Card sorting case study

Card sorting government services: topic or task?

By UXbeam, information architecture tools and services since 2021 · Updated September 2026

Government digital teams have spent years reorganizing services around user needs instead of internal government structures. This study asks a direct question: can card sorting investigate the organizing principle itself, rather than only how participants group the cards?

In a 20-card open sort around major life transitions, one topic-based mental model dominated. When the study had 69 responses, UXbeam also detected a smaller action-based mental model. The two mental models did not just place cards differently; they reflected different classification rules.

At 69 responses, UXbeam found two mental models

Model A · 57 participants

The 57-person mental model organized services largely by topic or domain, with participant category names such as Housing, Education, Employment, Childcare, Care, and Financial support.

Model B · 9 participants

The 9-person mental model organized the same services much more strongly around actions, with category names such as Find, Apply, comparing and understanding options, and challenge decisions.

Three participants were treated as outliers in this result. The minority mental model was not just a different arrangement of topic categories. It organized the cards by a different rule. The topic model asks what is this service about?, while the action model asks what am I trying to do?

Averaged into one hierarchy, these two patterns can describe a compromise IA that neither group actually proposed.

What the two mental models were actually organizing around

A mental model tells us that a subset of participants organized the cards in a similar way, but it does not by itself tell us which classification scheme they were using. We identified the organizing principle from each participant-created category name and the cards inside it.

Education was coded as topic/domain. Find was coded as task/goal only when the cards inside it were also united by finding something.

The two analyses answer different questions:

Analysis	Question
Mental models	Which distinct mental models appear in the participant data?
Organizing principles	What classification rule is each mental model using?

At 69 responses, the two analyses aligned perfectly. Among the 61 participants with a codeable dominant principle, all 8 members of the smaller mental model were task/goal-dominant, while none of the 53 members of the dominant mental model were. The association was statistically significant (Fisher's exact test, p < 0.001).

Both analyses use the same card-sort sessions: mental-model detection uses placements, while organizing-principle coding uses category names and their contents. The result therefore measures agreement between two analyses of the same behavior, not evidence from an independent dataset.

Topic, task and audience are established IA organizing schemes

The distinction between organizing by topic, task or audience is not new. In Information Architecture for the World Wide Web, Louis Rosenfeld and Peter Morville described organization schemes such as topic, task, audience and metaphor, alongside hybrids that combine more than one schemeRosenfeld & MorvilleO'Reilly MediaInformation Architecture for the World Wide WebRosenfeld & Morville · 3rd edition, 2006 · oreilly.com.

The organizing principle determines the rule people use to decide what belongs together; an audience scheme, for instance, asks who the content is for. The same services can support several plausible schemes, and each can produce a materially different structure.

Government IA has already changed its organizing principle

First-generation government websites often mirrored government itself. Departments owned services, so departments became the navigation. That made sense from inside the organization, but it was much harder for citizens who did not know which department owned the thing they needed.

In the UK, this became a central problem in government digital reform. Martha Lane Fox's 2010 review of Directgov argued that people should not have to navigate the departmental structure of government before finding the service or content they need, and recommended a single government domainLane FoxCabinet OfficeDirectgov 2010 and Beyond: Revolution Not EvolutionMartha Lane Fox · November 2010 · gov.uk.

The following year, the team building the GOV.UK alpha put user needs onto cards. Literally: they wrote the top 100 needs onto physical cards, prioritized them, and sorted them into categories on the wall. GDS's 2011 project history still preserves the photographGDSGovernment Digital ServiceA GDS Story 2011Photo: Richard Pope · gds.blog.gov.uk.

Three boards, one shift in the unit of organization. The middle frame carries the 2011 GDS photograph of user needs cards in publication (credit Richard Pope; the GDS page publishes under the Open Government Licence v3.0 except where otherwise stated).

By October 2012, GOV.UK had launched. GDS described the new website as planned, written, organized and designed around what users needed to get done, rather than around the way government wanted people to do itGDSGovernment Digital ServiceGOV.UK: The start of a new way of delivering public servicesFrancis Maude · 16 October 2012 · gds.blog.gov.uk. The organizing principle had shifted away from internal government structure and toward user needs and tasks.

Victoria describes its public content as organized using a life-events approach, with its information architecture informed by how people experience information and services rather than by departmentvic.gov.auVictorian GovernmentUse a life events approach for digital contentvic.gov.au. A life event such as having a baby, buying a home or dealing with a death can cross several agencies, so the principle can affect more than navigation. It can expose gaps between services, show where journeys break across organizational boundaries, and create reasons to join previously separate services.

Twenty of 28 countries surveyed by the OECD in 2025 had adopted some form of life-event approach, and 13 had begun integrating some services for at least one life eventOECDOECDGovernment at a Glance 2025Seamless and accessible public administrative services · 2025 · oecd.org. In the lighter version, government gathers information and links in one place. In the integrated version, the life event becomes a single entry point backed by coordination between providers and data sharing across services.

Department, task, life event: these are not alternative labels for the same hierarchy. They can change which services are considered part of the same journey, who has to coordinate with whom, and what the digital service itself becomes. They are different answers to a more basic IA question:

Card sorting usually answers where the items belong

In modern UX practice, an open card sort usually starts with a content or feature inventory. The items become cards, participants group them and name the categories, and the researcher uses the category labels, similarity matrix and dendrogram to develop an information architecture. The method is naturally bottom-up: begin with the items, observe which ones participants place together, and work toward a structure.

That works well when the question is how should these items be grouped? But the cards themselves can already point toward one kind of grouping. Billing reads like a topic or object; Pay an overdue invoice reads like a task; When you move home reads like a life event.

Card-sorting guidance warns about this directly. Nielsen Norman Group recommends neutral wording because terminology can influence how participants group cardsNN/gNielsen Norman GroupCard Sorting: Pushing Users Beyond Terminology Matches2024 update of Jakob Nielsen's 2009 article · nngroup.com. If several labels repeat the same words or grammatical pattern, participants can match the terminology rather than the underlying concepts.

That matters if the research question is not yet “what should the sitemap be?” A team may first want to know whether people naturally think about the same services by topic, task, audience, life event, object, or another organizing principle. If the card set already strongly favors one of those schemes, a conventional content inventory may not be the best instrument for answering that question.

So we designed the card set differently.

From recurring IA question to a repeatable method

We had run into this question in earlier IA work, where the useful decision was sometimes the organizing principle rather than the final hierarchy. That pushed us to formalize a repeatable way to study the choice directly. This government-services study is a deliberately constructed example of that approach.

Designing a sort where several principles can compete

We created 20 cards covering government services around work, education, housing, care, disability, retirement and other life transitions. Examples included:

• Find financial help after losing your job
• Apply for funding to retrain for a new career
• Get childcare support while attending training
• Find accessible housing after developing a disability
• Compare education options after compulsory schooling
• Report a death and find out what must be done next

The cards were deliberately selected so several plausible classification cues recurred across the set. Participants could therefore privilege topic, task, life event, or another organizing principle in an otherwise normal open card sort:

Organizing principle	Examples available in the set
Topic / domain	housing, education, work, care
Task / goal	find, apply, compare, understand, challenge
Audience	parents, carers, older people
Life event / context	moving, losing a job, retiring
Object / entity	benefits, courses, qualifications

To participants, it was a normal open card sort. The difference was how the 20 cards were selected and how the results were analyzed.

We were not trying to make every card neutral or deliberately ambiguous. We designed the set so several plausible classification schemes had enough recurring evidence to compete.

When the organizing principle is the million-dollar question

Sometimes the sitemap is the deliverable. Sometimes the more consequential decision comes earlier: what should people experience as one service, and what should organize it?

The government examples above show why. A life event can cross agencies, turning an organizing decision into a decision about service boundaries, handoffs, and which systems must coordinate.

That is a much larger design question than deciding whether Benefits belongs above or below Employment. A public-service team may be deciding whether services currently owned by different departments should become one journey, where the handoffs should disappear, and which teams or systems need to coordinate behind the scenes.

The same decision appears in products. Imagine a work-management platform. Organizing it around objects might make Projects, Tasks, People and Reports the primary model. Organizing it around goals might foreground Plan, Review, Analyze and Configure. Organizing it around workflow stages might make Setup, Execute, Monitor and Review the backbone. Those are different product models, not cosmetic rearrangements of the same menu.

A multi-audience product can raise another high-stakes version of the question. If managers and contributors disagree about one or two placements, the team may need better labels or cross-links. If they consistently use different organizing principles, the implication can be much larger: different entry points, different navigation models, or different product structures for different audiences.

The principle can shape the service itself, or become a testable structure. Often both.

A useful result at this stage may be: most participants understand this domain as a set of topics, while a smaller mental model understands it as a set of actions. That finding can change service boundaries, product models, entry points and research priorities before a sitemap exists.

Each mental model can become a candidate sitemap

If the team needs navigation, the two mental models provide two explicit hypotheses: one IA organized primarily by topic or domain, another by task or action. UXbeam can open either mental model as an editable sitemap and send it to Tree Testing. The alternatives remain separate rather than being averaged into one hierarchy.

The mental-model result changed as responses accumulated

At 69 responses, UXbeam reported a 57-participant topic-oriented mental model and a 9-participant action-oriented mental model.

In the current production result, which contains 82 response rows, the action-oriented mental model is no longer reported separately, and the UI reports One mental model dominates.

Two results from the same study. Mental-model results can change as recruitment continues.

The earlier nine-person action-oriented mental model did not disappear because it became too small to display. As more responses arrived, its participants were absorbed into the dominant mental model.

The 69-response result is the relevant state for comparing the two detected mental models, because that is when UXbeam reported both. It also leaves a product question worth asking: should an earlier coherent mental model remain visible after later responses change the result?

What mental models and organizing principles add to card sorting analysis

Mental-model analysis separates participant patterns that would otherwise be blended in one aggregate dendrogram. Organizing-principle analysis identifies whether each mental model is organized by topic, task, audience, life event, object, or another classification scheme.

Method note

Sources

• Louis Rosenfeld & Peter Morville, Information Architecture for the World Wide Web, O'Reilly Media.
  Organization schemes include topic, task, audience, metaphor and hybrids.
• Martha Lane Fox, Directgov 2010 and Beyond: Revolution Not Evolution, Cabinet Office, 2010.
  The review argued that people should not need to navigate government's departmental structure to find services or content.
• Government Digital Service, “A GDS Story 2011.”
  Documents the GOV.UK alpha team's top 100 user needs being written on cards and sorted into categories on the wall. Photograph credited to Richard Pope. The page is published under the Open Government Licence v3.0 except where otherwise stated.
• Government Digital Service, “GOV.UK: The start of a new way of delivering public services,” 16 October 2012.
• Victorian Government, “Use a life events approach for digital content.”
  Explains how vic.gov.au uses life events to organize public content and how the approach can expose gaps or opportunities to join services.
• OECD, Government at a Glance 2025, “Seamless and accessible public administrative services.”
  Reports life-event adoption in 20 of 28 surveyed countries, with 13 beginning to integrate services for at least one life event.
• Nielsen Norman Group, “Card Sorting: Pushing Users Beyond Terminology Matches.”
  2024 update of Jakob Nielsen's 2009 guidance. Card-label phrasing can influence grouping; neutral labels reduce keyword matching.

Study data: UXbeam card sorting sessions shown on this page.