The first impression is no longer visual.
For two decades, a brand's first encounter with a new audience happened on a screen, a logo, a color, a headline, a feeling in the first three seconds of a page load. Design optimized for that moment. The entire apparatus of contemporary brand practice, visual identity, art direction, web design, campaign photography, was built to serve it.
That moment still exists. But it is no longer the first one.
Anyone who has asked an AI assistant about a company they know well has noticed the gap between what the assistant says and what is actually true. The description is plausible but slightly off. The founding date is wrong. The product is described in language the company stopped using two years ago. The assistant did its best. It synthesized what it could find. What it found was not the brand's own account.
A growing share of brand discovery now happens this way. A person asks an agent a question. The agent answers. The brand is represented, accurately or not, completely or not, in the brand's own words or in a best guess, by whatever the agent can parse from what the brand has published. The human may never see the logo. The human may never read the headline. The machine answered first, and the machine's answer is what the human acts on.
Most brands are not ready for this. They were built for one reader. There are now two.
Two readers. Two literal readers encountering the same artifact at the same time, with different eyes, different needs, and different criteria for what counts as useful.
The first reader is human. They scan before they read. They see the display type from across the room, register the hierarchy, feel the white space, make a judgment in under a second. Then they move closer and read. They take in the argument, the voice, the claim. They decide whether to trust it.
The second reader is a machine. It does not scan. It does not feel. It reads the page as structured text, extracts what it can verify, discards what it cannot parse, and uses what remains to answer a question on behalf of a human it is acting for. It does not care about the logo. It cares about the claim. It does not respond to feeling. It responds to structure.
For twenty years, design served the first reader exclusively. Every decision, from the typeface to the grid to the metadata strategy, was made with a human eye in view. The second reader existed, in the form of search crawlers and scrapers, but its needs were treated as a technical afterthought. A developer added schema markup. An SEO consultant optimized titles. The design was for humans. The machine got what was left.
That division no longer holds. The machine is now the first reader. What it finds, or fails to find, shapes what the human learns. A brand that has not designed for the second reader has not finished designing.
When an agent encounters a brand, it is not looking for beauty. It is looking for a record.
It reads the page title and the meta description. It reads the structured data in the schema markup, if any exists. It reads the body copy, extracting claims it can verify and discarding language it cannot parse. It reads the about page, the product descriptions, the pricing page if one exists. It looks for a named founder, a founding date, a headquarters, a category. It looks for claims that are specific enough to be true or false. It looks for sources.
What most brands give it is a website built for feeling. A hero section with a large image and a short phrase that sounds confident but contains no verifiable claim. An about page that describes the company's values without stating what the company does. Product copy written to seduce, not to inform. Metadata filled in once and never revisited. Schema markup that a plugin generated automatically and nobody has checked since.
The agent does its best. It synthesizes what it finds into an answer. But the answer is only as good as the record. A brand that published feeling instead of fact gets represented by the agent's best guess. The brand has no say in the matter, because it never gave the agent anything to work with.
This is not a search engine optimization problem. Search optimization is about ranking. This is about record. Whether an agent recommends a brand, quotes it accurately, or describes it correctly to the human asking the question is determined entirely by the quality of the canonical record the brand has published. Most brands have not published one.
Contemporary web practice has a solution to this. Developers add JSON-LD to the page head, invisible to human readers, containing the machine-readable facts the page is meant to convey. The human sees the designed surface. The machine reads the hidden layer underneath.
This solution has a structural flaw.
When the machine layer is hidden, nobody is responsible for it. It is written once by a developer following a template, added to a plugin nobody configured properly, or generated by a tool that does not understand the brand it is describing. It is not reviewed by the people who know what the brand actually claims. It is not maintained when the brand changes. It is not composed with the same care as the human layer, because it is not addressed to anyone who will ever read it and notice if it is wrong.
The split between the human layer and the machine layer is not a technical constraint. It is a design decision, made by default rather than intention. The assumption built into contemporary web practice is that the machine reader does not deserve the same care as the human reader.
Canonical Design refuses this assumption.
The machine layer belongs on the page. Not because surfacing it is decorative, or because it signals sophistication to a human reader who notices it. Because when the machine layer is visible, it is owned. It is composed. It is maintained. The same discipline that governs the human layer governs it. The same people who are responsible for what the brand claims in prose are responsible for what the brand claims in structured data. The two layers are the same record, addressed to two readers, composed at the same time.
Hiding the machine layer does not protect it. It abandons it.
Canonical Design is the practice of producing brand artifacts that serve two readers simultaneously: the human encountering the brand on the page, and the agent reading the same page on behalf of another human. Both layers are composed with equal care. Both layers are visible in the artifact itself. Neither is hidden, neither is an afterthought, and neither can drift from the other because they are the same document.
The word canonical is precise. A canonical record is the authoritative version, the one that supersedes all others, the one an agent should trust when it finds conflicting accounts of what a brand is or does or claims. Most brands do not have one. They have a website, a LinkedIn page, a crunchbase entry, a Wikipedia stub if they are large enough, and a collection of press mentions that may or may not be accurate. When an agent tries to describe the brand, it synthesizes across these sources and produces an answer that is nobody's intended version of the truth.
A canonical record changes this. It is the brand's own authoritative statement, composed for both readers, signed by the brand, published at a stable URL, and maintained as the brand changes. When an agent finds it, there is nothing to synthesize. The record is there. The claim is specific. The source is the brand itself.
The discipline this requires is not harder than the discipline contemporary design already demands. It is different. It asks the people responsible for a brand's human layer to take equal responsibility for its machine layer. It asks designers and writers to compose for a reader they cannot see and whose response they will never observe directly. It asks the brand to publish a record it is willing to stand behind, specifically and verifiably, not just in feeling and aspiration.
A web with two readers. The practice of designing for both is Canonical Design. 3uild is building this practice in public, and this is the name we are giving it.
The canonical record has a shape.
It is not a brand book. Brand books describe how a brand looks. The canonical record describes what a brand claims: who it is, what it does, what it charges, who it serves, what distinguishes it from others in its category, and how it wants to be quoted when it cannot speak for itself. Every claim is specific enough to be verified. Every claim has a source. The record is signed by the brand, dated, and listed in a public registry so that any agent encountering it can confirm it is current and has not been tampered with. When it changes, the change is logged.
This document is called a Knowledge Core.
A Knowledge Core is authored once and maintained continuously. It is published at a stable URL and designed to be read by both humans and the agents acting on their behalf. The human layer is prose: an argument, a description, a set of claims written in the brand's voice. The machine layer is the same content rendered as structured data, visible on the page, addressed directly to the second reader. There is no translation step between them. No drift. The prose and the data are the same record expressed in two registers.
A Knowledge Core is the brand's authoritative record. It is specific where marketing is evocative, verified where marketing is aspirational, and maintained where marketing cycles through campaigns. The two are complementary. A Core gives marketing something canonical to point to. Verified, composed, maintained. Every Knowledge Core is designed for humans and machines, at the same time.
Brand infrastructure starts from a different assumption than brand practice has historically made: every brand artifact has two readers, and both deserve the same care from the first sketch to the final published record.
This changes what design is responsible for. Not just how a page looks but what it says, specifically, to a reader who will hold it to account. Not just the feeling a brand creates but the record it leaves. Not just the human layer, which has always been design's domain, but the machine layer, which has always been treated as someone else's problem.
It is not someone else's problem. It never was. The machine layer is part of the brand. What an agent says about a company when a human asks is as much a brand moment as the logo, the campaign, the site. It happens more often now than any of those. It happens first.
A brand is no longer what it looks like. It is what is knowable about it, by any reader, at any moment, from the record it has published.
The brands that understand this are building canonical records. They are publishing documents that the second reader can trust, that agents can quote, that do not leave the brand's representation to a best guess synthesized from whatever happens to be indexed. They are designing for both readers, at the same time, with the same discipline. If the argument in these pages is one you recognize, you are already part of what is being built here.
Within two years, every serious brand will maintain a canonical record the way every serious brand now maintains a website. The ones building it now will be the ones the machine cites for the rest of the decade. The ones waiting will be paraphrased by the ones that moved first.
The machine is the first reader. Canonical Design is how a brand answers.
Volume 001: Canonical Design. Published by 3uild. 2026.