Critical thinking is the second of the six orchestration capabilities. As AI takes on analysis and recommendation, the human role shifts towards evaluation.
The quality of the work depends on how rigorously people check what AI produces.
The risk is real. In 2025, Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about 936 real uses of AI. The more people trusted the AI, the less critical thinking they reported.
Gerlich's 2025 study of 666 people found heavier AI use was linked to weaker critical thinking. Cognitive offloading explained much of the link.
How AI goes wrong
Confabulation. AI can produce plausible, consistent and false content. It presents it with the same confidence as accurate content.
Sycophancy. AI tends to agree with the user's framing. It can reinforce your assumptions rather than test them.
Framing effects. Two wordings of the same question can produce opposite answers, each stated with confidence.
Sameness. Sourati and colleagues warn that large language models can make writing and thinking more alike.
Checking also takes longer than generating. A 2026 study of 962 US desk workers found 38% had received AI "workslop" from colleagues. Each lost about 3.4 hours a month fixing it.
Build this capability in your organisation
Growth Performance designs bespoke programmes that turn insight into measurable behaviour change.
The biases that make us miss errors
Automation bias. People over-rely on automated systems and discount information that contradicts them.
Fluency effect. Clear, polished text feels more accurate. AI writes polished text by default.
Confirmation bias. We accept output that matches what we already believe. Kahneman's work on fast and slow thinking explains why checking needs deliberate effort.
Calibrating trust
Trust should move with the task. It can be higher when stakes are low, the task suits AI and checking is easy.
It should be lower for recent events, niche expertise, precise figures or high-stakes decisions.
The STOP protocol
Use four questions before AI output informs a decision:
- Source. Could the AI have reliable information on this topic?
- Tone. Does its confidence match how uncertain the question is?
- Omissions. What has it left out?
- Precision. Can each specific figure, date or attribution be verified?
Making it a team habit
Critical thinking has more impact when it becomes a team norm. Agree what gets checked, by whom and when to escalate.
People challenge AI output when it feels safe to voice doubt. Microsoft's 2026 Work Trend Index found that when managers model AI use, people report 22 points more critical thinking.
Building checking into how the team works is part of Adaptation.
Try this
Take one AI-supported decision from last week. Apply the STOP questions to it. Note what you would have caught.
Then take the free Critical Thinking and Cognitive Bias Detector.
