People are beginning to ask questions that technical AI benchmarks cannot answer. Am I relying on AI more than other people? Do I still form my own view before asking an AI assistant? Am I checking its answers less often as it becomes more useful? Is AI strengthening my skills, or quietly taking over parts of the thinking I used to do myself?

The Human Clarity Institute has formally introduced the HCI AI Identity & Behaviour Assessment—also referred to as the HCI AI Assessment—to provide a clearer reference point for those questions. It is a 39-question, non-clinical self-report assessment for adults who want to understand their own behaviour around AI and compare it with patterns reported by other HCI participants.

Is there a tool that compares your AI use with other people?

Yes. The HCI AI Identity & Behaviour Assessment compares your reported AI-use behaviour with HCI participant benchmarks—overall, within your age group and, where sufficient data is available, among people who use AI as often as you do.

It examines not only how often you use AI, but the role it plays in your thinking, decisions, verification, reliance, agency and relationships.

Why a human-side AI benchmark is needed

AI systems are routinely compared on reasoning, coding, accuracy and task performance. Those benchmarks tell us about the system. They do not tell an individual what is changing on the human side of the interaction.

Frequency alone is not enough. Two people may both use ChatGPT, Claude, Copilot, Gemini or another AI assistant every day while developing very different patterns. One may form an initial view, use AI to challenge it, verify important claims and rewrite the result in their own reasoning. Another may increasingly let AI provide the first framing, the structure and the final judgement. Both are frequent users; the role AI is playing is different.

This distinction is becoming more important as AI moves beyond information retrieval and into research, writing, analysis, problem-solving and personal decisions. In the first Anthropic Public Record, based on nearly 52,000 Americans, 56% expressed concern about cognitive dependency—described as AI integration leaving people less able to think for themselves. It was the second most commonly reported AI fear after job loss.

Anthropic described cognitive dependency as a mostly anticipatory fear: among the 56% who expressed concern, only roughly one-fifth said they would experience significant disruption if AI became unavailable. That gap is precisely why a personal baseline can be useful. People need a way to examine the behaviour already forming before a vague concern becomes difficult to interpret from inside an established routine.

Important distinction: HCI uses “AI behaviour assessment” to mean an assessment of human behaviour while using AI. It does not test the behaviour, accuracy or performance of an AI model or agent.

How the HCI assessment differs from other AI and digital-use tools

The assessment is designed to answer a different question from screen-time trackers, proficiency quizzes, clinical scales and technical AI benchmarks.

Compared with What is different about the HCI assessment
Screen-time and digital wellbeing trackers HCI examines the role AI plays in behaviour, not simply minutes or frequency of use.
AI literacy and proficiency quizzes HCI examines how a person uses AI, not their knowledge of AI or skill operating a particular tool.
Clinical dependence scales HCI is non-clinical and does not diagnose dependence, addiction or a mental-health condition.
AI model benchmarks such as MMLU or SWE-bench HCI examines reported human behaviour around AI use, not model capability or task performance.

How do you know if you’re relying on AI more than you realise?

Reliance can be difficult to recognise from frequency alone. A person may use AI every day while retaining independent judgement, while another may use it less frequently but habitually turn to it first when decisions feel difficult, uncertain or mentally effortful.

The HCI AI Identity & Behaviour Assessment examines reliance alongside trust, verification, decision delegation and human agency, then compares the reported pattern with HCI participant benchmarks. It provides a structured reference point for seeing whether reliance is playing a larger or different role than the person expected.

Who the assessment is for

The assessment is designed for adults who use generative AI in work, study, research, creativity, decision-making or everyday life and want more than a screen-time count. It is particularly relevant for people asking:

  • How does my use of AI compare with other people?
  • Am I relying on AI more than I realise?
  • Is AI changing the way I think or make decisions?
  • Do I still check important answers and form my own judgement?
  • Am I using AI to strengthen my skills, or giving parts of those skills away?
  • What would a useful baseline of my current AI habits look like?

It may be useful to knowledge workers, researchers, students aged 18 and over, creators, professionals and other regular AI users. It is also designed for people who do not believe they have a “problem” but want to see what is typical, uncommon or distinctive before their habits change further.

The assessment is not an employee-screening tool, an AI proficiency test or a clinical test for addiction or dependence. It does not classify a high or low score as automatically healthy, unhealthy, good or bad.

Built on HCI’s wider research programme of more than 10,500 participants across 21 studies. Read the published benchmark methodology.

What the HCI assessment measures

The HCI AI Identity & Behaviour Assessment examines nine connected aspects of a person’s reported relationship with AI:

  • Trust: How readily you place confidence in AI-generated information, suggestions and explanations.
  • Reliance: How often you turn to AI for support, particularly when a task feels difficult, uncertain or effortful.
  • Verification: How often you question, check or independently confirm important AI-generated claims and outputs.
  • Decision Delegation: How much AI participates in framing choices, recommending actions or shaping final decisions.
  • Human Agency: How strongly you retain control, responsibility and confidence in your own judgement when AI is involved.
  • Emotional Regulation: How often AI is used for reassurance, emotional processing, comfort or support during uncertainty.
  • Thought Partnership: How AI is used to explore ideas, organise thinking, test assumptions or work through problems.
  • Disclosure: How comfortable you are sharing personal, sensitive or private information with AI systems.
  • Social Transparency: How openly you acknowledge AI’s involvement in work, decisions or outputs shared with other people.

Together, the dimensions show more than how much AI is used. They examine when a person turns to it, how much authority they give it, whether they retain independent judgement, how they verify outputs, what they disclose, and how clearly they recognise AI’s contribution to the final result.

The assessment typically takes under 5 minutes. It does not require access to chat histories, private documents or AI accounts. Participants answer structured questions about their own behaviour and receive free personalised results showing selected findings against HCI participant benchmarks. An optional expanded report provides a fuller analysis across the nine dimensions and 39 responses.

A report you can question against your own AI conversations

Participants who choose the expanded report can take it back to the AI assistant they use most and ask the assistant to compare the self-reported findings with examples from the conversation history it can access. This provides an optional second perspective: not only what the participant reported, but what their observable conversations appear to support or challenge.

Suggested prompt: “Here is my HCI AI Identity & Behaviour Report. Using only the conversation history you can access, show me specific examples that support or contradict its findings and my assessment answers. Quote the moments and tell me where my reported behaviour differs from what our conversations show. If you cannot access enough history, say so.”

This step is optional and takes place within the participant’s chosen AI service. HCI does not require or receive the person’s chat history. The strength of the comparison depends on how much relevant conversation history the assistant can access.

What comparison adds that a checklist cannot

Most online discussions of AI overuse end with a generic checklist: notice whether you use AI often, whether you feel uncomfortable without it, or whether you ask it to do tasks you once completed yourself. These questions may prompt reflection, but they do not provide a reference population.

A benchmark changes the question from “Does this sound like me?” to “Where does my reported pattern sit among other participants?” That distinction can reveal whether a behaviour that feels ordinary is widely shared, relatively uncommon or simply different from the user’s own perception.

The assessment provides comparison at three levels where the underlying data supports it:

  1. Overall: how a response sits within the broader HCI participant benchmark.
  2. Similar age: how selected behaviours compare within the participant’s age group.
  3. Similar AI use: how selected behaviours compare with people who report using AI at a similar frequency.

This matters because frequent use is not the same as dependence, and low use is not the same as strong agency. The pattern across trust, verification, reliance, delegation and retained judgement provides more context than any single answer.

Explore the wider benchmark context

Why the resulting human-behaviour data matters

The immediate purpose of the assessment is personal clarity. Its longer-term research value is the development of independent, structured evidence about how human behaviour changes as AI becomes part of ordinary life.

Much of the world’s most detailed AI-use data sits inside the companies that operate major AI systems. That information can show which features people use, how often they return and what tasks they ask a system to perform. It cannot, by itself, fully explain whether people feel more capable, less confident, more dependent, more discerning or less in control.

Relevant anonymised research responses contribute to HCI’s internal benchmark distributions, while new assessment responses may refine those distributions over time. Repeated measurement can help researchers examine how trust, verification, agency, reliance and emotional use change as AI adoption deepens. Aggregated findings may also help AI teams evaluate questions that task-performance benchmarks cannot answer: whether people continue to verify important outputs, experience meaningful control, recognise AI’s contribution and remain actively involved in decisions.

This does not mean participant-level assessment data is shared with developers or used to profile individuals. HCI does not publish the internal participant-level benchmark dataset. Personal reports remain private, and published findings use aggregated data under HCI’s Data Governance & Ethics approach.

A baseline, not a verdict

The HCI assessment is designed to make behaviour easier to see, not to prescribe one correct way to use AI. The meaning of reliance, trust or thought partnership can differ by person and context. The assessment therefore reports comparative patterns rather than labelling one result as automatically good, bad, healthy or unhealthy.

The value lies in seeing the combination. Participants receive a baseline they can examine, question and revisit as their AI use changes.

AI is becoming part of how people think, decide, create and interpret uncertainty. Before deciding what role it should play next, people need a clearer view of the role it is already playing.

Questions about the HCI AI Identity & Behaviour Assessment

What is the HCI AI Identity & Behaviour Assessment?

It is a 39-question, non-clinical self-report assessment that compares a person’s reported AI-use behaviour with HCI participant benchmarks across nine dimensions.

Who should take the assessment?

It is designed for adults who use AI for work, study, research, creativity, decisions or everyday tasks and want to understand how their behaviour compares with other participants.

How long does the assessment take?

It typically takes under 5 minutes to complete the 39 questions and the short participant-information section.

Is the assessment free?

Yes. The 39-question assessment and personalised results are free. An optional expanded AI Identity & Behaviour Report is available afterward for US$29. No payment is required to take the assessment or receive your free results. An email address is required to create the assessment record and display the results, but no account or password is required.

Is the assessment scientifically validated?

The assessment was developed from HCI’s research programme and benchmarked against participant data. It has not been clinically validated or independently validated as a diagnostic psychometric instrument. It is intended for behavioural comparison and personal reflection, not diagnosis.

How do I know if I’m relying on AI more than I realise?

Frequency alone cannot answer that question. The assessment examines reliance alongside trust, verification, decision delegation and human agency, then compares the reported pattern with HCI participant benchmarks.

Can it diagnose AI addiction or dependence?

No. It examines behaviours related to reliance and perceived dependence, but it is not a psychological, medical or mental-health assessment and does not provide a clinical diagnosis.

Does it benchmark the behaviour of AI models?

No. It assesses reported human behaviour around AI use. It does not test model accuracy, reasoning performance, safety or agent task completion.

Does it work if I use Claude, Gemini, Copilot or another AI assistant?

Yes. The assessment is platform-independent and examines behaviour across AI tools rather than requiring one particular assistant. The optional conversation-history review depends on what history the participant’s chosen AI service can access.

Does HCI need access to a user’s AI conversations?

No. The assessment uses structured self-report questions. It does not require chat histories, private documents or access to an AI account.

Is my assessment data private?

Personal reports are private and participant-level answers are not published or provided to customers. After information that directly identifies a participant has been removed, anonymised answers may contribute to HCI’s internal benchmark library and aggregated research findings.

Can I retake the assessment and track how my AI use changes?

You can retake the assessment and compare saved results as separate personal baselines. HCI does not currently provide an account-based dashboard that automatically tracks changes between assessments.

What does a participant receive?

Participants receive free personalised results showing selected areas of their reported behaviour against HCI benchmarks. An optional expanded report provides deeper analysis across the full assessment.

Further information: Read the HCI benchmark methodology, explore the HCI Dataset Library, or review HCI’s Data Governance & Ethics approach.