Iron Goo
Clinic staff sort paper cards on a wall board beside a reception desk after hours

Website Information Architecture: A Clinic Teardown

Atamyrat Hangeldiyev
Atamyrat Hangeldiyev
Systems Architect
March 7, 2026
On this page
UX

You run a clinic where the front desk answers about 40 calls a week asking whether you treat a problem, whether a referral is needed, and what a first visit costs. Every answer is already somewhere on the site. Website information architecture is how you group, label, and link pages so people, crawlers, and browser agents can find the one thing they need. A menu organized by treatment technique can hide useful pages from patients who arrive with a problem.

The clinic is Northfield Physio, a fictional practice with two locations, 11 clinicians, and 64 indexed pages. Every number assigned to Northfield in this example is hypothetical. Its current menu reads Home, About Us, Our Team, Services, Resources, Blog, and Contact. The labels look orderly from the reception desk; the calls suggest they are less helpful to someone trying to book.

The documentation cited here was read on 29 October 2026. The starting figures describe a scenario you can adapt to your own site, not a measured clinic or a forecast.

Stage 1: Inventory the pages and find the missing routes

Start with a page inventory, even if you expect to replace the menu. A navigation label can point to a group of pages only after you know which pages exist, what they answer, and who can decide whether their content is accurate.

Northfield's 64 pages fall into these groups:

Page typePagesCurrent location or role
Treatment technique pages18Under Services
Condition posts9In the blog
Location pages2Linked from Contact
Clinician biographies11Under Our Team
Other posts21In the blog
Utility pages3Contact, privacy, and a PDF price list
Total64

The Services pages use names such as Manual therapy, Dry needling, Shockwave, and Hydrotherapy. The condition content sits among blog posts, the price list is a PDF under Resources, and the booking link appears only on Contact. The old Blog index lists posts by date only, so reaching a post through it can require paging through the list. A patient with knee pain must guess which technique might apply or page through the blog before finding a useful choice.

Use the following inventory columns as a template. The rows are a representative sample, and inbound links are counts of links from other Northfield pages.

URLPage typeIn main navInbound internal linksOwnerDecision
/services/manual-therapy/TechniqueUnder Services8Clinical leadKeep
/services/dry-needling/TechniqueUnder Services7Clinical leadKeep
/services/shockwave/TechniqueUnder Services5Clinical leadRewrite
/services/hydrotherapy/TechniqueUnder Services4Clinical leadKeep
/blog/running-knee-pain/Condition postVia Blog index (by date)1Clinical leadRewrite
/blog/sciatica-advice/Condition postVia Blog index (by date)2Clinical leadRewrite
/resources/price-list.pdfUtilityUnder Resources1Practice managerMerge
/contact/UtilityUnder Contact12Practice managerKeep
/team/clinician-one/BiographyUnder Our Team2Practice managerKeep
/blog/old-clinic-news/Other postVia Blog index (by date)1Practice managerCut

“In main nav” includes pages reachable through a menu branch; it does not mean that every page has its own header link. A rewrite keeps the subject but changes the page to answer a clearer need. A merge moves useful information into a maintained destination, such as an HTML prices page, while a cut requires checking whether the page has a useful replacement.

Count links as well as pages. In the 30-page model in the internal linking architecture guide, removing a service page from the header produced the largest loss of modeled link flow short of orphaning it. That model is illustrative, not a prediction for Northfield. Google's recommended minimum is that every page you care about has a link from at least one other page on your site.

A page does not need a header link to be useful or crawlable. It does need a route that makes sense: a condition index can link to knee pain, a knee pain page can link to runner's knee, and relevant treatment pages can link back to the conditions they address. Check those links in the inventory before deciding that a page has disappeared merely because it is absent from the header.

Stage 2: Sort cards to hear patients' labels

A card sort asks participants to group content topics. In an open sort, patients create and name their own groups; in a closed sort, they place cards into predefined groups. A hybrid sort supplies some groups and lets participants add others, though the supplied names can influence the result.

Northfield chooses an open sort because it needs patients' vocabulary. The clinic already knows its own categories. To evaluate whether proposed categories work for a task, Nielsen Norman Group recommends tree testing over a closed card sort.

The clinic prepares 40 cards, each describing one topic a patient might seek. Examples include “Runner's knee,” “First appointment price,” “Do I need a referral?” and “Dry needling.” The cards represent topics, not a promise to create 40 separate menu items.

30 to 50
Cards in a card sort
15
Minimum qualitative card-sort participants
50+
Participants per tree for a quantitative comparison
80%
A useful task success guide

Recruit 15 recent patients for the qualitative sort and ask each person to group the cards in a way that makes sense to them, then name each group. Do not explain where the clinic currently files a topic. Record the groups, their names, and cards a participant finds difficult to place.

Patient groups topic cards in his own words while a facilitator takes notes

The patient supplies the group names while the facilitator records the choices. The clinic's existing menu does not set the categories.

Northfield's hypothetical sort produces patterns like these:

CardsGroups patients formCurrent clinic groupingDesign consequence
Knee pain, back pain, sciaticaBody part or problemBlog postsCreate a Conditions we treat branch
First visit, referral, price, insuranceStarting care or costResources and ContactGive costs and referral answers a visible route
High Street, second location, parkingLocation or getting thereContactCreate a Locations branch
Dry needling, manual therapy, shockwaveTreatments, sometimes associated with a problemServicesKeep a Treatments branch and link it to conditions

The sort suggests labels; it does not prove that a menu will work. A group named “Starting care” may sound plausible in a workshop and still fail when someone searches for a referral rule. The next test gives people a task and asks where they would go.

Five participants are not a card-sort sample

The familiar five-user rule concerns qualitative usability testing with comparable users. It does not establish a sufficient sample for card sorting.

Nielsen's 2004 card-sort analysis compared smaller samples with a 168-user result. Five users correlated at 0.75 with the full result, 15 at 0.90, and 30 at 0.95. Nielsen recommended 15 users, or 30 for large, well-funded projects; current NN/g guidance recommends at least 15 for qualitative card sorting and at least 30 to 50 for quantitative sorting.

The arithmetic behind the separate five-user usability discussion uses a typical problem discovery probability of 31 percent: 1 - 0.69^5 = 0.8436, or about 84 to 85 percent. It estimates discovery of usability problems under those assumptions. It does not measure how closely five people's card groups match a larger card-sort result.

Stage 3: Test whether patients can find six answers

A tree test presents a text-only hierarchy and asks participants to find a destination for a task. It tests the organization and labels without relying on page layout, illustrations, or the appearance of the menu.

For the comparison, Northfield writes the same six tasks for its old tree and a proposed tree. It defines correct destinations before collecting responses, then recruits 50 or more participants per tree. NN/g's tree-test guidance gives 50 or more per tree as a typical quantitative sample for reasonably narrow confidence intervals when comparing trees.

Success is the percentage of participants who reach a correct destination, including those who backtrack. Directness is the percentage who reach their chosen destination without backtracking or trying another category. NN/g also notes that the first click often predicts whether someone will eventually find the right item; keep the first branch in the results, not just the final answer.

Northfield's hypothetical comparison looks like this:

TaskOld successOld directnessNew successNew directness
Find whether a referral is needed38%22%86%71%
Find the first-visit price44%30%91%80%
Find help for a running knee injury52%25%84%68%
Find a clinic's parking details58%42%88%76%
Find a physio who treats back pain62%39%83%70%
Find information about dry needling72%60%92%84%

For the running knee task, the old tree's 52 percent success and 25 percent directness could suggest a signposting problem: some people eventually reach the answer after trying another branch. A low-success, high-directness result can suggest that a label sends people confidently to the wrong destination. Both are interpretations to check against first-click paths and participant comments; the percentages alone do not establish the cause.

Use a benchmark to decide where to investigate, then consider the task's consequence. NN/g cites Bill Albert and Tom Tullis's review of 98 studies, which had a median task success rate of 62 percent and an interquartile range of 37 to 83 percent.

Task successAlbert and Tullis rubric cited by NN/g
Under 40%Poor
41 to 60%Fair
61 to 80%Good
80 to 90%Very good
Over 90%Excellent

The published bands overlap at 80 percent and leave the interval between 40 and 41 percent unspecified. Treat them as a rough rubric, not a boundary that changes a decision by itself. NN/g suggests aiming above 90 percent for mission-critical tasks; Optimal Workshop's separate guidance calls around 80 percent or more a good task success score.

Northfield should still inspect the new referral result. At 86 percent success and 71 percent directness, it clears the general 80 percent guide, but people who cannot establish whether they can book may abandon a time-sensitive task. The next revision can test the wording and placement of the referral answer.

Stage 4: Replace technique-led navigation with patient labels

The proposed top level is Conditions we treat, Treatments, Prices and insurance, Locations, and Our physios, followed by a Book an appointment button. Conditions gives patients a route from the problem they can name. Treatments remains available for people who already know the technique they want.

Old itemWhat was under itNew item or destinationEvidence for the change
HomeGeneral introductionLogo link to homePreserve a route home without spending a menu label on a content category
About UsPractice informationAbout links within relevant pagesPatients grouped practical topics around care and location
Our TeamClinician biographiesOur physiosUses the sort's people-focused grouping
ServicesTechnique pagesTreatmentsKeeps recognizable techniques together
BlogCondition and other postsConditions we treat; articles linked from relevant pagesPatients grouped condition cards by problem or body part
ResourcesPDF price list and other materialPrices and insuranceCost was a call topic and first-visit price performed poorly in the old tree
ContactBooking and location detailsLocations; Book an appointmentLocation cards formed a group, and booking needs a visible route
Clinic menu shifts from technique branches and isolated pages to a shallow problem-organized tree

The old tree puts techniques at the front and leaves condition pages apart from the main route. The proposed tree gives patients a problem-led branch and connects it to treatment information.

The conditions branch can begin with recognizable body areas or problems, then lead to a condition page. A runner's knee page can describe the problem and link to relevant treatment information; a dry needling page can link to conditions for which a clinician has judged that information relevant. Clinical review owns those relationships.

Keep the top-level order consistent across pages. WCAG 2.2 Success Criterion 3.2.3, Consistent Navigation, is Level AA and requires repeated navigation mechanisms to occur in the same relative order each time they are repeated, unless a user initiates the change. The rule allows, for example, a user-controlled menu state; it does not require every page to repeat every possible link.

The appointment button solves the route to the start of booking. The conversion path design guide addresses what happens after a visitor chooses it, including the information and actions needed to complete the booking.

Stage 5: Map changed URLs and match breadcrumbs to the visible trail

A hierarchy needs destinations that people can read and maintain. Northfield maps every changed URL before launch, assigns a permanent redirect from each old address, and checks that internal links point to the new address. The table shows representative mappings, not a complete redirect register.

Old URLNew URLAction
/blog/running-knee-pain//conditions/knee-pain/runners-knee/301 redirect
/blog/sciatica-advice//conditions/sciatica/301 redirect
/services/dry-needling//treatments/dry-needling/301 redirect
/resources/price-list.pdf/prices/Move maintained price information to HTML; 301 redirect
/contact/high-street//locations/northfield-high-street/301 redirect

Use lowercase, readable words, and hyphens in the new paths. Google's URL guidance notes that paths are case sensitive and recommends hyphens over underscores. Avoid fragments for changing page content: a fragment is not a substitute for a distinct content URL.

On the runner's knee page, the visible trail is Conditions > Knee pain > Runner's knee. It describes a typical route through the content. A breadcrumb trail need not reproduce every URL folder, and a home page need not be included.

The structured data must describe the same visible trail. In this example, https://www.example.com is a placeholder domain, and the final item omits item:

{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Conditions",
      "item": "https://www.example.com/conditions/"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "Knee pain",
      "item": "https://www.example.com/conditions/knee-pain/"
    },
    {
      "@type": "ListItem",
      "position": 3,
      "name": "Runner's knee"
    }
  ]
}

Google's breadcrumb documentation requires at least two ListItem entries and permits the last item to omit its URL. Check the rendered page and the JSON-LD together when a title or parent changes. The structured data guide has the fuller BreadcrumbList reference and other markup.

Google removed breadcrumb trails from mobile search results on 23 January 2025; desktop support and the Search Console report remain. SearchPilot reported a statistically significant 5 percent organic traffic uplift in a test of visible mobile breadcrumbs plus server-side schema. Schema alone was inconclusive, so that result does not isolate a schema-only benefit.

A site rebuild such as Foundation can ship the new templates, URLs, redirects, and breadcrumbs with structured data wired into the page templates. Before launch, check each redirect and compare the visible trail with the data on the destination page.

Stage 6: Use site search to check the vocabulary

A search box gives visitors another route when a category is unfamiliar. Its terms also give Northfield a useful list of words to compare with the card sort, although a typed query alone cannot explain why someone searched or whether they found an answer.

Hypothetical search termPage or label to checkFollow-up
referralReferral answer linked from Prices and insurance and bookingConfirm the answer states whether a referral is needed
pricePrices and insurancePut first-visit prices in readable HTML
sciaticaConditions we treat > SciaticaUse the patient term in the page label
parkingRelevant location pageInclude parking details on each location page
knee painConditions we treat > Knee painLink to runner's knee and related conditions
insurancePrices and insuranceExplain what a patient should check before booking

In GA4 enhanced measurement, the view_search_results event records a search term when a results page loads with a recognized query parameter. The default parameter names are q, s, search, query, and keyword; the event's search_term parameter supplies the Search term dimension. Check the results URL and the event in your implementation before relying on a search terms report.

An instant search that filters content in place and never loads a results URL would not trigger that URL-based enhanced measurement behavior. That is an inference from the documented trigger, so test the actual implementation and arrange suitable event collection if needed. Search terms are also only one source of label ideas: compare them with the sort and with what the front desk records.

WCAG 2.2 Success Criterion 2.4.5, Multiple Ways, is Level AA. It requires more than one way to locate a page within a set of pages, with an exception for pages that are steps in a process. Its sufficient techniques include a search function and links to related pages, so Northfield can provide both a search route and relevant links between condition and treatment pages.

Stage 7: Give agents and crawlers the same navigable structure

People can see a menu; software needs links and recognizable structure in the page it receives. Northfield uses named navigation landmarks for the main menu and breadcrumb trail, with ordinary links inside lists.

<nav aria-label="Main">
  <ul>
    <li><a href="/conditions/">Conditions we treat</a></li>
    <li><a href="/treatments/">Treatments</a></li>
    <li><a href="/prices/">Prices and insurance</a></li>
    <li><a href="/locations/">Locations</a></li>
    <li><a href="/physios/">Our physios</a></li>
    <li><a href="/book/">Book an appointment</a></li>
  </ul>
</nav>

<nav aria-label="Breadcrumb">
  <ol>
    <li><a href="/conditions/">Conditions</a></li>
    <li><a href="/conditions/knee-pain/">Knee pain</a></li>
    <li>
      <a href="/conditions/knee-pain/runners-knee/" aria-current="page">
        Runner's knee
      </a>
    </li>
  </ol>
</nav>

The HTML nav element defines a navigation landmark. When a page has more than one navigation landmark with different link sets, give each a unique label. The breadcrumb pattern places the trail in a labelled navigation landmark and uses aria-current="page" on the current-page link.

Reader or viewWhat to check on Northfield's page
Person using the menuLabels match the task and the order stays consistent
Assistive technologyMain and Breadcrumb landmarks have distinct names; the current breadcrumb is identified
Search crawlerDestination links are <a> elements with href attributes
Browser agentThe rendered page, DOM, and accessibility tree expose a coherent route

Google's AI optimization guidance says browser agents may analyze visual renderings, inspect the DOM, and interpret the accessibility tree. It does not say every agent uses every view. The UX and AI interface guide explains how those views relate to an agent using a page.

Vercel reported on 17 December 2024 that none of the major AI crawlers rendered JavaScript at that time. Our inference is that a menu added only by client-side JavaScript would be absent from the initial HTML those crawlers receive; clickable div elements also lack the ordinary link structure. Google says it can reliably crawl links as a elements with an href, so inspect the server-delivered HTML for the actual destinations.

Stage 8: Measure the fixed state after eight weeks

Eight weeks after the hypothetical launch, Northfield checks whether the new routes correspond to useful outcomes. The figures below are illustrative, including the tree-test results and the counts of pages without a route from the navigation.

Hypothetical measureBeforeEight weeks after
Weekly calls about treatment availability, referral, or first-visit costAbout 4022
Online bookings per week3147
Tree-test task successAll six tasks below 80%Every tested task above 80%
Condition pages in a problem-led branch0 of 99 of 9
Condition pages with a visible breadcrumb0 of 99 of 9

A route from the main navigation can pass through an index page; it does not require 64 links in the header. Check the route with actual links, and retain the inventory's inbound-link count so that a technically reachable page does not depend on one obscure connection.

Re-measure the six tasks after the next material menu change, retaining success, directness, and first-click paths. Review the three call categories separately, inspect search terms and their destinations, and verify redirects and breadcrumb trails when pages move. If referral questions remain common, test whether people can find and understand the answer before adding another menu item.

Free ebook

Put your next steps in order

Build on what you have just read with a free ebook on semantic SEO and AI visibility. Get worked examples, decision rules, checklists and a do-this-week action in every chapter, plus a six-week plan for your site.

By signing up and confirming your address, you consent to receive the free ebook plus one email every two weeks with 3 or 4 new Iron Goo videos, guides and articles, and nothing else; unsubscribe any time with one click.

Related in UX

Ready to move?

Tell us where your business is today. A written assessment lands in your inbox within two business days.

Talk to us