail. The other is accurate. Click the one you'd actually send.
That's the drill.
What is the "Eloquence Trap", and why do I keep falling for it?
Fluent and wrong looks exactly like fluent and right.
A generative AI is a language model — it is optimised to produce text that reads well, not text that is true. When it doesn't know something, it doesn't go quiet or hedge. It produces a clean, specific, confident sentence anyway. That invented-but-plausible output is called a hallucination.
Why the confidence is the dangerous part.
We use fluency as a shortcut for competence. A message that's well-structured, uses the right jargon, and states a precise figure feels trustworthy. The AI gives you all three for free — even when the figure is made up. The Eloquence Trap is mistaking the quality of the writing for the quality of the facts.
So what do I do instead?
Read for the specific claim, not the tone. Any number, date, citation, rule, or guarantee is a claim that must be checked against a real source before it goes to a client. The more precise and confident the AI sounds, the more important it is to verify — because precise-and-confident is exactly the shape a fabrication takes.
The one-line version.
Confidence in language is not confidence in truth. The AI does the intellectual work; you keep the accountability. You're the one who has to defend the message after it's sent.