The design

When people compare AI with humans, they usually compare two different partners, so any difference could just be what each one said. We held the partner fixed. Every participant talked to the same language model every time, and was told before each session that the partner was either another person or an AI.

What happened

The same replies under a human label were rated as more immediate, more equal and more relational. The effect showed up in the first chat and didn't fade across three sessions, so it isn't just novelty. The reaction was to who people thought they were talking to, and that puts a measurable cost on disclosing an AI.