Generative AI tools like ChatGPT and Microsoft Copilot have rapidly become embedded in everyday knowledge work, reshaping how tasks are initiated, executed, and evaluated. The promise is clear: faster workflows, richer ideation, and unprecedented access to synthesised information. But what happens to critical thinking when AI becomes a default collaborator?

A recent study of 319 knowledge workers, who collectively shared 936 real-world GenAI use cases, offers some insight into this shift. The findings reveal that GenAI doesn’t simply reduce or replace critical thinking; it reshapes it, redistributing cognitive effort in ways that both empower and endanger knowledge workers.

How Knowledge Workers Enact Critical Thinking With GenAI

The study shows that workers still engage in critical thinking, but the nature of that thinking has changed. Participants described three recurring phases:

Goal and Query Formation

Before prompting, workers set objectives, clarify constraints, and refine queries. This early-stage reflection is essential because the quality of the AI’s output depends heavily on the clarity of the input. As the report notes, workers define critical thinking as “setting clear goals, refining prompts, and assessing AI-generated content to meet specific criteria and standards.”

Inspecting Responses

Workers evaluate AI outputs using both objective criteria (accuracy, compliance with instructions) and subjective ones (tone, relevance, feasibility). Verification becomes central. Participants frequently cross‑checked facts against external sources or their own expertise, especially in high‑stakes domains.

Integrating Responses

Rather than accepting outputs wholesale, workers selectively incorporate, edit, or reframe AI-generated content. This includes aligning style and tone, merging multiple outputs, or discarding suggestions that don’t meet professional standards.

When and Why Critical Thinking Happens

Interestingly, critical thinking is more likely when workers have high confidence in their own abilities and lower confidence in the AI. Workers described engaging critically to:

• improve the quality of their work
• avoid errors or negative consequences
• maintain or develop their own skills

However, barriers such as time pressure, lack of awareness, and difficulty verifying unfamiliar content often reduce critical engagement. The study warns that routine reliance on AI, especially for low-stakes tasks, may gradually erode independent problem‑solving.

Shifts in Cognitive Effort: From Doing to Stewarding

One of the most significant findings is the shift in where cognitive effort goes:

• Recall & comprehension: from gathering information to verifying it
• Application: from solving problems to integrating AI-generated solutions
• Analysis, synthesis & evaluation: from executing tasks to stewarding them

Workers increasingly act as overseers rather than producers. GenAI shifts effort “from material production to critical integration,” a pattern previously observed in controlled studies but now confirmed across diverse real-world workflows.

Designing GenAI to Strengthen, Not Weaken, Critical Thinking

The study argues that GenAI tools must be designed to support critical thinking rather than silently diminish it. This includes:

• proactive prompts that highlight opportunities for reflection
• reactive support when users request deeper reasoning
• explanations of AI logic
• guided critiques and verification tools
• features that build user confidence and domain expertise

In the end, GenAI may not be the downfall of human thought so much as a persistent reminder that thinking was never optional in the first place. The tools can draft, summarise, and speculate to their heart’s content, but someone still has to make sure the whole thing isn’t confidently wrong (poo).