Anthropic studied nearly 10,000 conversations with Claude to understand what makes people effective at working with AI.

A key finding: When AI produced polished, finished-looking work like a doc or app, people were less likely to question the reasoning, notice gaps, or check facts. TLDR: The more competent the output looked, the less people scrutinized it.

This is the 2026 Edition of automation bias. This tendency to trust automated outputs over your own judgment is something human factors researchers have studied for 40+ years. One fix: Make the AI's output look like a draft, not a decision.