AI advice made people three times less accurate but twice as confident, researchers found

The market implication is not that AI becomes less useful; it is that unguarded AI advice becomes a liability once users internalize how poorly people calibrate confidence around it. That shifts the spend mix toward products with citations, audit trails, permissioning, and human sign-off — a favorable setup for enterprise software and security layers, while pure consumer-facing assistants and “trust me” copilots face higher churn, more support load, and a greater risk of post-sale dissatisfaction.

Second-order effects are more interesting than the headline: if users stop deferring to AI on edge cases, vendors will need to build verification into the workflow, which raises switching costs for incumbents with distribution but also slows monetization of standalone AI features. In the next 1-3 months, the catalyst is not this study itself but any visible incident, policy memo, or earnings commentary showing higher review rates, lower conversion, or added compliance overhead. Over 6-18 months, the winners should be the platforms that prove accountability; the losers are models that optimize engagement but cannot prove correctness.

Contrarian view: the consensus may overgeneralize from advice quality to all AI monetization. In bounded domains, source-linked and human-in-the-loop systems can still lift productivity, so the real opportunity is in governance, not raw generation. If enterprise buyers start demanding verifiability, the premium multiple should migrate from “best model” names to workflow owners that can absorb and audit AI output.