Skip to content

AI made people 11.7 points more confident even though about half its answers were faulty

Buyer behaviour


Reader Note:
Confidence was self-rated after solving lab reasoning questions, not measured in a live business workflow.
Evidence:
Experimental
Vendor interest:
None
Sample:
Three preregistered experiments with 1,372 total participants in pooled analysis; university lab and Prolific samples solving 7 CRT reasoning items, with participants randomly assigned to AI access and other conditions.

Good for

Backing budget and pricing decisions in similar situations.

Not for

  • Situations far from the one tested, without checking again.
  • This was measured in one market. It may not hold where you are.
  • The source never says when this data was collected. It could be considerably older than the publication date suggests.

Coming soon

This feature isn’t available yet.