AI Worth Trusting: Why AI Adoption is Not Proof of AI Success
AI adoption dashboards can look reassuring: more employees are using the tools, more teams are experimenting, and more workflows are incorporating AI. But here’s the thing: adoption is not proof of the level of trustworthiness of an AI system, and certainly not proof of success.
Research on how people respond to algorithms reveals a complicated picture. Sometimes people abandon an algorithm after seeing it make a mistake, even when it continues to outperform a human alternative. In other circumstances, people place too much weight on advice simply because it comes from an algorithm.
Trust also varies by person and context. Domain experts tend to be more skeptical of algorithmic advice than non-experts; People are more likely to find an AI recommendation persuasive when it supports an outcome they prefer. None of these reactions are indicative of a reliable measure of a system’s capability. This can create a serious problem for leaders who use adoption as a proxy for success. A widely used system may be operating without sufficient scrutiny. A less popular system may be demonstrably reliable but poorly understood by the people expected to use it.
What should organizations measure instead? Adoption remains useful but what we really should be measuring is evidence of task performance, error patterns, consistency over time and the quality of human oversight. Leaders should also ask whether users know when to question an output, escalate a concern, or decline to delegate a consequential decision.
Fast uptake tells you that people are using AI. It does not tell you whether the AI is worthy of their reliance. Leading organizations are leveraging assurance tools to effectively manage privacy impact, AI risk, and vendor/third-party risk assessments end to end. At each stage and in each area, organizations need systems with specific and measurable human oversight.
Adoption metrics often miss this important practical step in their governance models. The objective is not to slow adoption for its own sake. It is to ensure that adoption is informed and proportionate to what the system can actually do and to what the organization actually needs.
Adapted from: Kathleen McGinn and Michael Pascu, “Stop Building Trust in AI. Start Building AI Worth Trusting,” published June 9, 2026
