Looks good, must be right? We seem to scale back our questioning once AI starts building. This is a key finding from the latest Anthropic report on AI fluency.
When AI produces artifacts (apps, code, documents, visuals, or interactive tools), users are significantly less likely to verify the work:
- 3.7 pp less likely to check facts
- 3.1 pp less likely to question reasoning
- 5.2 pp less likely to identify missing context
This suggests an interesting conundrum: as AI moves from being a chat/conversation partner to a builder, our skepticism fades. We are far more likely to question a text response than a functional piece of code or a formatted document, even though the latter often requires the most oversight.