AI Glossary / Governance, ethics and safety / Advanced

User-centered design

Definition

User-centered design is an iterative approach to creating systems around an explicit understanding of the people who will use them, their goals, and their working conditions. Research informs requirements, designs are evaluated with relevant evidence, and findings guide revisions. For AI products, this includes deciding how automation should fit the human task.

What it means

The starting question is what people need to accomplish. “Add an AI assistant” is a proposed solution, not a user need. A more useful starting point might be helping staff locate current instructions and produce accurate responses within their available time.
User-centered design extends beyond the appearance of an interface. It can affect the division of work between people and software, the information available for checking outputs, recovery from mistakes, and the support needed to use the service.
The approach is closely associated with human-centered design. Terminology varies, but neither label should excuse ignoring people indirectly affected by a product. User convenience is one consideration alongside other relevant responsibilities and constraints.

How it works

Understand the context through appropriate research, such as observation, interviews, and analysis of existing work. Identify relevant differences in experience, abilities, equipment, and circumstances. A convenient group of colleagues may not represent the intended users.
Translate findings into requirements that explain the outcomes the design should support. Produce candidate solutions, often beginning with prototypes. Evaluate them against the requirements and use the evidence to revise both the design and, where necessary, the understanding of the problem.
This is a repeated process rather than a single interview followed by implementation. Different disciplines contribute different evidence. Involving people with disabilities throughout development can reveal needs that might otherwise be missed, while standards-based evaluation provides complementary checks.

Three practical examples

• Research for an AI search tool shows that users need to distinguish current guidance from archived material. The team treats identifying the applicable version as a requirement, then tests ways of presenting that information.
• Staff using a drafting assistant often work on small screens. A design that is convenient on a large monitor leaves important review information difficult to access. The team evaluates the actual devices and workflow before choosing a layout.
• Users want to recover from an unwanted AI edit without losing their own changes. The team prototypes recovery options and observes whether users can understand and use them, rather than assuming that an undo icon is sufficient.

Think of it this way

Think of designing a workspace after observing the work people do there. The layout follows the tasks, constraints, and people rather than an attractive arrangement chosen in isolation. User-centered design applies that principle repeatedly as evidence reveals what works and what needs adjustment.

Worked example

Consider a constructed design review for an AI-assisted reference tool. The team defines a task: find the currently applicable instruction and identify its supporting document without facilitator help.
An initial evaluation includes 20 desktop users. Eighteen succeed, producing an observed success rate of 18/20 = 90%. The team later recognizes that mobile use is also part of the intended context and evaluates 20 mobile users. Ten succeed, or 50%.
Across these 40 participants, success is (18 + 10)/40 = 70%. More importantly, the separate results identify a context the initial evaluation failed to cover. The combined percentage does not explain the cause of the mobile difficulties.
The team investigates the observed barriers and revises the design. It then evaluates the relevant task again under specified conditions. User-centered design lies in this connection between context, requirements, evidence, and revision; calculating a percentage alone is not the process.
These invented study results are descriptive. They do not estimate all users' performance or establish a causal effect of device type without further evidence.

Illustrative case study

Fictional company Willow Reference plans an AI assistant for staff who answer internal questions. The first concept focuses on generating a polished response immediately. Interviews and observation reveal that staff's main difficulty is deciding which instruction is current and applicable.
The team reframes the requirement around locating and verifying guidance. It prototypes an interaction that makes document dates, scope, and supporting passages easier to inspect before drafting. Staff try realistic tasks, including conflicting and outdated documents.
Findings lead to further changes in the presentation and review process. The team also considers users who navigate with keyboards or assistive technologies.
This fictional case illustrates how research can change the proposed solution. It does not claim that interviews alone reveal every need or that the resulting prototype is proven suitable before evaluation.

Common misconceptions and mistakes

“User-centered design means building every requested feature.” Requests are evidence to understand, not automatically the best solution.
“One survey completes the process.” Design decisions need continuing research and evaluation appropriate to the project.
“Usability testing at the end is enough.” Earlier assumptions can shape the entire product and become costly to change.
“An average user represents everyone.” Important differences can disappear in an imagined average.
“User convenience overrides every other concern.” Design must also account for affected people and relevant constraints.

Business use

Connect research findings to explicit requirements and design decisions. Record why a feature exists, what user outcome it supports, and what evidence would show that it works. This helps distinguish a demonstrated need from an attractive technical possibility.
Involve relevant users early enough for their input to change the product. Include the surrounding workflow, training, support, and recovery processes in the design scope. For AI services, test how people handle uncertain or incorrect outputs as well as ideal demonstrations. Revisit assumptions when the user population, task, or system behavior changes.

People also ask

Is user-centered design only for interface designers?

No. Product, research, engineering, operational, and domain expertise can all contribute.

Does it require a finished product?

No. Prototypes can help test ideas before substantial implementation.

Is it the same as usability testing?

No. Testing is one evaluation activity within a broader design approach.

Check your understanding

Why was the desktop-only evaluation incomplete?

Answer: Mobile users were part of the intended context but were not initially represented.

What distinguishes a user need from a proposed feature?

Answer: The need describes an outcome or problem; the feature is one possible solution.

Why retain the reasoning behind a requirement?

Answer: Later teams need to understand the evidence and conditions on which it depends.

Related terms

• Usability
• Usability testing
• Context-of-use
• Human-centered design
• accessibility
• task analysis
• prototyping
• user requirements

Learn next

Choose one proposed AI feature. Identify the user outcome it is supposed to support, gather evidence about the current task, and describe a simple prototype test that could challenge the proposed solution.

More in Governance, ethics and safety

• Uncertainty Bias - Uncertainty bias, in the supplied Artificial Intelligence Ontology sense, occurs when a predictive system favors groups represented more strongly in its...
• Use And Interpretation Bias - Use and interpretation bias, in the supplied ontology's interpretive sense, is a systematic distortion in how ambiguous information, situations, or events...
• User Interaction Bias - User interaction bias arises when people's existing biases or patterned behavior shape their interaction with data, AI outputs, or results.
• Value alignment - Value alignment concerns whether an AI system's objectives and behavior correspond to the human values, interests, and intentions it is meant to serve.
• Value sensitive design - Value sensitive design is an approach to technology design that systematically considers human values throughout the design process.
• Variable - A variable is a symbol or named quantity that can take values from a specified set, called its domain.

Sources

NIST, Human Centered Design, archived overview of principles and activities: https://www.nist.gov/itl/iad/human-centered-technologies/human-factors-human-centered-design NIST SP 1270, human-centered design and context of use: https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.1270.pdf W3C Web Accessibility Initiative, Involving Users in Web Projects for Better, Easier Accessibility: https://www.w3.org/WAI/planning/involving-users/

Previous: Use And Interpretation Bias Next: User-defined function