How much do people rely on AI?
AI reliance is common, especially when people feel uncertain, overwhelmed or faced with difficult decisions. But reliance is not the same as dependence. Across HCI datasets, most respondents still verify AI outputs, override recommendations and feel in control, while stronger potential-dependence indicators remain less common. The central question is not simply how often someone uses AI, but whether judgement and independence remain active alongside that use.
AI reliance: the benchmark at a glance
58% of respondents in HCI's Decision Making & Digital Systems sample reported relying more on AI when decisions become difficult.
85% in that same sample reported verifying AI outputs before using them, and 84% reported overriding AI when they disagree.
74% of respondents in a separate HCI autonomy sample reported feeling in control while using AI, and 88% said they could ignore AI recommendations.
Potential-dependence indicators were less common in the autonomy sample: 21% regularly handed decisions over to AI, 15% felt uncomfortable without AI and 10% followed AI despite discomfort.
What This Benchmark Measures
This benchmark separates three ideas that are often treated as if they mean the same thing. AI use means interacting with an AI system. AI reliance means leaning on AI to support a task, judgement or decision. Potential-dependence indicators are behaviours that may suggest reduced independence, such as regular decision handover, difficulty functioning without AI, reduced checking or following AI despite discomfort.
The distinction matters because high use is not automatically high reliance, and high reliance does not automatically mean dependence. Reliance becomes more informative when it is read alongside verification, override capability, delegation and perceived independence.
What does HCI's survey evidence show about AI reliance?
AI is often reported as useful under uncertainty
This is an important starting point: people often turn to AI because it is useful under uncertainty, not because they have already surrendered independent judgement.
Reliance is often reported when decisions or tasks become difficult
This makes reliance partly contextual. Difficult decisions and cognitive pressure appear to be important conditions for understanding when people lean more heavily on AI.
What does retained judgement during AI reliance look like?
Retained judgement markers
- Verification remains active: users check important AI outputs before accepting or acting on them.
- External judgement remains present: users compare AI responses with other sources or their own knowledge.
- Override remains available: users can ignore AI recommendations when they disagree.
- AI is support, not replacement: users rely on AI to clarify, structure or extend thinking rather than outsource judgement entirely.
- Independence remains intact: users believe they can still operate without AI when needed.
- Confidence remains balanced: confidence in AI does not replace confidence in one's own evaluative ability.
This is the page's central counterweight: reliance can increase while verification, override and independent judgement remain active.
How should AI reliance patterns be interpreted?
HCI Reliance Pattern Framework
Support under uncertainty
AI is used to clarify options, structure information or make an unfamiliar task easier to approach.
Situational reliance
Reported reliance is higher when decisions are difficult, tasks are demanding or the user feels overwhelmed.
Reliance with retained judgement
AI supports the task while verification, source comparison, override and decision ownership remain present.
Elevated-reliance indicators
AI becomes a frequent default or a regular source of direction and decision handover.
Potential-dependence indicators
Reliance appears with reduced independence, discomfort without AI, perceived skill decline or following AI despite discomfort.
How to use the framework
- Read the groups as patterns, not stages: they organise behaviours without implying that everyone moves through them in sequence.
- Look beyond frequency: verification, override, delegation and independence change what a given level of reliance means.
- Consider context: the same person may rely on AI differently across familiar tasks, difficult decisions and periods of cognitive pressure.
The framework preserves the distinction between useful support, reliance with retained judgement and patterns that may warrant closer attention.
How common are dependence markers?
The important distinction is that reliance-related behaviours are common, while the stronger indicators associated with potential dependence remain minority patterns.
What changes when reliance combines with other behaviours?
Higher reliance + active verification
This combination would be more consistent with AI use that remains actively evaluated.
Higher reliance + strong override
This combination would be more consistent with retained decision ownership and agency.
Higher reliance + reduced verification
This combination would warrant closer attention because acceptance is less evaluative.
Decision handover + discomfort without AI
This combination would be more consistent with potential dependence than usage frequency alone.
Do people lose agency as they rely on AI?
How does the HCI benchmark relate to wider research?
AI overreliance is recognised as a major risk category
The MIT AI Risk Repository classifies overreliance and unsafe use as a distinct AI risk category linked to autonomy, dependence and unsafe decision-making. HCI adds a behavioural measurement layer by showing how often relevant reliance, verification, override and delegation behaviours appear within its survey samples.
Reviews propose possible pathways, but HCI has not tested a progression
Recent review literature maps possible links among automation bias, uncritical acceptance, cognitive offloading, overreliance and potential deskilling. This wider research helps explain how dependence might develop over time. HCI's current contribution is different: it measures the related behavioural signals without presenting them as a proven sequence.
Psychological factors are associated with overreliance
A 2026 SEM–ANN study examines how dependency, anxiety, fear of missing out and other psychological factors are associated with AI overreliance. HCI complements this psychological account by measuring behaviours such as verification, decision handover, override capability and discomfort without AI.
Confidence and critical-thinking effort show associations in knowledge work
Microsoft Research surveyed 319 knowledge workers and analysed 936 examples of generative-AI use. Higher confidence in generative AI was associated with less self-reported critical-thinking enactment, while higher self-confidence was associated with more. Of the 936 examples, 379 contained no self-reported critical-thinking enactment.
This reinforces the importance of confidence, verification and active evaluation. Reliance is not only about how often AI is used; it is also about whether the human evaluator remains engaged.
Methodology: Learn how HCI develops these benchmarks in How Human Clarity Institute Builds AI Behaviour Benchmarks.
What can be concluded from this AI reliance benchmark?
Reliance under difficulty is common in the HCI samples
More than half of respondents in two HCI samples reported relying more on AI when decisions or tasks become difficult. Difficulty, uncertainty and cognitive pressure are therefore central to understanding why reliance varies.
Verification, override and perceived control remain widespread
Large majorities reported verifying outputs, overriding recommendations or retaining control. Reliance does not automatically remove active evaluation or decision ownership.
Stronger potential-dependence indicators are less common
Regular decision handover, discomfort without AI, perceived skill decline and following AI despite discomfort were reported by minorities in the autonomy sample. These behaviours provide clearer warning signals than usage frequency alone.
The combination of behaviours matters most
No single behaviour defines potential dependence. Reliance becomes more informative when it is considered alongside verification, override, delegation, confidence and perceived independence.
Why This Matters
The important question is not simply whether someone relies on AI. The more useful question is whether that reliance remains active, intentional and reversible.
AI can provide valuable clarity, structure and support while people continue to verify, override and retain decision ownership. Potential concern appears when reliance is accompanied by reduced checking, regular handover, discomfort without AI or weakened confidence in independent action. Behavioural benchmarks make that distinction visible. Individuals can compare these patterns using the HCI AI Identity & Behaviour Assessment.
How can I compare my AI reliance with other people?
Group-level findings can show how reliance appears across HCI samples, but they cannot determine where one individual sits. The HCI AI Identity & Behaviour Assessment is designed to provide that personal comparison.
The assessment takes 3–4 minutes and asks 39 questions across nine dimensions of AI behaviour. Free personalised results compare your responses with all HCI participants, people in your age group and people who use AI as often as you do.
This comparison helps show whether your reliance is accompanied by verification, decision ownership and human agency, or by more regular delegation and discomfort without AI. It provides a behavioural reference point, not a clinical diagnosis or an intelligence test.
Frequently asked questions about AI reliance
Does frequent AI use mean someone is dependent on AI?
No. Frequency measures how often AI is used. Potential dependence requires a broader pattern, such as reduced verification, regular decision handover, discomfort without AI or diminished ability to act independently.
When does AI reliance become concerning?
Reliance becomes more concerning when it appears alongside weaker checking, reduced override, routine delegation or difficulty functioning without AI. No single item establishes dependence.
Does HCI's evidence show that AI causes skill decline?
No. In one HCI autonomy sample, 17% reported believing AI was causing their skills to decline. That is a self-reported perception, not evidence of measured skill loss.
What indicates retained judgement during AI use?
Verification, source comparison, override capability, decision ownership and the ability to work without AI are behaviours consistent with retained judgement.
What is the HCI Reliance Pattern Framework?
It is a descriptive framework that organises AI reliance into support under uncertainty, situational reliance, reliance with retained judgement, elevated-reliance indicators and potential-dependence indicators. The groups are patterns rather than fixed stages.
How can I compare my AI reliance with other people?
The HCI AI Identity & Behaviour Assessment asks 39 questions across nine dimensions of AI behaviour and takes 3–4 minutes. Free personalised results compare your responses with all HCI participants, people in your age group and people who use AI as often as you do. It provides a behavioural reference point rather than a clinical diagnosis or an intelligence test.
Related HCI benchmarks
How does your AI reliance compare?
Compare your responses with HCI participant benchmarks across nine dimensions of AI behaviour—including reliance, verification, decision delegation, trust 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