Your organisation has real expertise. Your content is solid. And yet, when someone asks ChatGPT or Google’s AI a question you should own, a thinner competitor gets named and you do not. The instinct is to fix the content, add a few pages, refresh the copy, work the keywords. The reason usually sits further back than that.
AI search decides what to surface by reading how a site is built. It follows the structure to work out what an organisation covers, which parts are authoritative, and which passage answers the question in front of it. That structure, the hierarchy, the naming, the way pages relate to one another, is set early, at the design stage, well before a word of content is written.
This is the part most teams miss. Getting found in AI search is decided upstream, in the structure and naming of the site, and the organisations that get cited are the ones who settle that deliberately at the design stage.
What is information architecture for AI search?
Information architecture for AI search is the deliberate structuring and naming of a website, its hierarchy, navigation, URLs and topic grouping, so that both people and AI engines can find, interpret and trust what it covers.
Information architecture, or IA, is a web development discipline of organising a site so its contents make sense. It decides what pages exist, how they nest, what they are called, and how they connect. It sits above content and shapes how content is found.
For AI search, that framing does a specific job. Google’s own guidance for its AI features makes the point that the fundamentals of good SEO still apply because the content is drawn from the same Search index, and pages surface when they are genuinely helpful and show clear expertise. An engine reads your headings, navigation, and internal links to build a picture of what your organisation is expert in. A clear structure gives it a clean picture. A muddled one leaves it guessing, and an engine that has to guess tends to reach for a source it understands better.
Good content inside a weak structure is like a well written page filed in the wrong drawer. The quality is real, and no one can find it.
Why discoverability is decided before you write a word
Discoverability is decided at the design stage because structure and naming are settled there, and content is poured into that structure afterwards. Get the frame wrong and good content lands where neither people nor engines can find it.
The reason this matters more now is that ranking and traffic have come apart. AI search returns a synthesised answer over a list of links, so a page can hold its position and still lose the click. Pew Research Center found that when an AI summary is present, people click a traditional search result in about 8% of visits, against 15% when no summary appears. Being present in the results is no longer the same as being found. What gets used is the source an engine can read cleanly and cite with confidence, and that legibility is a property of structure.
Those structural decisions are made once, at the beginning, such as:
- Which pages exist
- How they are grouped
- What the navigation calls them
- How they link
Change them later and you are into redirects, broken links and re-mapped navigation across a live site, which is slow and costly. Settling the architecture early is the cheaper path by a wide margin, and it is the one that decides whether your expertise ever surfaces.
What good information architecture looks like for AI search
Strong information architecture for AI search comes down to a short set of upstream decisions. Each one makes your site easier for an engine to read and to trust.
- Hierarchy and depth. Important pages sit near the top, reachable in a click or two. Anything buried deep is harder for both users and crawlers to reach and value.
- Navigation labels in your users’ words. Menus and section names use the language people actually search with, not internal or sector jargon that only insiders recognise.
- URL and page naming that describes the content. A clear, readable URL tells an engine what a page covers before it reads a line of the body.
- Taxonomy and topic grouping. Related pages cluster around a clear parent topic, so the site shows depth on the subjects you want to be known for rather than a scatter of disconnected posts.
- Internal linking as connective tissue. Links between related pages, with descriptive anchor text, are how an engine infers which topics you cover thoroughly and treats you as a source worth citing.
Each of these is a structural choice, settled before content, and far easier to get right at the start than to unpick later. After this step, we focus on structuring and writing the content so an engine can lift a clean, self-contained answer. Read more about how to structure content for AI search.
Where this goes wrong for complex organisations
Large government, health, education and not-for-profit sites are built by many hands over many years, and the architecture debt that accumulates is exactly what makes them hard to find in AI search.
The symptoms are familiar to anyone who owns one of these sites:
- Orphaned pages that nothing links to and no one can navigate to.
- Duplicate or competing sections that fragment a topic across the site.
- Labels that only staff understand, so users and engines both miss the content.
- No clear home for a topic, so authority is spread thin instead of concentrated.
This matters more in these sectors because of what people ask. Informational queries such as “how do I apply”, “eligibility for” and “what does this policy mean” are the primary target for AI-generated answers, and these are precisely the questions a for-purpose organisation exists to answer.
Pew’s analysis of AI summaries found that government sites made up a larger share of cited sources in AI summaries than in standard results, around 6% against 2%, so these are precisely the organisations the answers draw on.
Topics that touch health, money and safety carry the highest bar. Google’s raters treat them as Your Money or Your Life topics that need the most scrutiny and reward demonstrable expertise. A confusing architecture undermines that at the source, because the engine cannot tell where your authority lives.
How a deliberate discovery stage made a complex health site findable
The same discovery and information architecture discipline that makes a complex site navigable for people is what makes it legible to a machine. Peter MacCallum Cancer Centre is a clear example.
Peter Mac is a world-leading cancer research, treatment and education institution, and its website serves patients, families, clinicians and researchers, each arriving with different needs. Before any build began, Butterfly ran a full discovery phase: stakeholder interviews across the organisation, competitor and heuristic analysis, user journey mapping and defined personas. That work settled the structure before the content was written.
The result was an architecture that directed each user type to what they needed quickly, with an expert finder to connect people to the right specialists and site search layered on top. The site met WCAG 2.1 AA, so it was accessible to all users by design.
The point for AI search is the sequence. Findability was decided upstream, in the structure and naming, through a deliberate discovery stage. A site built that way gives an engine the same clean picture it gives a patient, which underpins everything else.
The part of this that AI cannot do for you
AI is making the build itself cheaper and faster. That raises the value of the judgment that comes before the build: understanding real users, empathy-led design, and decisions about what a site should contain and how it should be named.
When writing code becomes cheap, code stops being scarce. What stays scarce is knowing which site to build in the first place: which user tasks matter, how people actually search, what the organisation is genuinely expert in, and how to structure it. That is research and design work, and it is human.
This is how Butterfly works: user-centric web design and hand-coded front-end craft. We apply our expertise where it creates the most value and make thoughtful decisions that serve the long term rather than the fastest path. The common failure we see is a build handed to developers who lack strategy or design alignment, so it results in a technically fine site that no one thought through, and it never gets found.
As WCAG 2.2 specialists, we treat accessibility and clear structure as the same discipline, because a site that is legible to a screen reader is usually legible to an engine too. A cheaper build makes that thinking more important.
What to get right at the design stage
Before you commission content or a build, settle these questions. These upstream decisions determine whether you get found, and a non-technical owner can lead the conversation.
- What are our users’ real tasks? List the actual jobs people come to the site to do, in their words.
- What should the site contain? Decide which pages need to exist to serve those tasks, and cut the ones that don’t.
- How is it grouped and named? Settle the topic groups and the labels, using the language your users search with.
- How do the pages link? Plan the internal links that connect related pages and signal where your expertise sits.
- Who owns the structure? Name the person or team responsible for keeping the architecture coherent as the site grows.
Answer these before the build, and content and code have a sound frame to sit in. Skip them, and you’ll likely be repairing structure later, at far greater cost. The technical build has its own set of questions to put to your web partner, from semantic markup to how the site is crawled and rendered. Those are covered in what to ask your web development partner about AI search.
Frequently asked questions about website structure for AI search
Information architecture for AI search is the deliberate structuring and naming of a website so people and AI engines can find, interpret and trust what it covers. It includes hierarchy, navigation, URLs and topic grouping, and it is set at the design stage, before content is written.
Yes. AI engines follow a site’s structure and internal links to understand what it covers and to locate the passage that answers a query. Clear hierarchy and naming make your content easier to find, interpret and cite. Poor structure leaves good content stranded where engines cannot use it.
Usually not a full rebuild. Most sites need targeted structural work: fixing navigation labels, grouping related pages, clarifying URLs and repairing internal links. A discovery pass identifies what to change. The gains come from getting the architecture right, which is far cheaper than replacing the whole platform.
Yes. Search engine optimisation remains the foundation. AI search builds on the same fundamentals: clear structure, direct answers, and genuine expertise. Strong information architecture improves both traditional rankings and AI citation, because both depend on an engine understanding what your site covers and trusting it.
Traditional SEO optimises pages to rank. Structuring for AI focuses on making content easy for an engine to extract and cite as a self-contained answer. Both reward clear architecture, so good information architecture serves them both.
AI finds organisations who actively plan to be found
The organisations AI surfaces are the ones that deliberately designed their structure at the design stage, with a real understanding of the people they serve. That decision shapes everything a search engine later reads about them. As development gets cheaper and faster, this upstream judgment becomes what sets good digital platforms apart.
The practical move is to treat structure as a design decision and get the thinking right first. Look at how your site is organised and named, and ask whether it reflects how your users actually search and what your organisation is genuinely expert in.
If you want a clear view of how your own site is structured and found, we can walk through it with you.
Start a discovery and information architecture conversation with Butterfly about how your own site is structured and found.
