Information Architecture: Structure for Users, Google, ChatGPT, Claude, Gemini & Co
A demand-driven information architecture unites business goals and user needs to build efficient organic touchpoints
A demand-driven information architecture works like a magnet: it shows users the direct route to their goal instead of prescribing detours. It rests on actual demand, on the surface where a brand is already visible, and on the goals of the company. This only succeeds when more than people understand what is being said: Google, ChatGPT and Claude have to consider the content relevant, or nobody sees it.
Information architecture starts with demand, existing listings and business goals
Information architecture describes the structuring, organisation and labelling of content within a digital offering. It defines which content belongs together, what it is called and by which routes users reach it. Established methods include content audits, card sorting, wireframes, usability tests and customer journey maps. Information architecture forms the logical foundation, navigation makes that foundation operable.
Usually the term stops at the edge of a company’s own website. That is exactly where the problem starts. The three most important inputs sit upstream of the structure:
- Actual demand. Which questions prospects ask, in search engines as well as AI systems, at which stage of their decision.
- The existing listings and mentions. On which channels, in which directories, trade media and comparison portals the brand already appears today.
- The goals of the company. Which services should grow, which audiences are addressed, which locations and markets count.
Planning a structure without these three rearranges existing content. Including them means building the content that is actually in demand and placing it where it gets found. The demand analysis delivers the first, the competitive analysis with benchmark the second.
From this follows a wider definition: information architecture covers all content from and about a company, including what stands about it outside its own domain. The section on offsite architecture below explains what that means in practice.
Research, design, build, learn: an architecture that optimises itself in a recursive loop
An information architecture is designed once and has to adapt to a rapidly changing environment from then on.
New requirements keep arriving. From outside: new standards from Google, ChatGPT and Claude, shifts in demand through seasonality and trends. From inside: new product categories, changed focus areas, additional locations, a different range, new priorities. Every one of these changes follows the same logic: designed at benchmark level, implemented fast and without errors. An agile loop is the structure for that.
Layers
01
Research
User intent and demand by decision stage, what competitors offer against it, current visibility per touchpoint.
02
Design
Structure and page types per touchpoint: content, design, goals & KPIs.
03
Build
Implementation in prioritised order, onsite and offsite across the whole content architecture.
04
Learn
Measure the effect per touchpoint, deviations go into the next round’s backlog.
Every layer of the architecture feeds its own metrics into this loop: rankings and clicks, visibility in AI answers, dwell time and exits, enquiries per entry page. What stands out goes into the backlog of the next round.
The layers of information architecture at a glance
- Website structure – navigation, sitemap, page types, taxonomies
- Internal linking – directing relevance to where it works
- Technology – structured data, loading order, metadata on all media
- The single document – composition, answer logic, extractability
- Offsite – directories, profiles, trade media, reviews, third parties
- Media classes – owned, earned and paid under one logic
- Special cases – location pages and listing pages in retail
The first four layers concern the company’s own website and account for most of the work. The last ones decide how much of that work becomes visible.
Onsite architecture: structure, linking, technology and documents
Website structure: navigation and page types along user questions
The structure mirrors how a subject actually works. Departmental boundaries do not interest users.
Demand produces topic clusters, clusters produce page types. A page type is a recurring format with a fixed composition: service page, location page, guide, case study, product detail. Large websites typically need ten to thirty page types, enough for real differentiation and few enough to maintain.
The sitemap arranges these page types into a hierarchy. Taxonomies connect them across it: by industry, use case, region or audience. The URL structure follows the hierarchy, stays short and contains no years. Navigation makes both operable and stays consistent across all areas, which is also a criterion for an accessible website.
The guide on planning a website structure describes the design process step by step. The structure is tested with users. Card sorting reveals how people group and name content. Tree testing measures whether they reach their goal within the proposed structure.
Internal linking: directing relevance to where it works
Internal links carry meaning. Every link tells the machine what the target page is about and the user why the click is worth it.
Three rules decide whether internal links transport meaning:
- One anchor text belongs to exactly one target. The same anchor text on two different pages devalues both.
- The anchor text names the focus topic of the target page or a synonym of its intent, never “here” or “read more”.
- Links sit where they contribute something.
How many internal links a page receives is a decision. Pages with commercial value get more, secondary pages fewer. A link matrix records which anchor text points to which target and remains the basis of all editorial work for years.
Technology: structured data, loading order and metadata on all media
Machine readability belongs to the architecture and does not happen retroactively.
Structured data following Schema.org describes what a page is and what it covers: organisation, service, location, article, question and answer. It connects entities and turns text into verifiable statements. A grounding page with the company’s core facts and an llms.txt pointing AI systems to the authoritative content do the same.
Loading order decides what a visitor sees first. The most important part is delivered first, everything else afterwards. Good Core Web Vitals make sure the structure arrives at all.
Metadata belongs on every medium: text, image, video and document. Every image gets a descriptive file name, an alternative text, a title and, where useful, a caption. Videos get transcripts and VideoObject markup, documents meaningful titles. Images and video are entry channels of their own, with demand of their own.
The single document: answer first, sections extractable
Every single page is a small architecture in itself. It starts with the answer and explains afterwards.
Headings name the content and the finding of the section that follows. Every section stands on its own and remains comprehensible when quoted alone. That is exactly what generative systems do: they split documents into passages and evaluate each one separately. A section that makes no sense without its predecessor does not get cited.
Lists, tables and question-and-answer blocks increase extractability. Plain language does the same for people: short main clauses, familiar terms, technical vocabulary where it creates precision and nowhere else.
Offsite architecture: directories, profiles and third parties carry the same structure
What a decision-maker sees of a brand is mostly not on its website. It sits in industry directories, map services, review portals, trade media, comparison platforms, social profiles and third-party coverage. Designing only onsite shapes a fragment and leaves the larger part to chance.
For AI systems this part carries particular weight. They answer questions from several sources and weigh what independent third parties say. A brand described identically everywhere is recognised as one entity. A brand with diverging names, services and details falls apart into several unclear records.
The architecture therefore secures three things across all touchpoints: the same terms for the same services, the same facts about company and locations, the same relationships between them. Where third parties report, preparation decides what they write. Templates, facts and wording belong in place before anyone asks. Details are covered under PR and offsite and local listings.
Owned, earned and paid follow one architecture
Three media classes, one vocabulary. What is called a service on the website carries the same name in the ad and in the trade article.
Paid campaigns test demand faster than organic visibility can answer it. What works in ads belongs in the structure. What resonates editorially deserves promotion. This feedback is why content, SEO, GEO and campaigns need the same architecture: each class supplies the others with data as soon as they speak the same language.
Special cases of information architecture: locations and listings
Two constellations follow their own rules because they manage content in large numbers and in constant flux.
Location pages: location data belongs in the entire architecture
A location page carries the complete location data, the offering of that location and the catchment area it actually serves.
In structured data this means LocalBusiness with address, geo coordinates, opening hours and phone number, areaServed for the catchment area, hasOfferCatalog for the services of that location and Offer for offers valid there. Every location page points to its Google Business Profile and to the directory entries where that location is listed.
What matters is consistency of location data across the entire architecture. Name, address and phone number appear identically in the on-page text, in the structured data, in the business profile, in directories and on social profiles. Divergences split a location into several unclear records, and with many locations the damage multiplies. Maintenance therefore runs through one record per location that feeds all channels. This includes local SEO and Google Business Profile management.
Listing pages in e-commerce and vehicle retail
In retail the listing decides visibility. Someone searching for immediately available vehicles of a model lands on a listing. The individual vehicle is sold tomorrow, the listing remains.
Four questions need answering for every listing. Which selection does the listing represent, and does that selection match a combination that is actually in demand. How many items does it hold, so it appears neither empty nor unmanageable. Which details belong in the item preview, such as price, availability, year, mileage or specification. And which parameters do better-ranking competitors already carry in their listings although they are missing from the individual documents.
That last point is overlooked most often. Filter parameters with their own demand deserve their own indexable listing. Filter combinations without demand stay out of the index, because they consume crawl budget and create duplicates.
Lifecycle management: sold items redirect to the most valuable listing
A sold item holds rankings, incoming links and visitors. A 404 wipes all of that at once.
The better solution is a redirect to the most valuable listing the item belonged to. A sold estate car of a particular model leads to the listing of available estate cars of that model. The visitor arrives at a genuine alternative, and the listing inherits the signals of the individual document.
Which listing is the most valuable follows from demand and rankings. This mapping is defined once and then automated, because it applies again with every sale.
Anyone who maintains original content instead of adopting manufacturer texts, labels their media completely, keeps the architecture consistent beyond their own domain and develops the result in sprints through a maintained backlog has created the conditions under which visibility grows predictably.
Information architecture takes shape in five steps
- Audit – capture the inventory: pages, page types, linking, technology, content.
- Benchmark – compare competition and existing listings, make distances measurable.
- Demand – gather keywords, questions and prompts along the customer journey.
- Design – sitemap, page types, taxonomies, content outline, link matrix, schema plan.
- Implementation and measurement – implement by priority, measure impact, maintain the backlog.
The result is a set of handover documents: a structure document with sitemap and page types, a content outline per page type, a link matrix with anchor texts, a schema plan and a prioritised backlog. Any editorial team and any development team can continue from there, including without the agency.
The effort depends on a few factors: number of page types, size of the inventory, number of languages and markets, number of locations, and whether an existing website is being ordered or a new one built.
Information architecture for existing sites and relaunches
On existing sites, ordering grown structures comes first. Typical findings are duplicate content, competing pages on the same topic, orphaned subpages and navigation that has been extended around exceptions for years. The work proceeds step by step without endangering rankings: merge, redirect, relink, measure.
In a relaunch the structure comes before the design. Only once page types, content and routes are settled is a draft worth making. The reverse produces websites that look good and get nobody anywhere. Anyone planning a relaunch schedules the architecture as the first work package.
Frequently asked questions about information architecture
What is information architecture?
Information architecture is the structuring, organisation and labelling of content within a digital offering. It defines which content belongs together, what it is called and by which routes users reach it. The aim is findability and comprehensibility, for people and for machines.
What is the difference between information architecture and a sitemap?
A sitemap lists pages. Information architecture is the model behind it: page types, categories, relationships and labels. The sitemap is one of its results.
How should a website be structured?
Along the questions of its users. Each level answers a question and leads to the next. Recurring content follows fixed page types so that structure survives growth.
How can a website structure be visualised?
As a tree diagram with page types and hierarchy levels, supplemented by wireframes for the most important page types and user flows for the most important routes. Together these three views form a concept document that editorial, design and development can work from.
How does structure affect visibility in AI answers?
Generative systems split documents into passages, resolve entities and weigh what independent sources say. Clear structure, unambiguous labels, structured data and consistent details outside the company’s own website raise the probability of being cited as a source.
Does this work without a relaunch?
Yes. Most projects start on existing sites. Structure, linking, technology and document composition can all be improved step by step without rebuilding the website.
What does information architecture cost?
The effort depends on the number of page types, the size of the inventory, languages, markets and locations, and on whether an existing website is being ordered or a new one designed. The range is estimated in the initial consultation based on these factors.
Who takes part in an information architecture project?
On the agency side structural design, editorial, technology and offsite. On the company side those responsible for website, marketing and, where present, sales. Without sales knowledge, the questions prospects actually ask are missing.
How many page types does a large website need?
Typically ten to thirty. Fewer means different content gets pressed into the same format. More means nobody can maintain the formats. What matters is that each page type has its own job and differs from the others.
How are several languages and markets handled?
Every language version gets the same structure but its own content and its own demand. Translated structures fail wherever one market searches differently from another. Technically, hreflang connects the versions; editorially, demand analysis per market decides the deviations.
How are many locations reflected in the structure?
Through a dedicated page type for locations with a consistent composition and location-specific content. What matters is that each location page carries genuine substance and matches the entries in directories and map services.
What happens to old URLs during a rebuild?
Every retired URL gets a redirect to the closest matching new address. Before the rebuild, a complete list of existing URLs with their rankings is recorded; afterwards it is checked whether rankings and clicks arrive at the new addresses.
How is success measured?
By rankings and clicks per page type, by visibility in AI answers, by the number of pages that generate entries at all, by exits on the most important routes and by enquiries per entry page. Measured against the state before the rebuild.
Initial consultation including benchmark and roadmap
The initial consultation takes 25 minutes and is free of charge. It comes prepared with a benchmark against the relevant competition, an assessment of the current structure, proposed solutions with effort estimates and a roadmap through to a running improvement loop.
A meeting is worthwhile if you are prepared for starting costs in the low four-figure range and want to make the structure of your digital presence a management matter. For corporations, mid-sized companies and growing businesses, in B2B as well as B2C.
Related services for more visibility and enquiries
Related articles
Book a free, no-obligation consultation
In the initial consultation, you receive an analysis and a concrete proposed solution.