How is AI changing the way people make decisions?
AI is becoming a regular input to human decisions, but it is not usually replacing judgement outright. In HCI's Decision-Making and Digital Systems sample, 47% used AI often for decisions and 71% mainly used it to inform their thinking rather than act for them. Reliance rose under uncertainty and difficulty, while verification, override and responsibility often remained high. The central issue is therefore not simply whether AI was used, but how it entered the decision and what remained human.
AI decision-making: the benchmark at a glance
How are people using AI to make decisions?
Most reported AI-supported decision-making in these HCI samples was advisory rather than fully automated. People used AI to clarify options, check their reasoning, reduce uncertainty or create a starting point. Direct delegation existed, but verification, override and personal responsibility remained common.
This creates a mixed picture. AI can expand the information available to a decision-maker while also changing which option feels salient, credible or safe. HCI therefore treats decision-making as a process, not a binary choice between human and machine.
What does this AI decision-making benchmark measure?
AI-supported decision-making occurs when an AI system contributes information, options, recommendations, reassurance or action to a choice that a person is making.
This benchmark measures five connected parts of that process: reliance on AI input, the conditions in which reliance rises, changes in confidence or judgement, perceived control and override, and the delegation of action or responsibility. Together these form the HCI Human–AI Decision System.
The benchmark does not determine whether any particular decision was objectively good or bad. It describes how participants reported incorporating AI into decisions and how that behaviour related to confidence, checking, control, oversight and ownership.
How common is AI-supported decision-making?
AI was a regular decision input for nearly half the sample
Frequency alone does not show whether AI was helpful, persuasive or decisive. The surrounding conditions and retained human behaviours provide the more informative benchmark.
When does reliance on AI increase?
Uncertainty, difficulty and pressure were common reliance conditions
What changes when people rely on AI for difficult decisions?
Reliance changed the starting point and acceptance more than reported checking
Checking an AI output is not the same as generating an independent alternative. A person can verify details while still allowing AI to define the initial frame, options or recommended direction.
How does frequent AI use relate to confidence and judgement?
Frequent users reported more confidence with AI—and more self-doubt when it disagreed
These cross-sectional differences cannot show whether use changed confidence, confidence encouraged use, or another characteristic influenced both.
Do people feel in control when AI influences a decision?
The perceived control gap
Perceived control describes how decision-making feels. It does not independently measure how much a recommendation changed the final choice.
What happens when people delegate decisions or actions to AI?
Delegation comfort was linked to lower oversight, not lower attention to outcomes
Can AI-supported decisions feel less personally owned?
Concern and ambiguity were more common than direct offloading
What is the HCI Human–AI Decision System?
The HCI Human–AI Decision System, developed in HCI's decision-making synthesis report, organises AI-supported decisions into five connected components. It is a descriptive model for locating where influence enters a decision—not a sequence everyone moves through or a score of good and bad use.
Reliance
How often AI is consulted, how central its input becomes and whether it supplies a default starting point.
Conditions of use
Whether uncertainty, difficulty, time pressure, overwhelm or fear of error increases reliance.
Cognitive influence
How AI affects clarity, confidence, option framing, self-doubt and the evaluation of one's own judgement.
Perceived control
Whether people feel able to override, resist or decide independently—even when suggestions steer them.
Delegation
What is handed to AI, how closely action is monitored and where responsibility is understood to remain.
How does decision-making connect with other AI behaviours?
Decisions × Reliance
Reliance describes how central AI becomes, while this benchmark locates that reliance within a decision process. The AI Reliance benchmark provides the broader behavioural context.
Decisions × Verification
Verification remained high across very different reliance groups. The AI Verification benchmark examines the strength and limits of checking as a separate behaviour.
Decisions × Trust
Trust can make acceptance or delegation feel more appropriate, but trust and action are not the same construct. The AI Trust benchmark shows how trust relates to checking, monitoring and override.
Decisions × Thinking partnership
Using AI to test or extend reasoning differs from allowing it to define the decision. The Thinking Partnership benchmark examines that boundary.
What does wider research show about AI-assisted decisions?
Human–AI decision-making has several distinct forms
A 2026 systematic review and bibliometric analysis of 627 peer-reviewed articles identified four broad human–AI decision paradigms: adaptive intuitive, programmed algorithmic, interpretive analytical and integrative hybrid decision-making. The review reinforces the need to distinguish the role AI plays rather than treating every AI-assisted choice as the same behaviour.
Effective assistance depends on human judgement and interaction design
Steyvers and Kumar describe three challenges for AI-assisted decision-making: recognising when AI should be used, developing accurate mental models and reliance strategies, and designing interactions that preserve useful human input. Their review notes that receiving AI advice after an initial human judgement can support more independent reflection than seeing the recommendation first.
Biased AI advice can influence later unaided judgements
Across three experiments, Vicente and Matute found that participants exposed to systematically biased AI recommendations reproduced related errors, including after the recommendations were removed. The experiments concern a controlled classification task, not everyday decision-making, but they demonstrate that AI influence can persist beyond one visible recommendation.
Access to AI advice does not produce equal decision outcomes
In a randomised field experiment with 640 Kenyan entrepreneurs, an AI business adviser had no statistically significant average effect on business performance. Higher-performing firms benefited while lower-performing firms performed worse, despite broadly similar use. The authors linked the difference to which suggestions people selected and how they implemented them.
Otis et al. (2024)Harvard Business School AI Institute (2026)
HCI's earlier reports locate the human signals behind the statistics
HCI's decision synthesis describes how reliance, conditions of use, cognitive influence, perceived control and delegation form one behavioural system. The companion Human Signals report focuses more closely on uncertainty, reassurance, self-trust and judgement stability. The present benchmark builds on both reports while adding verified cross-sample comparisons from the underlying datasets.
What can be concluded from this AI decision-making benchmark?
AI-supported decision-making was already common in HCI's samples, but most participants described AI as an input to judgement rather than an autonomous actor. Verification, override and final responsibility remained high.
The strongest behavioural shifts appeared when decisions became uncertain, difficult or pressured. Higher reliance was associated with making AI the default starting point and accepting its direction more readily, even though self-reported verification remained almost unchanged.
Frequent users reported greater confidence with AI support and greater use of AI to check their reasoning. They also reported more doubt in their own judgement when AI disagreed. Likewise, perceived control often coexisted with feeling steered, and delegation comfort coexisted with reduced oversight.
The evidence therefore supports neither a simple displacement story nor a universal augmentation story. AI changes decision-making by changing the conditions, starting points, confidence signals and oversight patterns within which human judgement operates. The relevant benchmark is whether people retain the ability to form, test, override and own a decision—not whether AI was present at all.
How can I compare the role AI plays in my decisions?
One measure—such as how often you ask AI for advice—cannot show whether you have retained decision ownership. The more useful comparison includes decision delegation, reliance, verification, trust, thinking partnership and human agency.
The Human Clarity Institute AI Identity & Behaviour Assessment asks 39 questions across nine dimensions of AI behaviour. It takes about 3–4 minutes and provides free personalised results comparing your responses with all HCI participants, people in your age group and people who use AI as often as you do.
The assessment provides a behavioural reference point for reflection and comparison. It does not test decision quality, diagnose dependence or determine whether using AI for a particular choice is appropriate.
Frequently asked questions about AI decision-making
How common is it to use AI for decisions?
In HCI's Decision-Making and Digital Systems 2026 sample, 47% reported often using AI to help make decisions. Seventy-one percent mainly used it to inform judgement rather than act on their behalf. These figures describe a non-probability sample and are not national population estimates.
When are people most likely to rely on AI for a decision?
Reliance was commonly reported under uncertainty and pressure. In one HCI sample, 61% relied more on AI when uncertain, while 58% did so when decisions were difficult, under time pressure or when overwhelmed. The results are descriptive and do not show that those conditions caused reliance.
Does relying on AI mean people stop checking it?
Not in HCI's difficult-decision comparison. Verification was reported by 86% of both high- and low-reliance respondents. However, high-reliance respondents were much more likely to use AI as a default starting point and accept decisions unchanged, showing that checking and influence can coexist.
Does frequent AI use affect confidence in decisions?
Frequent users in the AI Decision Dependence & Cognitive Caution 2025 sample were more likely to report confidence with AI support and to use AI to check their thinking. They were also more likely to doubt their own judgement when AI disagreed. Cross-sectional data cannot show which behaviour came first.
Can people feel in control while AI still influences them?
Yes. In HCI's Autonomy, Control & Perceived Independence 2026 sample, 75% reported high control. Among that group, 57% also felt steered by AI suggestions. Perceived control and perceived influence are related but separate experiences.
Are people comfortable delegating decisions to AI?
In HCI's Delegated Action & Responsibility 2026 sample, 42% reported high comfort delegating decisions or actions to AI. Across the full sample, 85% reported paying attention to outcomes. Delegation comfort was associated with reduced oversight but not with lower reported attention to results.
Does AI make decisions feel less like my own?
For a substantial minority, it can. In HCI's Cognitive Load, Fatigue & Decision Offloading 2025 sample, 39% reported feeling detached from an outcome when AI was involved, 38% reported blurred boundaries between their judgement and automated suggestions, and 34% questioned whether a decision was truly theirs.
What is the HCI Human–AI Decision System?
It is a descriptive model with five components: reliance, conditions of use, cognitive influence, perceived control and delegation. The model shows where AI can enter and shape a decision without assuming that everyone follows the same sequence or that one pattern is always better.
How can I compare my AI decision-making with other people?
The Human Clarity Institute AI Identity & Behaviour Assessment asks 39 questions across nine dimensions and provides free personalised comparisons with all HCI participants, people in your age group and people who use AI as often as you do. It provides a behavioural benchmark, not an evaluation of decision quality.
Related HCI benchmarks
Methodology: Learn how HCI builds benchmarks. Dataset sources are linked throughout the page and in the accompanying structured data.
How does your AI decision-making compare?
Compare your responses with HCI participant benchmarks across nine dimensions of AI behaviour—including decision delegation, reliance, verification, trust, thinking partnership and human agency.
See Where I Sit →- 01All participants in the benchmark
- 02People in your age group
- 03People who use AI as often as you do