Concepts

The library-science practice of establishing and maintaining a single authorized form for the name of each entity, with its variants recorded and its sources cited, so that every reference resolves to one heading.

In cataloguing, the record in an authority file that establishes the authorized form of a name (a person, an organization, a work, a subject), the variant forms it is not, and the sources consulted, so that every reference to the thing resolves to one heading. Applied to a business, it is one verified master record of who it is, what it does and what it has proven, kept true and published everywhere machines look.

Quantitative analysis of publication and citation patterns.

Grouping documents by the citations they share, so that bodies of related work emerge from citation patterns rather than from subject headings.

One exact name for every service, credential, location, and person, held consistent everywhere machines read. Ambiguity makes a machine hedge; precision lets it commit.

A standard set of fifteen core metadata terms (title, creator, subject, description, publisher, contributor, date, type, format, identifier, source, language, relation, coverage, rights) and their refinements, maintained by the Dublin Core Metadata Initiative and standardized as ISO 15836, for describing any resource.

The state in which machines resolve every reference to a business to the one entity it is, and neither confuse it with a namesake nor merge it with another.

The discipline of combining findings across sources into a defensible aggregate claim.

Non-indexed literature and the systematic problem of information that exists but cannot be retrieved.

The structural design of information environments (websites, intranets, apps and knowledge bases): how content is organized, labelled, connected and navigated, so that people can find what they need and machines can understand what is there.

The science of finding relevant material within large collections.

The branch of AI concerned with encoding facts and their relations in a form a machine can reason over: classes, properties, relations and instances, expressed in formal languages such as RDF, OWL and description logics.

Keeping a body of evidence continually current rather than periodically re-done.

Applied ML for continuous monitoring and screening of research literature.

Structured, machine-readable description of a resource: what it is, who made it, what it is about, when, in what language and under what rights.

A formal, machine-readable model of a domain: its classes of things, their properties, the relationships between them and the rules (axioms) that constrain how they combine, expressed in a language such as OWL.

Formal standards governing how research is declared and made checkable.

Peer review of search strategies for reproducibility and recall.

Formal methodology for exhaustive, reproducible literature identification and appraisal.

A controlled vocabulary arranged as a hierarchy, in which each concept has a broader concept above it and narrower concepts below, so that a thing can be classed by walking the tree.

A controlled vocabulary in which each concept carries a preferred label, alternate labels, a definition and scope notes, and is linked to other concepts by hierarchical (broader, narrower) and associative (related) relations, as standardized in ISO 25964.

A distinct, identifiable thing in the world or in thought (a person, place, organization, product, work or concept) that can be named, given an identifier and described by its relationships to other things.

Google's knowledge base of entities (people, places, organizations, things) and the relationships between them, assembled from licensed and public sources and used to understand queries as things rather than strings and to populate knowledge panels.

A graph of entities and the typed relationships between them, each entity identified and described, used to store and reason over knowledge about a domain.

A method of publishing structured data on the web in which things are named by HTTP IRIs that resolve to useful descriptions, and those descriptions link to other things by their IRIs, so that data from different sources can be joined.

A proposed convention for a plain Markdown file at a site's root (/llms.txt, with /llms-full.txt carrying the whole) that summarizes the site for language models at inference time.

An identifier for a resource that is globally unique, never reassigned or changed, and resolvable on the web to a description of the thing it names.

The middle term of a triple: the named relationship that links a subject to an object, drawn from a vocabulary so that the same relationship means the same thing wherever it appears.

Markup, usually JSON-LD in a page's head, that states in the shared schema.org vocabulary which entities a page describes, their properties and how they relate, so that search engines and assistants can read the page as data.

The unit of machine-readable fact in RDF: a subject, a predicate and an object, each identified so that the statement can be joined with others into a graph.

Data organized according to a defined model, with named fields and types, so that a machine can parse it without interpretation.

The design of sites, records and interfaces for use by software agents rather than by people: stable identifiers an agent can follow, typed links that say why one entity leads to another, a machine-readable form of each page, and facts kept beside their subject.

AI systems that pursue a goal on a person's behalf by planning, choosing actions and observing results in a loop, rather than producing a single answer: deciding what is needed, following a link or calling a tool, reading the result and continuing.

Supplier selection in which software screens, scores and shortlists vendors from the records it can read, before or instead of a human review.

The natural-language task of determining which expressions in a text refer to the same thing: that 'we', 'the company' and a name all denote one organization.

Google's framework for judging the quality of a source and its author: first-hand experience, expertise, authoritativeness and trustworthiness, as shown on a website and across the web.

AI that produces text, images or code by learning statistical patterns from large bodies of data (connectionist or neural AI; large language models are its best-known form) and generating what is probable given a prompt.

Search that answers a query by composing a response from retrieved sources, with citations, instead of returning a ranked list of pages: Google's AI Overviews and AI Mode, Copilot's answers in Bing, and the assistants that search the web before replying.

Retrieval-augmented generation in which the retrieved material is a knowledge graph (entities, their relations and the paths between them) rather than text passages, so that the model can answer questions that join multiple facts.

Tying a model's output to evidence it has retrieved, so that each claim in the answer can be traced to a source rather than generated from the model's parameters.

Output from a generative model that is fluent but unsupported by its sources or by fact: invented facts, wrong attributions, confident errors.

The discipline of building and governing symbolic knowledge: inventorying a domain, defining its concepts and relations, deciding what is true and current, and keeping it so.

An open protocol by which AI applications and agents connect to live, structured sources and tools through one standard interface, so that an agent can query a source directly instead of scraping a page.

Systems that combine a neural model with symbolic knowledge: the language model reads and writes natural language while a knowledge graph or ontology supplies verified facts and the logic that constrains the answer.

The knowledge a language model acquired in training and holds in its weights: general knowledge and the entities seen often enough to be memorized, fixed until the next training run.

The retrieval step that selects one passage of a few hundred tokens from a page or a post to answer one sub-query, rather than reading the document whole.

A method in which a language model retrieves relevant, current documents at the time of a query and conditions its answer on them, instead of answering from its training alone.

AI built on explicitly represented knowledge: facts stated as symbols, rules applied by logic, and every conclusion traceable to the facts and rules that produced it.

Inference over explicitly stated facts and rules: following a chain of relations across a graph, applying a rule, deriving a new fact, with each step recorded.

The two questions an AI system settles before it recommends a business: whether it can find the business and resolve it as one entity, and whether it can verify what the business claims.

The representation of words, passages and entities as points (embeddings) in a high-dimensional space in which distance stands for difference in meaning, so that relatedness can be measured rather than matched on keywords.

A closed cycle in which a fact is published at one surface, propagated to an organization's other surfaces and written into its graph, then checked back against its origin, so that the same statement is found everywhere it should be.

Proof in the customer's own voice, verified by the customer: the most convincing evidence a business can hold and the heaviest to collect (the asking, the consolidating, the approvals). Distinct from reviews (earned and answered on Google, never encoded) and from source-verified facts (the Record's admission standard).

Modelling an organization as a set of entities and the relationships between them before anything is published: inventorying what exists, naming each thing one way, giving it a stable identifier, defining how it relates to the others and mapping it to external references.

The practice of declaring an organization as one entity with one identity across every surface where it appears (its site, its profiles, the registries and graphs that describe it), so that every mention resolves to the same thing.

A 20-character alphanumeric code (ISO 17442) that uniquely identifies a legal entity, issued through the Global LEI System overseen by GLEIF, with a public record of who the entity is and who owns it.

The practice of collecting evidence for an organization's claims and publishing it in a form a machine can read as proof: structured, linked to the claim it supports and traceable to its source.

A published, machine-readable unit of evidence (a case study, a project spotlight, a credential, a documented result) linked to the claims it supports, so that a machine can follow the link from claim to proof.

Building a machine-readable business identity with sufficient depth, breadth, and fidelity that it holds up under multi-source verification, by any agentic system, in any context.

The system that holds the authoritative version of a given set of facts, to which other systems defer when versions disagree.

The teaching ladder for machine trust: Legible, Resolvable, Verifiable, Corroborated, Comprehensible, Proven, Current; the output is confidence, enough of it to recommend.

Symbolic layer (entity; what you declare and control) and probabilistic layer (brand; what AI infers from natural language). Without the symbolic layer, you cannot be verified. Without the probabilistic layer, you can be verified but never recommended.

Digitally signed, machine-verifiable statements about an entity (its identity, a role, a qualification), issued by a party the verifier trusts, that can be checked for integrity and authenticity without contacting the issuer.

The practice of declaring a business as one resolvable entity, verified against every record the world already holds on it, so machines can identify it before they reason about it.

The state in which the facts about a business are identical, not merely similar, on every surface where they appear, maintained by synchronizing each surface to the record.

The distance between what is true of a business and what a machine can be certain of: facts that exist but are undeclared, unverifiable, inconsistent across sources, or stated so generically that they do not distinguish the business.

Agreement on a fact across independent sources: an organization's own record, its profiles, its reviews and third-party mentions stating the same thing.

A business holds real, hard-won expertise that machine systems cannot see, verify, or reason about.

Enough corroborated edges around a claim that a machine asserts it rather than hedges.

A machine attributes a claim to the wrong entity, or invents one, because the real relationships were never declared.

The state in which an AI system holds enough verified, corroborated and current fact about a business that it resolves the business as one entity and asserts what it knows rather than hedging.

Appearing among the local business results (the map and the three listings beneath it) that Google shows for a query with local intent.

Checking a claim against the record kept by the party that makes it, published on that party's own domain, rather than against third-party descriptions of it.

Multi-location personal care and grooming services.

The sector of contractors and builders, residential and commercial, including the service and retail lines that often sit beside the build.

Member-owned, cooperatively governed financial institutions, usually operating several branches, whose values form part of their identity.

Credit unions, banking, financial advisory.

The sector of organizations that deliver, supply or support health care: health systems, hospitals and clinics, and the equipment, product and service suppliers around them.

The prevention of infection in healthcare and other settings through engineered controls: disinfection and sterilization equipment and systems, and the distribution and service behind them.

Manufacturers, precision machining and industrial fabrication.

Distribution and professional services for healthcare equipment.

The sector of grooming and personal services (barbering, hairdressing, aesthetics), often operated across several locations.

Consulting, legal, accounting and engineering firms: businesses that sell expertise.

Pool and recreational facility construction, service and retail.

Custom signage, electronic message centres, architectural signage, wayfinding.