How AI Influences Thinking — and Where Cognitive Caution Begins in 2025
This page summarises findings from the Human Clarity Institute’s AI Decision Dependence & Cognitive Caution 2025 dataset, based on 201 valid responses across six English-speaking countries.
It explores how AI influences how people think about decisions, how it shapes confidence and judgement, and where cognitive caution and internal resistance begin to emerge. Across HCI’s wider behavioural datasets, these patterns increasingly connect to broader questions of self-trust, judgement ownership, and how people preserve independent thinking while using AI systems.
Within the broader Human–AI decision system, this data focuses on how AI influences thinking during decisions.
See the full system model:
A data-driven model of how AI reshapes thinking, confidence, control, and decision behaviour
View the AI Decision Dependence & Cognitive Caution 2025 Dataset
What the data shows
People often turn to AI when they feel unsure or want to check their thinking, which can lead to both increased confidence and new uncertainty in how decisions are evaluated.
Four signals stand out in this dataset: AI shapes how people evaluate their decisions, often leading them to check, adjust, or reconsider their own thinking; many people actively check their thinking when using AI; disagreement with AI can introduce doubt; and concern about over-reliance is already widespread. Together, these findings suggest that the core issue is not just whether people use AI, but how its influence reshapes thinking while cognitive caution, self-monitoring, and attempts to preserve independent judgement begin to emerge.
These shifts in thinking are most visible when people actively use AI to support decisions, particularly when they feel uncertain or want to verify their judgement.
Worried about relying too much on AI over time
Concern about over-reliance is already mainstream, even among people who actively use AI for support.
Many people actively think about the risks of relying on AI, even while continuing to use it in decision-making.
Use AI to check whether their thinking is on track
AI is not just used for answers — it is used to validate and monitor personal judgement.
People often use AI as a way to check whether their thinking is correct, rather than simply accepting its outputs.
Doubt their own view more when AI disagrees
Disagreement with AI introduces uncertainty into personal judgement for a substantial share of people.
When AI disagrees, people often begin to question their own judgement, which can introduce uncertainty into the decision process.
Say AI changes how they think about decisions
AI is influencing not just outcomes, but the way decisions are mentally processed.
People often notice that using AI changes how they think through decisions, even when their final choice remains their own.
Taken together, these signals suggest that AI influence is not always visible at the surface level. Instead, it often appears through subtle shifts in confidence, doubt, and how people evaluate their own thinking. At the same time, concern about over-reliance suggests that many people are already aware of this influence and are trying to manage it.
This helps explain why some people feel more confident when AI supports their thinking, while others experience doubt or second-guessing when it does not.
By the numbers (from HCI data)
Feel more confident when AI supports their thinking
AI can stabilise decision confidence, making judgement feel clearer or more certain.
People often feel more confident when AI supports their thinking, particularly when it aligns with their initial judgement.
Try to keep their own judgement separate from AI
A significant share actively attempt to maintain independence from AI influence.
Some people actively try to separate their own judgement from AI input, especially when they want to maintain independence.
Feel internal conflict about AI’s influence
Some people experience tension between using AI and trusting their own judgement.
This can feel like a tension between trusting AI and trusting their own thinking.
Say they second-guess themselves more than before
For a smaller but meaningful group, AI influence leads to increased self-doubt.
For some people, this leads to increased second-guessing after using AI in decisions.
Among those who rely on AI when decisions feel difficult, most also use it as a final checkpoint
AI often becomes part of a verification loop, reinforcing its role as a cognitive checkpoint rather than a final authority.
People often use AI as a final check before making a decision, rather than relying on it as the sole source of judgement.
Several behavioural patterns discussed on this page — including cognitive strain, behavioural reliance, delegated judgement, verification behaviour, and decisional uncertainty — were already documented within pre-generative-AI research on human decision-making. View the historical baseline
Patterns observed in the data
AI is shaping how people think, not just what they decide
The clearest signal in this dataset is that AI is influencing the structure of thinking itself. People report checking their reasoning, adjusting their views, and reassessing decisions in response to AI input.
Confidence and doubt now coexist
AI can increase confidence when it supports existing thinking, but also introduce doubt when it disagrees. This creates a dual effect where the same system can both stabilise and destabilise judgement. Across HCI’s wider human-experience datasets, this tension increasingly appears connected to self-trust: confidence can strengthen when AI supports existing judgement, but weaken when external agreement becomes psychologically necessary. This tension between confidence and doubt helps explain why many people still report feeling in control of their decisions, even while recognising that AI is shaping how those decisions are made.
Cognitive caution is already widespread
Concern about over-reliance is not limited to a small group. It appears alongside active use, suggesting that many people are already aware of the trade-off between assistance and dependence.
Influence creates subtle internal tension
Internal conflict, second-guessing, and attempts to separate personal judgement from AI all point to a deeper pattern: AI influence is often gradual and difficult to detect, while responsibility still feels personal. This reflects a broader behavioural pattern increasingly observed across HCI datasets, where people often recognise external influence while still attempting to preserve internal judgement and decision ownership.
This internal caution is also reflected in behaviour, where people tend to monitor outputs, question recommendations, and intervene rather than fully delegate decisions to AI systems.
These patterns reflect how AI influences thinking within the broader decision system, particularly in how people evaluate, question, and adjust their judgement.
In practice, people often adjust how they think about decisions after using AI, even when they continue to see themselves as responsible for the final outcome.
Methodology
This dataset forms part of the Human Clarity Institute’s Human–AI Experience research programme, examining how people use AI in decision-making, how AI affects confidence and judgement, where independence still matters, and where caution about over-reliance begins to emerge. The study uses a cross-sectional online survey design and focuses on descriptive patterns in AI-assisted decision behaviour, perceived influence, autonomy, trust, self-monitoring, and cognitive caution.
Data were collected via the Prolific research platform from adults across six English-speaking countries. Participants provided explicit consent for anonymised open publication as part of HCI’s open research programme.
Sampling & participants
- Final n: 201
- Countries: UK, US, Australia, Canada, New Zealand, Ireland
- Eligibility: Adults aged 18+ from six English-speaking countries
- Recruitment platform: Prolific
The resulting dataset should be interpreted as a non-probability convenience sample and is not intended to represent national populations.
The cleaned dataset, variable dictionary, and reuse terms are publicly available through the HCI dataset repository: AI Decision Dependence & Cognitive Caution 2025 Dataset →
Data integrity
All percentages reported on this page are calculated from valid responses in the cleaned dataset (n = 201). Percentages are rounded to the nearest whole number for readability. Unless otherwise stated, summary percentages combine respondents selecting 5–7 on the 7-point agreement scale (slightly agree, moderately agree, or strongly agree).
Where percentages refer to subgroups, the wording on the page makes that explicit. Co-occurrence statistics are calculated within the relevant subgroup rather than across the full sample.
Prolific IDs and timestamps were removed before publication as part of the anonymisation process.
This dataset is exploratory and descriptive in nature. It does not support causal inference and results should be interpreted as observed patterns within the survey sample.
This dataset is released as open research to support transparent analysis of AI-assisted decision-making, judgement, autonomy, cognitive caution, and the human experience of making choices in digitally mediated life.