Mental model

On this page 7

A mental model is the simplified internal representation a person builds of how a system works, which they use to predict outcomes and decide what to do next. In UX design, the term describes the gap or fit between what users believe a product will do and what the product actually does, and that fit largely determines whether an interface feels intuitive or confusing.

Mental models are incomplete, unstable and partly wrong by nature. People form them from prior experience, analogies and visible cues, not from documentation. The task of interaction design is therefore not to teach users the true architecture of a system but to present a conceptual model that maps cleanly onto beliefs users already hold or can acquire quickly.

How users form a mental model

Users form a mental model by generalizing from past experience and adjusting it as the interface responds to their actions. The cognitive scientist Kenneth Craik introduced the idea in 1943, and Donald Norman later brought it into interface design by separating the designer’s model, the system image and the user’s model.

The designer cannot speak to users directly. Everything users learn arrives through the system image: labels, layout, icons, feedback, error messages, onboarding and even marketing copy. When the system image is ambiguous, users fill the gaps with assumptions, and those assumptions harden into persistent misconceptions that are difficult to correct later.

The following table separates the three models that interact in every product and shows who owns each one.

ModelHeld byWhat it containsTypical risk
Designer’s modelDesign and product teamThe intended concept and its rulesAssuming users share internal knowledge
System imageThe product itselfInterface, feedback, content and documentationCues that suggest the wrong concept
User’s modelEach individual userBeliefs about cause and effectGaps filled with incorrect analogies
The three models in Norman’s framework

Formation follows a trial, feedback and revision loop. Each action produces a result that either confirms the current belief or forces an update. Immediate, legible feedback accelerates accurate learning, while delayed or silent responses leave users guessing and encourage superstitious behavior, such as clicking a button several times because nothing appeared to happen.

Properties of mental models in practice

Mental models in practice are functional rather than accurate: users keep a belief as long as it produces acceptable results, even when it is technically false. Many people believe that closing an app saves battery or that a pressed elevator button calls the car faster, and these beliefs survive because they rarely cause visible harm.

The table below lists the properties that researchers repeatedly observe and the design implication of each one.

PropertyWhat it meansDesign implication
IncompleteUsers model only what they needExpose detail progressively
UnstableUnused details fade over timeKeep core patterns consistent
BorrowedUsers transfer models from other productsFollow established conventions
ParsimoniousUsers prefer the simplest explanationAvoid hidden modes and exceptions
IndividualModels vary with experience and cultureResearch several user segments
Observed properties and their design implications

Because models are borrowed across products, conventions carry great weight. Jakob’s law states that users spend most of their time on other websites and expect yours to work the same way. A shopping cart icon, a magnifying glass for search or a trash can for deletion work because they reuse a model users already hold, which lowers cognitive load during the first visit.

A mental model differs from neighboring concepts because it lives inside the user’s head and can only be inferred, never directly observed. Related artifacts such as a persona or a user journey are team-made documents that summarize research, whereas the mental model is the belief those documents try to capture.

ConceptWhere it existsMain question it answers
Mental modelIn the user’s mindWhat does the user believe will happen?
Conceptual modelIn the design specificationWhat concept should the product present?
Information architectureIn the product structureWhere does content live and how is it labeled?
User journeyIn a research artifactWhat steps and emotions occur over time?
PersonaIn a research artifactWho is the archetypal user and what do they need?
How mental models relate to neighboring concepts

The most important distinction is between the mental model and the conceptual model. The conceptual model is a deliberate design decision, for example presenting cloud storage as folders rather than as tagged objects. A good conceptual model is chosen because it matches users’ existing mental models closely enough to be learned with minimal effort.

Methods for uncovering mental models

Teams uncover mental models through generative and evaluative user research that asks people to explain, sort or predict rather than simply to state preferences. Users rarely articulate their models directly, so researchers infer them from language, groupings and errors.

MethodWhat it revealsBest stage
Contextual interviewsVocabulary and real-world workflowsEarly discovery
Card sortingHow users group and name contentBefore structuring navigation
Tree testingWhether a hierarchy matches expectationsAfter drafting the structure
Think-aloud usability testingPredictions, surprises and recovery attemptsPrototype and live product
Mental model diagramsTasks and beliefs aligned against featuresStrategy and roadmap planning
Research methods and what they reveal

In think-aloud sessions, the most valuable signal is a failed prediction. When a participant says “I expected this to save automatically” and it did not, the researcher has found a precise point of mismatch. Indi Young’s mental model diagram extends this approach by mapping clusters of user intentions above a line and the supporting product features below it, which exposes unsupported needs at a glance.

Designing for mismatched mental models

When a user’s model and the product’s behavior disagree, designers have two options: adapt the product to the model or adapt the model to the product. Adapting the product is usually cheaper for users and is the default recommendation, while reshaping the model is justified only when the new concept delivers a clear advantage.

SituationRecommended strategyReason
Users hold a strong, common conventionMatch the conventionRelearning costs exceed any novelty benefit
The new concept is genuinely superiorTeach with onboarding and feedbackShort guidance creates a durable new model
Expert and novice models differLayer the interfaceDefaults serve novices, shortcuts serve experts
Users hold a harmful false beliefMake system state visibleVisible status corrects the belief at the moment of error
Choosing a strategy for a mismatch

Common mistakes include naming features after internal teams, introducing hidden modes that change what an action does, and redesigning a familiar layout without a transition path. Each of these breaks a working model and produces errors that users attribute to their own incompetence, which damages trust more than an obvious bug would.

Explained without jargon

Everyone carries a small picture in their head of how things work. You do not know how a thermostat is wired, but you believe that turning it up makes the room warmer. That picture is a mental model, and it lets you act without reading a manual.

Apps and websites work the same way. People arrive with expectations from other products. When a design behaves the way people expect, it feels easy. When it surprises them, they feel lost, even if the design is clever. Good designers study those expectations first and build around them.

A typical use case

A team redesigning a hospital’s patient portal noticed that many users never found their test results. In think-aloud sessions with twelve patients, most looked under “Messages”, because they believed results arrive the way a doctor’s letter does. The portal, however, stored them under a section called “Health record”, which reflected the database structure rather than how patients thought about it.

The team did not try to retrain patients. Instead, they sent a notification message with a direct link whenever a new result became available and renamed the section “Test results and records”. In a follow-up round of testing, all twelve participants located a result in under one minute, and support calls about missing results dropped noticeably over the next quarter. The fix worked because it matched the existing mental model instead of fighting it.