Terms that appear in my writing. Anyone who argues that brands should make their rules explicit should start with their own. The first part holds my own coinages, each with a date and a place it is developed. The second holds terms from the field that I use and that should be lookupable.
Terms I define
Brand Infrastructure Brand Specification Machine Readable Brands Agentic Brand Readiness Machine Consensus Agent Authority Accountable Decision Consensus Taste The Bet Stateful Identity File Open Brand Repo
Terms I use
GEO Answer Engine / AEO llms.txt Schema.org / JSON-LD Model Context Protocol Machine Customers Agentic Commerce
The machine-readable layer of a brand: explicit, parametrizable, for systems.
It complements brand identity, which addresses people, rather than replacing it. A brand needs both at once: identity for people, infrastructure for systems. Older uses of the expression exist in a different sense, for example for organizational brand processes.
The brand in a form that can be applied rather than interpreted.
Positioning, language rules, facts, and boundaries, written for people and systems at once. A brand book describes a brand and leaves the reading to the reader. A specification states what holds, and can therefore be executed. That difference decides whether a brand appears in automated processes at all.
Brands whose identity, language, and design logic exist as a specification that people and systems can work with equally.
Third stage of a maturity model: being read, producing with AI, becoming applicable. A brand book describes a brand; a specification can be applied. The expression is also used by others since late 2025, mostly descriptively.
The measurable maturity level of a brand toward AI systems: being found, described correctly, and recommended.
Stage one of the ladder, and the only one you can test yourself in ten minutes: three questions to ChatGPT or Perplexity, four points of evaluation.
The picture of a brand that AI systems assemble from every source they can find.
Not what a brand says about itself, but what systems synthesize from its website, trade press, reviews, and forums. For anyone asking an AI, that consensus is the brand. It cannot be controlled, but it can be influenced: through the sources that feed into it.
The deliberately granted, bounded, and accounted-for mandate a brand gives its agents.
Granted means: decided rather than drifted into. Bounded means: defined what the agent may promise and where a human takes over. Accounted for means: there is an address that stands behind it. The expression comes from the law of agency and is carried into the brand context here.
The judgment about which option is right: justified, dated, versioned, with a named sender who stands behind it.
What remains of brand work when machines take over execution. An opinion is stated and left behind. An accountable decision carries consequences. Opinions are free, including from machines.
Taste that can be learned from many examples, and therefore reproduces the middle.
What many people consider good can be trained. When that middle comes out of a machine, we call it slop. The objection is not aimed at the machine but at the assumption that taste is an average. A decision that risks something is by definition not the middle.
The recommendation that commits: thesis, named failure signal, sender who bears the consequences.
What strategy work sells when analysis is available in minutes. Language models deliver plausibility, meaning the most likely consensus. A commitment is the opposite of that.
A brand carries its entire history with it. Every statement lands against everything said before.
A language model is stateless: each request starts from zero, with no memory of the last. Brands work the other way around. What you promise today is measured against what you promised three years ago. So it is not enough to tell a system how a brand sounds. It has to know what the brand has already said.
The file that states who a brand is and what it acts on.
In software circles the name SOUL.md circulates for this. Identity file is the more precise expression: less pathos, clearer about the thing. It is not a document about the brand but the place where the brand exists for systems. What is not in it does not get applied.
The brand specification as a public, versioned repository.
Identity, language, and facts readable for agents and verifiable for people, while the boundaries of what an agent may promise stay internal. Trust through verifiability instead of assertion. Every rule a documented decision, with date and sender.
Terms from the field that appear in my writing. Not coined by me, explained here.
Optimizing content for the answer a generative system gives, rather than for a ranking in a list of results.
Generative Engine Optimization. SEO gets you found; GEO gets you into the answer. Both remain work on discoverability. Machine Readable Brands start one step earlier: with the question of whether anything applicable exists that a system can work with.
A system that answers a question directly instead of linking to results. AEO is the practice of optimizing for it.
The difference from a search engine is the missing list: there is one answer, not ten options. If you are not in it, you are not there. In practice AEO overlaps heavily with GEO; the split is more the market's vocabulary than a real distinction.
A text file summarizing a website's most important content for language models.
Proposed in 2024 and heavily promoted since. For coding and browser agents that are handed the file directly, it is useful. As a visibility lever it is, by measurement, ineffective: a 2026 analysis found that 97 percent of all llms.txt files receive no AI request at all. This site has one, for the first reason, not the second.
A shared vocabulary for marking up who or what a website is. JSON-LD is the usual format for it.
The least exciting and most effective part of brand infrastructure. A person can see that a page holds a person, a product, or a price. A system has to be told. Without that markup, a statement stays prose a system has to interpret instead of adopt.
An open standard through which AI systems reach external tools and data sources.
Introduced by Anthropic, established as a common standard in late 2025, now held by a neutral foundation. For brands the technology matters less than the consequence: there is now a standard route by which an agent reaches brand data. Anyone with nothing available along that route does not exist for agents.
Systems that buy on their own or prepare purchasing decisions.
Coined by Gartner. The forecast is that such systems could account for 15 to 20 percent of revenue in relevant categories by 2030. What matters is not the number but the shift in addressee: brand communication then speaks to a customer that sees no images.
Commerce in which an agent on at least one side compares, negotiates, or buys.
An actual purchase by an agent is currently the rarest case. More common, and already more consequential, is the step before it: agents that compare, filter, and recommend. The recommendation decides earlier than the transaction.