Generative AI is reshaping how knowledge workers think. Rather than removing critical thinking, it shifts the work from producing content to verifying, contextualizing, and taking responsibility for AI-assisted outcomes. A Microsoft Research and Carnegie Mellon study provides useful evidence for how this shift is happening in real work.
What the research found
The study surveyed 319 knowledge workers who shared 936 real-world examples of using generative AI (GenAI) at work. The researchers examined when workers engaged in critical thinking and how GenAI affected the effort required for activities such as analysis, synthesis, and evaluation.
Participants reported applying critical thinking in 555 of the 936 examples—about 59.29% of cases. They also generally perceived that critical-thinking activities required less effort when using GenAI than when working without it.
Confidence changes behavior
One of the study’s most important findings is that confidence strongly influences critical engagement. Higher confidence in GenAI was associated with less critical thinking, while higher confidence in one’s own abilities was associated with more.
This creates a practical risk: when workers trust an AI output too readily, they may skip verification, accept generic answers, or fail to adapt the result to their organization’s context. Conversely, workers who feel capable of evaluating the task are more likely to challenge, refine, and validate AI-generated work.
The new role of knowledge workers
GenAI reduces the burden of routine production—drafting, summarizing, searching, and generating initial ideas. But it increases the importance of human oversight: checking accuracy, integrating outputs with business context, identifying gaps, and ensuring the result meets quality standards.
In other words, the role is shifting from content producer to AI supervisor. This is especially visible in knowledge retrieval, where AI can automate information gathering but workers must still invest effort in validating accuracy.
What critical thinking looks like with AI
Workers apply critical thinking when using GenAI by:
-
Setting clear goals and defining success criteria before prompting.
-
Refining prompts iteratively to improve relevance and quality.
-
Verifying outputs against trusted sources and their own expertise.
-
Assessing whether the response meets task-specific standards.
-
Integrating AI output into broader work rather than accepting it as final.
This reframes critical thinking as an active workflow practice—not merely a final review step.
Barriers to critical engagement
The research identified three main barriers:
-
Awareness: Workers may not recognize when AI output needs scrutiny.
-
Motivation: Time pressure, routine work, and narrow job scope can discourage deeper evaluation.
-
Ability: Workers may struggle to improve or challenge AI responses in unfamiliar domains.
These barriers matter because efficiency gains can become risks when people rely on AI without retaining the ability to question it.
What leaders should do
Organizations should not treat GenAI adoption as simply a productivity initiative. They should design workflows that preserve human judgment.
-
Require verification for high-stakes or factual outputs.
-
Teach employees to evaluate AI responses, not just generate prompts.
-
Encourage domain expertise alongside AI fluency.
-
Build review checkpoints into AI-assisted workflows.
-
Avoid using AI as a substitute for accountability.
The goal is not to slow AI adoption. It is to ensure that AI amplifies informed judgment rather than replacing it.
Important limitation
This is a self-reported survey, not a controlled experiment. It measures how knowledge workers perceive their critical thinking and effort, rather than directly measuring cognitive performance. Still, its large set of real-world examples makes it a valuable signal for how GenAI is changing modern knowledge work.
The takeaway
Generative AI can make critical thinking feel easier, but that does not mean it should become less important. The most effective AI-enabled professionals will be those who combine AI’s speed with strong verification habits, domain expertise, and clear accountability.
References
https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf