AI use and cognitive skill self-reports (2025-2026)

What this data is, and isn't

The numbers on this page are drawn from published 2025-2026 survey work — Pew Research, BCG, Microsoft + Carnegie Mellon, and the BCG/HBR analysis that drove the "AI brain fry" coverage. All of it is self-report data: it describes what workers and consumers say about their AI use and cognitive skills, not what measurable cognition does in response to AI use.

They are not Senwitt-collected data. Senwitt does not run surveys.

They are not evidence of causation. The studies are mostly cross-sectional correlations: they show that heavier AI use is associated with certain self-reported patterns, but they cannot cleanly establish whether AI use causes the patterns, whether people with the patterns are drawn to AI use, or whether both are caused by something else.

The data

What US workers say about AI's impact

AI mental fatigue (the 'AI brain fry' study)

Critical thinking and AI use (cross-sectional studies)

Microsoft + Carnegie Mellon: critical thinking among knowledge workers

The Microsoft + Carnegie Mellon study (covered widely in 2025) examined critical-thinking patterns in knowledge workers using generative AI. The headline finding was that workers using AI more frequently demonstrated weaker critical-thinking patterns on the specific tasks studied, particularly among workers with lower confidence in their own judgment.

The study is correlational, not causal. It is also specific to the tasks examined — extrapolating to general cognitive ability is exactly the kind of overclaim the broader category has been criticized for.

What patterns hold across the surveys

Three things show up consistently:

AI use is associated with self-reported cognitive load. Workers who use AI heavily report feeling more mentally fatigued than less-exposed peers. This is robust across the surveys.

Heavier AI use correlates with lower self-reported confidence in unaided judgment. Workers who use AI more report feeling less sure of their own conclusions when AI isn't in the loop.

The effect is sharper for younger workers. This pattern shows up in both the Gerlich and Microsoft+CMU work. The interpretation is contested — it could be that younger workers are more AI-exposed, that they are more honest in self-report, or that they are still forming the skills that older workers had built before AI arrived.

What the data does not show

Three things worth flagging.

It does not show that AI use causes cognitive harm in any clinical sense. All the studies are self-report. None show measurable degradation in actual cognitive ability over time as a result of AI use specifically.

It does not show that quitting AI would reverse the patterns. Even if causation ran from AI to lower critical thinking, the policy response isn't obviously "use less AI." The research-supported response is "use AI deliberately and keep deliberate practice on the calendar."

It does not predict the future. The 2026 surveys describe the early years of mainstream AI adoption. The longer-term patterns are not yet measurable. We will know more in 2027 and 2028.

Methodology note

All values on this page are drawn from published surveys, papers, and news coverage of those papers. We have synthesized rather than directly cited specific percentages from any single dataset, because the underlying sources vary in methodology and population definition. The figures should be read as orientation, not as precision measurements.

What this is not

This is not evidence that AI is harming cognition. It is also not evidence that AI is not harming cognition. It is a snapshot of what knowledge workers report about their experience of AI, in the early years of mainstream AI use. Read it accordingly.

Related reading

• AI brain fry — full blog post
• AI dependency self-assessment
• Cognitive offloading — research page
• AI overreliance — research page