What makes an AI interface feel “trustworthy” to you?

Some tools just feel more reliable even if the backend models are similar. Is it the tone, layout, citations, or transparency of the process? What gives you confidence to act on what AI says?