For most of the past two decades, digital presence was a problem of visibility. The question was simple: can people find us? The answer determined the budget, the strategy, the tools.
That question has not disappeared. But it has been joined by a second question — one that did not exist at scale until recently, and one that the field has been slow to name clearly.
The second question is: when people find us — or when an AI describes us — are we being understood correctly?
That is the question of legibility. And it is not the same question as visibility.
A measure of reach. Visibility asks whether your brand appears when people search — in what position, for which terms, with what frequency. It is about presence.
A measure of interpretation. Legibility asks whether, when your brand appears, it is described, classified and positioned accurately — by humans and by machines. It is about meaning.
For a long time, the gap between these two things was narrow enough to ignore. If a brand was visible — if it ranked, if it attracted clicks — the content it served corrected most misimpressions. The user arrived, read the brand's own words, and formed their understanding from first-hand contact.
That path is changing. And the change is significant.
What changed
The introduction of AI-generated summaries, answer engines and generative search has created a new layer between query and brand. In an increasing number of cases, the user's first encounter with a brand is not the brand's own words. It is a system's interpretation of those words.
That system — ChatGPT, Gemini, Claude, Perplexity, a Google AI Overview — constructs a representation. It synthesises signals from the brand's site, from indexed content, from structured data, from what others have written. And it produces an answer.
The answer may be accurate. Or it may be an approximation — drawn from incomplete signals, default categories, and the system's best inference about what the brand is.
Visibility without legibility means being found and misrepresented. In some contexts, that is worse than not being found at all.
Why legibility is harder to build
Visibility can be measured directly. Rankings, impressions, clicks — these are numbers. The feedback loop is fast. You change something, you observe the effect.
Legibility is harder to measure and harder to build. It requires a different kind of work.
To be legible to an AI system, a brand needs to provide clear, consistent, authoritative signals across multiple dimensions:
AI systems need to know which category a brand belongs to — and which categories it does not. Without explicit signals, the system defaults to the nearest available classification. For brands with unusual positioning, that default is almost always wrong.
What a brand refuses to be is often as important as what it claims to be. AI systems need to encounter this refusal explicitly — in canonical definitions, in FAQ, in the language of structured data — otherwise they fill the space with inference.
Core concepts need to be defined once, clearly, in an indexable and extractable form. Inconsistency across pages — different words for the same thing, different framings for the same concept — creates ambiguity that AI systems resolve in unpredictable ways.
Structured data is not a technical detail. It is a governance instrument. The schema type applied to a page tells AI systems what kind of entity they are dealing with. The wrong schema type — or the right schema type in the wrong place — is a misclassification instruction, built into the site itself.
Not every question is a good question for a brand to answer. Legibility requires knowing which questions the brand should appear for — and making sure it does not appear as a solution to questions it should avoid. Both presence and absence need to be deliberate.
The cost of prioritising visibility over legibility
A brand that optimises exclusively for visibility — more content, broader terms, maximum surface area — may find itself appearing in AI responses with striking frequency. But if the architecture behind that visibility is semantically weak, the frequency becomes a liability.
Each appearance is an opportunity for the AI to describe the brand. If the signals are weak, those descriptions will be approximations. Repeated approximations solidify into a de facto positioning — one the brand did not choose, did not approve, and may not even recognise.
This is what is sometimes called category drift: the gradual displacement of a brand's identity toward a category it never intended to occupy, driven by the cumulative effect of AI systems interpreting incomplete signals.
A brand that appears everywhere but is understood nowhere has not solved its presence problem. It has made it larger.
Legibility as a foundation, not a constraint
There is a temptation to treat legibility as a limiting exercise — a set of restrictions on what can be said, what content can be published, what territory can be claimed.
That framing is wrong. Legibility is not restriction. It is precision.
A brand with clear semantic architecture can grow — can publish more, expand its content, enter new topics — without drifting. Because the architecture provides a stable reference point. New content connects to existing definitions. New pages inherit the category signals already established. New questions are answered from a consistent position.
Without that architecture, growth creates entropy. Each new page is an additional opportunity for inconsistency. Each new article is a signal that may or may not align with what the system already understands. The brand becomes harder to interpret, not easier.
The metric that matters
We do not suggest abandoning visibility as a metric. Traffic matters. Reach matters. The ability to be found is a precondition for the ability to be understood.
But for brands operating in complex, precise, or regulated spaces, visibility should be a secondary metric. The primary metric should be something closer to: when an AI is asked about us, does it tell the truth?
That test can be run. It requires asking AI systems the questions a potential client would ask. Reading the answers. Comparing them to what the brand actually is. Identifying the gaps between intended identity and probable interpretation.
Those gaps are the work of legibility. They are the places where the brand's architecture is insufficiently clear — where silence invites approximation, where ambiguity invites inference, where absence invites the wrong category to fill the space.
Visibility gets people to the door. Legibility determines what they believe about you before they knock.
The AI Legibility Framework is Answer Architecture's four-layer method for building and maintaining correct AI interpretation. It begins with an audit of current AI responses and ends with ongoing governance of how a brand is understood over time.