Audiences rarely meet a company directly. They meet it through the sources they trust, the summaries they read, and increasingly the answers a model gives them. These are the ideas SFG uses to explain what changes, what still matters, and where companies need to be clearer.
The trusted intermediary is the party an audience relies on to decide what is true. Communications is the discipline of delivering a company's signal to its audience through that intermediary - it is not the intermediary itself.
Read idea →Share of Model is a measure of how often, and how accurately, an AI model represents a company when it answers. It is the successor to share of voice for an AI-mediated market.
Read idea →The invisible shortlist is the set of people, companies, and sources an AI model considers before it answers, even when no ranked list appears to the user.
Read idea →A message house is the structured map of a company's position, proof, audience logic, and repeatable language. In the model era, it has to be legible to people and machines.
Read idea →Semantic triples are subject-predicate-object statements that make a relationship clear. They help people and systems carry the same meaning from one page or answer to the next.
Read idea →The set of operations an AI system performs between a human question and a human decision: retrieval, classification, comparison, narrowing, routing, and in supported contexts execution.
Read definition →The discipline of making a product's legitimate decision context machine-legible enough for AI systems to classify, compare, constrain, and route it accurately.
Read definition →The precondition, upstream of the social license, that an institution's meaning can be accurately summarized and attributed by the AI systems that now mediate between it and its stakeholders.
Read definition →Optimizing your content and entity so AI answer engines retrieve and repeat you as the answer, rather than ranking a link.
Read definition →The AI-mediated surfaces where questions are answered by synthesis from retrieved sources — assistants, answer engines, AI search overviews and agents — rather than by ranked links.
Read definition →The structural equivalence between brains and language models as surprise-minimizing prediction engines, making low-surprisal language optimal for human trust and machine retrieval at once.
Read definition →A biologic highly similar to an already-approved reference biologic, with no clinically meaningful differences in safety or effectiveness; distinct from a generic, which is a bioequivalent copy of a chemically synthesized drug.
Read definition →Source information transformed into lower-load, higher-structure output — the work cHP measures.
Read definition →Output completed under a defined policy, risk, provenance, and approval profile.
Read definition →The degree to which the complete, substantiated decision context survives AI synthesis and the subsequent operations of inclusion, comparison, constraint, and routing.
Read definition →The founder-specific practice of becoming the cited answer when buyers, press, and partners ask AI about your category.
Read definition →Shaping how generative models represent and recommend you when they synthesize an answer about your category.
Read definition →The discipline of keeping institutional meaning intact as it passes through AI systems, by placing a closed control loop around the model that compares outputs to an authorized reference and corrects drift.
Read definition →Supplying stakeholders ready-to-use, accurate-but-framed information to reduce information asymmetry and shape the decision environment.
Read definition →The societal permission an institution needs to operate; legal ownership alone has never been sufficient. Won by closing the gap between institutional reality and public belief.
Read definition →The discipline of treating institutional language as structured, traceable, verifiable data — engineering the signals AI systems read so that inference produces accurate representation.
Read definition →Signal Fidelity Group's control architecture that compares AI-mediated outputs against authorized reference signals — correcting or escalating drift — to make institutional meaning enforceable rather than aspirational.
Read definition →The point at which a pharmaceutical product's patent and regulatory protections end and generic or biosimilar competition can enter the market.
Read definition →The point at which the market — and the AI models that now mediate it — repeat your company's framing of its category accurately and unprompted.
Read definition →Structured catalog data supplied to a commerce or discovery system. A feed serializes an existing decision model; it cannot create or repair one.
Read definition →Drafting speed — the rate of fluent candidate output, before any verification.
Read definition →The time, attention, and risk absorbed verifying probabilistic outputs before release.
Read definition →An adversarial action that exploits language as an attack surface, injecting signals that look legitimate but carry intent designed to corrupt an AI system's synthesis and behavior.
Read definition →Measurable displacement between authorized meaning and an AI system's reproduction of it, quantified as cosine distance in vector space.
Read definition →The end-to-end flow of claims about your company — from your own content through third-party corroboration — that determines the triples AI ends up holding about you.
Read definition →A unit of machine-readable fact in the form subject–predicate–object (e.g. 'Acme — is — the leading X'), the form knowledge graphs and language models use to store and retrieve facts.
Read definition →The reduced option set an AI intermediary presents to a decision-maker. Inclusion in the shortlist, and the terms of comparison within it, determine commercial outcomes upstream of any transaction.
Read definition →Whether the meaning that arrives still carries the intent, evidence, and authority of the meaning that was sent.
Read definition →A communications asset structurally designed to seed an AI knowledge graph with precisely defined coordinates of identity, utility, and credibility.
Read definition →Surprisal is the improbability of the next word given context; perplexity is its computational twin in a language model. Lower values mean easier processing for both.
Read definition →Being the asset an AI agent selects and acts on — verifiable, machine-legible, and provenance-tracked.
Read definition →The precondition for institutional legitimacy in the AI era: if the inference engines that summarize an institution to its stakeholders cannot represent it accurately, the institution loses access to the conversation in which its social license is granted.
Read definition →The output-maximizing communications model valid only while information was scarce and attention abundant; obsolete after generative AI.
Read definition →The shift in which content becomes infinite and free while attention stays fixed, turning content into entropy and moving value from production to verification.
Read definition →Being what the model already believes before it searches — earned through dense, consistent presence in training-grade sources.
Read definition →Winning the cited answer in live answer-engine responses — through trusted earned media and clean, structured, liftable content.
Read definition →Authoring and corroborating the subject–predicate–object facts you want AI to associate with you, instead of letting models infer them from scraps.
Read definition →Output that passes evidence, factuality, and semantic-preservation checks.
Read definition →