Information Architecture
Date: 2026-08-17
How content is organised, labelled and related so people can find it. It’s the layer beneath navigation — navigation is one view of the architecture, and changing the menu without changing the structure moves the problem rather than solving it.
Information architecture (IA) is the structural design of shared information: what things exist, how they’re grouped, what they’re called, and how they relate.
IA is the structure. Navigation, search, filters and URLs are each one interface onto it.
All four are views of the same underlying model. A site where the menu, the filters and the URLs imply different structures is one where the IA was never decided — it accreted.
The components
ORGANISATION SCHEME
how things are grouped
by type · by audience · by task ·
by brand · alphabetically
LABELLING
what each group is called
— Taxonomy and Labelling
NAVIGATION
how people move between groups
— Navigation Patterns
SEARCH
the escape hatch from the structure
— Search and Findability
Hierarchy shape
The recurring trade:
| Broad and shallow | Narrow and deep | |
|---|---|---|
| Shape | Many top-level items, few levels | Few top-level items, many levels |
| For | Fewer clicks, more visible options | Each individual choice is simpler |
| Against | An overwhelming top level | More steps, and easy to get lost |
Broad and shallow generally wins for retail, because scanning many labelled options is faster than making a sequence of correct choices — and a wrong turn deep in a hierarchy is expensive to recover from.
The real constraint is that products belong in several places. A “hyaluronic acid serum for sensitive skin under £30” sits in a brand, an ingredient, a concern, a format and a price band simultaneously — which no single hierarchy can express. That’s what faceted filtering exists for, and it’s why the hierarchy should be shallow and the facets should carry the specificity — Faceted Filtering.
Polyhierarchy, and its cost
Allowing an item to sit in multiple categories:
BENEFIT found via more routes
COST which URL is canonical?
duplicate content
"where am I?" becomes
ambiguous
crawl budget spent on
near-duplicates
Pick one canonical location per product and treat the others as filtered views, with the canonical URL declared — otherwise the IA problem becomes an SEO problem — Technical SEO, Faceted Navigation and Crawl Budget.
Evidence to build it from
- Card sorting — how people group things, and their vocabulary — Card Sorting and Tree Testing
- Site search logs — the highest-value and most-ignored source. What people type is what your labels should say
- Zero-result searches — what you don’t have, or don’t call by their name
- Support tickets — “where do I find…”
- Analytics — which paths are used, which categories are dead
- Competitors — conventions your customers already know
Site search is the single best IA input you already own. People typing “P&P” when your label says “Delivery” is a labelling finding delivered for free — Search and Findability.
Commercial reality overrides the research
Card sorting produces a user-centred structure; the business has other legitimate inputs.
merchandising priority
margin by category
seasonal ranges
supplier agreements
SEO — what people actually search for
These conflict, and that’s normal. The failure isn’t the conflict — it’s resolving it silently in favour of whoever configured the platform. Decide it deliberately and record why.
Where it goes wrong
- Mirroring the org chart. Categories matching internal departments is the classic sign IA was never user-tested
- Renaming the menu without restructuring. Cosmetic, and the findability problem persists
- Growth without governance. New products get filed wherever, and after two years the structure describes the order things were added
- No maintenance. IA is not a project. Ranges change and the structure has to change with them
- Testing it only in the abstract. Tree testing validates the structure; the live page also has images, filters and merchandising, which change findability considerably
Validating it
Tree test before building, then watch behaviour after.
BEFORE tree testing on the proposed
structure — success and
DIRECTNESS
AFTER site search rate (high = the
structure isn't working)
category page exit rate
depth reached before conversion
zero-result searches
A rising site-search rate is the clearest signal an IA has stopped working, because search is what people do when browsing fails.