Skip to content
UXE Award

Information architecture that scores: labels, depth and the way back

Information architecture is one of the seven UXE criteria. What editors test — labels people would use themselves, depth that matches the task, search that works, and a way back from anywhere.

The UXE editors5 min read

Information architecture is the criterion teams score worst on without realising it, because IA problems don't look like problems. They look like a product that's "a bit hard to learn", a support queue full of "where do I find…", and a search box doing the work navigation should have done.

Here's what editors test when they score it.

1. Labels people would use themselves

The most common IA failure is naming things after the system rather than the task. Entities, Resources, Modules, Workspaces — words that mean something specific to the team that built the product and nothing to anyone else.

Labels, translated
Compared onThe user's wordThe system's word
A list of peoplePeopleEntities
Things you're paid forInvoicesBilling objects
Saved searchesViewsQuery configurations
Where work happensProjectsWorkspaces
The binDeletedArchive (which also means something else)
The test: could a new user guess what's behind the label before clicking? If two labels could plausibly hold the same thing, one of them is wrong.

2. Depth that matches the task, not the org chart

Navigation often mirrors how a company is organised, which is invisible to a user and irrelevant to their job. The better arrangement follows the frequency of tasks: the thing people do daily is one click away; the thing they do quarterly can be three.

  • Count the clicks to your three most common tasks. If any is more than two from the home screen, that's the finding.
  • Beware the level that exists only to hold one item. A category with a single child is an extra screen for nothing.
  • Beware the flat structure with thirty siblings. Seven is comfortable; thirty is a list you scan, not a menu you use.
  • Watch for the same destination appearing in two places with two names — editors notice, and so do users, who then never trust either.

3. A way back from everywhere

Orientation is the quiet half of IA. At any point, the interface should answer three questions without being asked: where am I, how did I get here, and how do I get back?

The four orientation questions
  1. 01

    Where am I

    The current location marked in the navigation, not only in the page title.

  2. 02

    What is this part of

    Breadcrumbs, or a visible parent, for anything more than one level deep.

  3. 03

    How do I go back

    Browser back must work — including after a modal, a filter or a tab change.

  4. 04

    Can I link to it

    A URL that restores the state. If a view can't be shared, it doesn't exist for teams.

That last one is worth dwelling on. In a product used by teams, a filtered view that can't be sent to a colleague isn't a view; it's a private arrangement that has to be described in a message. Editors test it by copying the URL and opening it in a new window.

4. Search that does its job

Search is IA's safety net, and in most products it's an afterthought that matches on exact strings. What's tested: a misspelling, a partial word, a synonym, an ID, and a query that returns nothing.

The no-results state is where the score is decided. Keeping the query visible, suggesting the nearest match, and offering a path (browse the category, ask for help, create the thing) turns a dead end into a route. Most products show an empty page and a magnifying glass.

5. Evidence that the structure was tested

IA is the criterion where evidence is most persuasive, because the alternative is an argument about taste. Two methods carry real weight in an entry:

  • Tree testing — give people the labels with no interface and ask them to find things. A success rate before and after a restructure is the strongest IA evidence there is.
  • Card sorting — how your users group things, which is rarely how your database does.

Search logs are a close third: the queries people type are a list of everything your navigation failed to surface. A team that can show the top ten queries and what they changed as a result has demonstrated the criterion better than any diagram.

What to send

A navigation map or sitemap, the search results page for a real query, the no-results state, and a screen three levels deep so we can see how you handle orientation. If you restructured something, send the before and after with whatever you measured.

IA is weighted most heavily in Web Experience, Service Experience and UX Research & Strategy, but it's scored everywhere. Enter a project.

Growth: how good IA goes bad

Almost no product launches with poor information architecture. It arrives over three years, one reasonable decision at a time: a feature that didn't fit anywhere gets its own nav item, a section gets renamed for a campaign, an acquisition brings its own vocabulary.

How structure decays
  1. Year 1

    Five clear sections

    Designed together, named consistently.

  2. Year 2

    Seven, plus a "More"

    Two features had nowhere to go.

  3. Year 3

    Nine, two of them overlapping

    Nobody is sure whether Reports or Insights holds the export.

  4. Year 4

    Search carries the product

    Which is the finding, not the fix.

The defence is a standing rule: nothing gets a new top-level home without something else losing one. It's an unpopular rule and it is the only one that works.

URLs are information architecture

For anything on the web, the URL is the most durable expression of your structure — it is what gets shared, bookmarked, indexed and pasted into a ticket. Readable, stable, hierarchical URLs are an IA decision that also happens to be the foundation of search visibility.

  • Words rather than identifiers, where the thing has a name.
  • Hierarchy that matches the navigation, so a person can delete the last segment and land somewhere sensible.
  • Stable over time, with redirects when something must move.
  • State in the URL: filters, tabs and sort orders that survive a reload and can be shared.

Questions we get asked

  • How many top-level navigation items is too many?

    Seven is comfortable, nine is a stretch, and beyond that people scan rather than choose. If you're over, the answer is usually grouping — or removing something that no longer earns its place.

  • Is a mega menu a sign of bad IA?

    Not necessarily. A mega menu that reveals a clear structure helps; one that lists forty links because nobody could decide what mattered is the structure problem made visible.

  • Should search replace navigation?

    No. Search serves people who know what they want; navigation teaches people what exists. A product where search is the only viable route has stopped explaining itself.

  • What's the cheapest way to test IA?

    A tree test: strip away the interface, give people the labels, and ask them to find five things. An afternoon's work, and it produces a number you can put in an award entry.

Assessed work that scores highest on information architecture

Chosen by score from the work the editors have assessed, not by hand.

  • information architecture
  • navigation
  • scoring

Written by

The UXE editors

The team that selects the work, uses it, scores the seven criteria and writes every verdict on this site.