MoroAK — Professional Authority Publishing
    Back to Directory
    Business

    AK-04 - The Architecture of Professional Discovery - MoroAK

    Evelyse Carvalho-Ribas6 Aug 2026

    Track Your Progress

    Sign in to earn XP, take quizzes, and unlock AI tools.

    Sign In

    AK-04 · DISCOVERY ARCHITECTURE · AUTHORITY INFRASTRUCTURE

    The Architecture of Professional Discovery

    AI Cards, Brand Memory, and the Retrieval Layer That Surfaces Expertise

    Series AK-04
    Domain DISCOVERY ARCHITECTURE · AUTHORITY INFRASTRUCTURE
    Year 2026
    Platform moroak.com

    Abstract

    AK-01 named the professional authority crisis. AK-02 specified the four-layer retrieval stack through which AI systems evaluate professional authority. AK-03 mapped the six educator archetypes that compound in that stack. AK-04 opens the layer that sits on top of all three: the discovery surface itself — the AI cards a user actually sees, the brand memory that makes a given professional retrievable over time, and the architecture that turns atomic cards and persistent memory into a coherent act of professional discovery.

    We advance a three-layer architecture. AI cards are the atomic unit: entity-centred, typed, provenance-carrying knowledge objects descended from the knowledge-graph and entity-linking literature (Singhal; Bollacker et al.; Dong et al.; Milne & Witten; Ferragina & Scaiella).[1][2][3][5][6] Brand memory is the persistence layer: the structured, reinforced associations a professional accumulates across training cycles and retrieval indexes, drawing on Keller’s customer-based brand equity framework, Aaker’s brand-asset model, and Krishnan’s memory-association work.[13][14][15][16] Professional discovery is the emergent system: the retrieval-and-generation pipeline through which those cards and that memory are surfaced to a querying user, descended from dense retrieval, RAG, and citation-grounded generation (Lewis et al.; Karpukhin et al.; Nakano et al.; Menick et al.; Edge et al.).[8][9][10][11][12]

    The three layers interact. Cards without persistent memory are ephemeral artefacts that do not reinforce authority. Memory without structured cards degrades into diffuse brand impression that cannot ground citations. Discovery without either is keyword matching. A professional who publishes under MoroAK is, in architectural terms, producing cards that feed brand memory that drives discovery.

    The paper specifies (i) a typology of AI cards — entity, synthesis, provenance, comparison, instructional — and the retrieval properties of each; (ii) the encoding, decay, and reinforcement dynamics of brand memory under model retraining, retrieval reindexing, and information-foraging user behaviour; and (iii) the integration with AK-02’s four-layer stack and AK-03’s six archetypes, so each archetype’s publishing cadence can be translated into card-and-memory terms. AK-04 is the interface paper between retrieval mechanics (AK-02), production profiles (AK-03), and the artefact format analyses that follow in AK-05 and beyond.

    Keywords: AI cards  ·  entity cards  ·  brand memory  ·  professional discovery  ·  retrieval-augmented generation  ·  knowledge graphs  ·  citation grounding  ·  customer-based brand equity  ·  information foraging  ·  authority architecture

    DEFINITIONS — Core Concepts and Workable Definitions

    AI CARD
    A structured, entity-centred knowledge object surfaced by an AI system in response to a query or embedded in a generated answer. Descended from knowledge-graph entity cards in the sense of Singhal (2012) — "things, not strings" —[1] and from the structured entity representations pioneered by Freebase and Knowledge Vault.[2][3] Contains: entity identifier, type, canonical claims, provenance links, and often a display surface.
    ENTITY CARD
    A sub-type of AI card centred on a single named entity — a person, firm, jurisdiction, instrument, or concept. The entity-linking literature (Milne & Witten; Ferragina & Scaiella) specifies how ambiguous query text is resolved to a specific entity card.[5][6]
    BRAND MEMORY
    The persistent, structured associations a professional or institution accumulates across (i) the training corpora on which foundation models are pre-trained, (ii) the retrieval indexes used at inference time, and (iii) the published artefact graph that reinforces both. After Keller’s customer-based brand equity framework: brand memory is the node-and-association structure in which a name, when cued, activates a distinctive set of attributes.[13][15]
    PROFESSIONAL DISCOVERY
    The end-to-end process by which a user (human or agentic) issues a query to an AI-mediated system and arrives at a specific professional as the answer or citation. It is an emergent property of cards + memory + retrieval pipeline — not of any single layer alone.
    RETRIEVAL-AUGMENTED GENERATION (RAG)
    The architectural pattern in which a generative model’s output is conditioned on documents retrieved from an external corpus at inference time.[8] The retrieval component typically uses dense vector matching (DPR and successors)[9] and may include web browsing, verified-quote extraction, or graph-level aggregation.[10][11][12]
    INFORMATION FORAGING
    After Pirolli & Card (1999): a cognitive-ecology model of how users navigate information environments, adapting their strategy to the information scent of available paths. Directly applicable to how users interact with AI cards and decide whether to follow a citation, expand a card, or reformulate the query.[17]
    CARD-MEMORY-DISCOVERY STACK
    The three-layer architecture introduced in this paper. Cards are the atomic unit; brand memory is the persistence layer; discovery is the emergent system. The stack sits above AK-02’s four-layer evaluation stack and is the surface on which AK-03’s archetypes actually compound.

    Conceptual Boundaries and Scope of the Paper

    This paper specifies an architecture, not a specific platform. The argument holds across retrieval-augmented generative systems that combine a card-like structured layer, a memory-like persistent layer, and a discovery-like retrieval pipeline. Specific vendor implementations differ; the architectural claims do not.

    We deliberately bridge the machine-learning and brand-management literatures. This is unusual but necessary: the retrieval literature has a precise account of cards and no account of persistence across model generations; the brand-equity literature has a precise account of memory and no account of retrieval-indexed surfacing. Neither field alone is sufficient to describe professional discovery today.

    Scope: professional domains in which AI-mediated discovery is already non-trivial — law, tax, finance, digital assets, policy, medicine, cultural economics, specialised technology. Consumer product discovery and mass-media brand dynamics are adjacent but are governed by different retrieval and attention economics.

    We do not claim that cards, memory, or discovery are new phenomena. Knowledge-graph cards precede LLMs by a decade; brand memory precedes search engines by a century. The novelty is their integration inside a retrieval-augmented generative system — and the consequences for how professional authority must now be produced.

    Key Questions

    Four questions frame the architecture analysis.

    What is the atomic unit of AI-mediated professional discovery, and what structural properties does it require?

    How does brand memory persist across model retraining, retrieval reindexing, and shifting discovery surfaces?

    How does the card-memory-discovery stack integrate with AK-02’s retrieval evaluation stack and AK-03’s educator archetypes?

    What must an educator actually produce — in cards, in reinforcement cadence, in provenance links — to compound in this architecture?

    SECTION I

    Three Layers — Cards, Memory, Discovery

    The three-layer argument is not a metaphor. Each layer corresponds to a distinct literature, a distinct set of engineering constraints, and a distinct kind of professional production. We introduce them separately before examining their interaction.

    The card layer — atomic, entity-centred, typed

    The idea of a card as the atomic unit of machine-readable knowledge is traceable to the knowledge-graph programme. Singhal’s 2012 announcement of the Google Knowledge Graph framed the shift in exactly the terms still used today: "things, not strings."[1] Under the hood, Freebase had already demonstrated a collaboratively-created graph database of structured entity nodes and typed relationships;[2] Knowledge Vault then showed that large-scale probabilistic fusion across heterogeneous sources could produce a web-scale card index,[3] and YAGO2 added systematic spatio-temporal grounding to each entity record.[4] Each of these systems treats the world as an index of typed entities with attached claims, not an index of documents.

    The entity-linking literature specifies how a noisy query or a passage of text is resolved to a specific card. Milne and Witten’s Wikification work established the baseline of learning to link natural-language mentions to Wikipedia entities;[5] Ferragina and Scaiella’s TAGME extended this to real-time annotation of short, noisy text fragments;[6] Gabrilovich and Markovitch’s Explicit Semantic Analysis represented entities in high-dimensional semantic space so that relatedness could be computed directly.[7] Together they answer the question: given a professional query, which card is the correct one to surface?

    The memory layer — persistence across retraining

    Where the card layer is architectural and engineering, the memory layer is behavioural and associative. The brand-equity literature is where this has been worked out most precisely. Keller’s 1993 customer-based brand equity framework defines a brand as a network of associative memory: a node (the name) linked to a distinctive set of attribute and evaluation nodes, activated when the name is cued.[13] Aaker’s brand-asset model operationalises this as awareness, perceived quality, associations, and loyalty — an intangible asset that persists across cycles.[14] Krishnan’s empirical work measured the strength, uniqueness, and origin of brand-memory associations and showed that these properties map onto differential retrieval under cueing.[15] Keller’s CBBE pyramid then formalised the stages of memory building from awareness through meaning through response to resonance.[16]

    We adopt this framework without modification at the level of professional authority. A professional’s brand memory is the associative structure that activates when their name — or the class of problem they address — is cued to an AI system. That memory is encoded in (i) the pre-training corpus of foundation models, (ii) the retrieval indexes that augment those models at inference, and (iii) the published artefact graph that reinforces both. It decays under retraining without reinforcement and compounds under structured, cross-referenced publishing.

    The discovery layer — emergent from cards + memory + retrieval

    Discovery is what the user experiences. The architecture beneath it is the retrieval-augmented generation pipeline Lewis et al. introduced in 2020: a generative model conditioned on documents retrieved from an external corpus at inference time.[8] The retrieval component is typically a dense-vector matcher such as DPR,[9] sometimes extended to browser-based search (WebGPT)[10] or to verified-quote extraction (GopherCite)[11] or to graph-level community summarisation over hierarchical card structures (GraphRAG).[12] Pirolli and Card’s information-foraging theory describes the user side: the cognitive-ecology model of how users navigate an information environment, adapting their strategy to the information scent of available paths.[17] Put together, discovery is the emergent outcome when a foraging user encounters a retrieval pipeline that is drawing from a card index that is in turn shaped by brand memory.

    If cards are atomic and memory is persistent, what exactly does the educator produce to ensure discovery surfaces them rather than a competitor?

    SECTION II

    The Anatomy of an AI Card

    Not all AI cards are alike. A working taxonomy — drawn from current observable behaviour in retrieval-augmented systems and grounded in the knowledge-graph and citation-generation literatures — distinguishes five functional types. Each has different retrieval properties and is produced by different kinds of published work.

    Five functional card types
    • 1. ENTITY CARD

      Centred on a named entity (person, firm, jurisdiction, instrument, concept). Surfaces canonical attributes and disambiguates ambiguous queries. Produced by: authoritative profile-level content, structured CVs, canonical method descriptions.

    • 2. SYNTHESIS CARD

      Summarises a multi-source position on a topic. Surfaces a structured take that integrates across a field. Produced by: cross-domain frameworks, field reviews, structured taxonomies — the Synthesiser archetype’s natural output.

    • 3. PROVENANCE CARD

      Carries the citation chain supporting a claim. Surfaces the "because X said Y" pathway that grounds a generated answer. Produced by: properly cited structured PDFs with stable URIs and well-formed bibliographies.

    • 4. COMPARISON CARD

      Presents a side-by-side structured comparison (jurisdictions, instruments, approaches). Especially frequent in legal, tax, and technical domains. Produced by: tables, matrices, comparative frameworks published as structured content.

    • 5. INSTRUCTIONAL CARD

      Supplies an actionable procedure or decision framework. Surfaces "how to do X" answers. Produced by: structured implementation guides, decision trees, named-test frameworks.

    Retrieval properties of each card type

    The card types are not retrieved identically. Entity cards are surfaced by name-bearing queries and by any query that implicitly disambiguates to an entity; they depend on entity-linking precision (Milne & Witten; Ferragina & Scaiella)[5][6] and on strong brand-memory associations that mark the entity as canonical for its class.[15] Synthesis cards are surfaced by topic-level queries and are rewarded by graph-structured retrieval that can aggregate a hierarchical card community (GraphRAG).[12] Provenance cards are surfaced when the generator is instructed to cite — increasingly the default behaviour in consumer and enterprise AI products — and are rewarded by the quote-grounding discipline described in GopherCite.[11] Comparison cards are surfaced by implicitly or explicitly contrastive queries and are rewarded by well-formed tables and matrices that dense retrievers such as DPR can index cleanly.[9] Instructional cards are surfaced by how-to queries and are rewarded by named, structured procedures with clear steps.

    Card composition — what a well-formed card contains

    ElementPurposeSource-of-truth literature
    Entity identifierCanonical name and disambiguating contextSinghal 2012; Milne & Witten 2008
    TypeCard class (entity, synthesis, provenance, comparison, instructional)Bollacker et al. 2008; Hoffart et al. 2013
    Canonical claimsThe small set of factual attributes a retrieval system can surfaceDong et al. 2014 (Knowledge Vault)
    Provenance linksStable URIs to source documents supporting each claimMenick et al. 2022 (GopherCite)
    Cross-referencesLinks to adjacent cards in the same graph / corpusEdge et al. 2024 (GraphRAG)
    Display surfaceShort, retrievable summary suitable for in-answer embeddingSinghal 2012; Lewis et al. 2020 (RAG)

    Which of your existing published artefacts already behave as well-formed cards, and which are missing provenance, cross-references, or a display surface?

    SECTION III

    Brand Memory as the Persistence Layer

    A card surfaced once is a lucky retrieval. A card that reliably surfaces across models, queries, and time is evidence of brand memory. The persistence layer is what turns an artefact into authority.

    Encoding — where brand memory lives

    Brand memory of a professional or institution lives in three places in the current AI stack: (i) the pre-training corpora used to train foundation models (public web, licensed books, Wikipedia, structured knowledge bases such as Wikidata and their derivatives); (ii) the retrieval indexes used at inference (vector databases, web search engines, enterprise document stores); and (iii) the cross-referenced artefact graph that connects published work to itself. Keller’s associative-memory model applies at each layer: a name cues a set of attributes with varying strength, uniqueness, and favourability.[13][15] In AI systems, this plays out as: named queries return a canonical card; topic-level queries return the professional’s work in the relevant cell of the authority graph; citation-grounded answers include the professional’s stable provenance links.

    Decay — what erodes brand memory

    Brand memory is not permanent. Three decay dynamics matter specifically for professional authority in an AI-mediated environment.

    Three decay dynamics
    • RETRAINING DILUTION

      Each foundation-model retraining cycle draws on a new snapshot of the web and licensed corpora. A professional whose published work is not refreshed is progressively diluted by newer content from competitors.

    • REINDEXING DRIFT

      Retrieval indexes are periodically rebuilt. Changes in embedding models, index structure, or scoring can reduce the retrievability of cards that were well-indexed under a prior regime.

    • ASSOCIATION DEGRADATION

      When a professional’s name is cued less often in fresh content, the strength and distinctiveness of their associated attributes weaken — the brand-equity literature’s standard decay dynamic.

    Reinforcement — what compounds brand memory

    The CBBE pyramid specifies the stages of brand-memory building: salience, performance, imagery, judgments, feelings, resonance.[16] In professional authority terms: (i) salience = being surfaced at all when a relevant query is issued; (ii) performance / imagery = being associated with the right class of problem; (iii) judgments / feelings = being cited with the right kind of authority, not just with the right frequency; (iv) resonance = being the default citation target in downstream work. Reinforcement across these stages is produced by the same mechanism: structured, cross-referenced publishing at a sustained cadence, across artefact types that feed each of the five card classes.

    LayerWhat encodesWhat erodesWhat reinforces
    Pre-training corporaPublic indexed web presence, Wikipedia / Wikidata, licensed archivesRetraining dilution; corpus-curation shiftsStable URIs; Wikipedia / Wikidata-grade authority; licensed archive deposits
    Retrieval indexesStructured documents with clean embeddings; well-formed cardsReindexing drift; embedding-model changeConsistent structured output; schema discipline; clean anchors
    Artefact graphCross-references inside a published corpus; citations receivedAssociation degradation when cadence dropsCross-referenced AK-series; citing prior work explicitly; being cited

    In which of the three encoding layers is your current brand memory weakest, and what reinforcement cadence would restore it?

    SECTION IV

    The Emergent Architecture — Integrating Cards, Memory, Discovery with AK-02 and AK-03

    The card-memory-discovery stack does not replace AK-02’s four-layer retrieval evaluation stack (retrieval layer, source priors, constitutional calibration, citation grounding). It sits above it, on the discovery surface, and interacts with it at every layer. Nor does it replace AK-03’s six archetypes; it specifies what each archetype actually produces in card-and-memory terms.

    Integration with the AK-02 retrieval stack

    AK-04 layerAK-02 layer it most directly drivesMechanism
    Cards (atomic unit)Retrieval layerWell-formed cards are indexed cleanly by DPR-class retrievers; ambiguous or unstructured text is not.
    Cards + provenanceCitation groundingProvenance links are exactly what quote-grounded generators need to cite (GopherCite dynamic).
    Brand memorySource priorsRepeated, cross-referenced presence across corpora raises the prior on a source being authoritative.
    Discovery (emergent)Constitutional calibrationWhen the model decides whether to trust a source, it is integrating card precision, memory strength, and retrieval confidence simultaneously.

    Integration with the AK-03 archetypes

    Each AK-03 archetype has a signature card mix. We specify the combination that compounds fastest for each.

    ArchetypeDominant card typeSecondary card typesMemory reinforcement strategy
    Domain SynthesiserSynthesis cardEntity; comparisonQuarterly structured authority PDFs cross-referenced into a corpus graph
    Jurisdictional SpecialistComparison cardProvenance; entityRegulatory-change-driven cadence with stable URIs per jurisdiction
    Cross-Domain IntegratorSynthesis cardComparison; entityCross-domain frameworks that bridge card communities across graphs
    Methodology TranslatorInstructional cardSynthesis; entityNamed frameworks and decision procedures that become default citations
    Frontier NavigatorProvenance cardEntity; synthesisRapid-cadence structured commentary with heavy source citation
    Cohort ArchitectEntity card (cohort-native)Instructional; comparisonDistributed authorship that produces many linked cards reinforcing a single cohort brand

    Information foraging and the user side

    All of the above is on the supply side. On the demand side, Pirolli and Card’s foraging model predicts user behaviour in a card-dense environment: users follow the scent of relevance, expand cards with high expected yield, and abandon low-scent paths quickly.[17] For professionals, the implication is exact: a card that does not carry strong relevance scent (clear entity, typed claims, visible provenance) will be passed over even if it is technically retrievable. Discovery is not only about being surfaced; it is about being surfaced in a form that the foraging user will actually follow.

    For your archetype, are the dominant card types you produce carrying sufficient relevance scent for the foraging user, and do they reinforce each other across the corpus graph?

    SECTION V

    Implications for Educator Practice

    If the three-layer architecture is accurate, educator practice cannot be optimised by attending to any single layer. Publishing more artefacts without card structure produces cards that do not compound memory. Building name recognition without structured artefacts produces memory that does not ground citations. Gaming retrieval without either produces neither. Four specific implications follow.

    Implication 1 — Publish for cards, not for prose

    The artefact that compounds is the one that decomposes cleanly into cards. This is not a stylistic preference; it is a direct implication of how retrieval indexes and quote-grounded generators operate on documents. Structured PDFs with named sections, typed tables, explicit entities, and stable URIs decompose well; long-form undifferentiated prose does not. The discipline is to produce work that a dense retriever can chunk without losing the underlying structure of your argument.[9]

    Implication 2 — Reinforce memory at the three layers, not one

    A professional who appears in licensed archives but publishes nothing fresh is encoded in pre-training corpora but missing from retrieval indexes. A professional who publishes frequently on a personal site but does not cross-reference their work is in retrieval indexes but not in the artefact graph. The three encoding layers require different reinforcement mechanisms — corpus-grade authority, structured consistent output, and explicit cross-referencing — and the three must be reinforced together.

    Implication 3 — Provenance is part of the product

    In a world where generators are increasingly required to cite,[11] the provenance card is not a by-product of scholarly work; it is the primary product in retrieval terms. Stable URIs, correctly formed footnotes, DOIs where available, and a clean bibliography are not ornamental — they are the raw material from which citation-grounded discovery constructs its answers.

    Implication 4 — Measure discoverability, not output

    Output metrics (number of PDFs, word count, download counts) are weakly correlated with discovery. Useful measurement asks whether the right cards surface for the right queries: does your name return a canonical entity card? Does the query that defines your domain return your synthesis? Are your structured claims cited verbatim in generated answers? These are the signals of a card-and-memory architecture that is actually compounding, and they are the ones that information-foraging users respond to in practice.[17]

    EducatorsLearnersInstitutions
    Shift from prose-first to card-first: structured PDFs, named frameworks, explicit typed tables, stable URIs, disciplined cross-references.Look for educators whose corpus decomposes into well-formed cards across the five types; those are the corpora that compound under AI.Audit institutional archives against the three encoding layers; invest in what is weakest, and measure discoverability, not throughput.

    Given your current corpus, which of the four implications represents the highest-leverage change in the next six months?

    Practical Implementation

    For Educators

    Map your existing corpus to the five card types

    For each published artefact, identify which card type(s) it produces. A corpus that only produces one or two types has structural gaps that weaken brand memory. Plan new artefacts to fill the missing card classes.

    Adopt a structured-PDF discipline

    Publish in a format that decomposes cleanly: named sections, typed tables, explicit entities, stable URIs, clean bibliography. The MoroAK AK-series format is designed for exactly this.

    Reinforce memory at all three encoding layers

    Check presence in (i) pre-training corpora via licensed or canonical archives, (ii) retrieval indexes via consistent structured output, and (iii) the artefact graph via explicit cross-references. Address the weakest layer first.

    Measure discoverability, not output

    Build a small test bank of queries that represent your domain. Run them against current AI systems quarterly. Track whether the right cards surface, not how many artefacts you produced.

    For Learners

    Prefer educators whose corpora produce well-formed cards

    A Synthesiser whose work decomposes into clean synthesis and comparison cards will surface in your future AI-mediated work far more reliably than one whose work is undifferentiated prose.

    Build your own portfolio as cards from day one

    Case notes, structured implementations, named frameworks — the same card discipline that compounds for educators compounds for learners. Start early.

    Use AI-mediated discovery as a diagnostic

    If the query that defines your target archetype does not return you after six months of structured publishing, something in your card-memory-discovery stack is misaligned. Diagnose, adjust, publish again.

    For Institutions

    Audit the institutional corpus layer by layer

    Pre-training presence, retrieval indexability, and cross-reference density are three distinct audits. Most institutions over-invest in the first and under-invest in the other two.

    Invest in card-native publishing infrastructure

    Migration from unstructured long-form publishing to structured, card-producing artefacts is a one-time architectural investment that compounds across every educator the institution supports.

    Measure institutional discoverability, not institutional voice

    Named educators on a platform drive institutional authority in AI-mediated discovery. Measure whether the right educator cards surface for the right queries; that is the institutional KPI the old metrics miss.

    Conclusion

    Professional discovery in an AI-mediated environment is a three-layer architecture. AI cards are the atomic unit, descended from the knowledge-graph and entity-linking literature. Brand memory is the persistence layer, descended from customer-based brand equity and associative-memory research. Discovery is the emergent system, descended from retrieval-augmented generation and information-foraging theory. Each layer has its own engineering and behavioural constraints, and the three layers interact at every retrieval.

    The architecture sits above AK-02’s four-layer retrieval evaluation stack and specifies what AK-03’s archetypes actually produce in card-and-memory terms. The educator who publishes structured, provenance-bearing, cross-referenced artefacts is not simply producing content; they are building a card index that reinforces brand memory that drives discovery. The educator who publishes unstructured prose is producing artefacts that the retrieval layer cannot cleanly index, that the memory layer cannot cleanly associate, and that the discovery layer cannot cleanly surface.

    AK-05 turns to the structural properties of the primary artefact this architecture requires — the structured authority PDF — and specifies why its format advantages compound in AI-mediated discovery. AK-06 develops the authority graph across educator corpora. AK-07 applies the full framework to the first domain-specific case: legal and tax expertise, where the card-memory-discovery dynamics are already most visible.

    This Authority PDF is published by MoroAK Professional Knowledge Infrastructure. MoroAK is a controlled hybrid model operating as SaaS Infrastructure, Educational Marketplace, and Technology Intermediary — not advisory. All educator profiles, structured courses, cohorts, trainings, and authority PDFs are available through moroak.com. Platform onboarding, institutional licensing, and mandate scoping: contact through moroak.com.

    FREQUENTLY ASKED

    Frequently asked

    What three layers make up the architecture of professional discovery?

    Cards, Memory, and Discovery. The paper specifies an architecture, not a specific platform: the argument holds across retrieval-augmented generative systems that combine a card-like structured layer, a memory-like persistent layer, and a discovery layer.

    What is the card layer?

    The card is the atomic, entity-centred, typed unit of machine-readable knowledge. The idea is traceable to the knowledge-graph programme — Singhal's 2012 announcement of the Google Knowledge Graph framed the shift toward entities as the unit of machine understanding.

    What is the memory layer?

    Where the card layer is architectural and engineering, the memory layer is behavioural and associative — persistence across retraining. The paper grounds it in the brand-equity literature, notably Keller's 1993 customer-based brand-equity work, where associative memory has been worked out most precisely.

    Why does this architecture matter for educators?

    Because durable discoverability depends on building both a structured card presence and a persistent brand memory that survive model retraining and surface at the discovery layer. The paper sets out the implications for educator practice on that basis.

    What is this paper for, and what is MoroAK's role?

    AK-04 gives educators the architecture behind how professionals are discovered by AI systems. It is part of the MoroAK AK-series, the platform's public authority layer for professional knowledge in AI-mediated markets.

    WORK WITH MOROAK

    MoroAK platform infrastructure is designed to turn educator output into well-formed cards that reinforce brand memory and surface in AI-mediated professional discovery — across every layer of the three-layer architecture specified in AK-04.

    → AK-01 — THE PROFESSIONAL AUTHORITY CRISIS— Why traditional signals of professional authority are losing discriminatory power, and the Invisible Middle cohort for whom authority infrastructure is strategically decisive.

    → AK-02 — HOW AI SYSTEMS EVALUATE PROFESSIONAL AUTHORITY— The four-layer evaluation stack (retrieval, source priors, constitutional calibration, citation grounding) that AK-04’s three-layer discovery architecture sits above.

    → AK-03 — EDUCATOR ARCHETYPES— Six structural profiles that compound authority; AK-04 specifies the signature card mix and memory reinforcement strategy for each.

    → AK-04 — THE ARCHITECTURE OF PROFESSIONAL DISCOVERY (THIS PAPER)— AI cards, brand memory, and the retrieval layer that surfaces expertise — the three-layer discovery architecture.

    → → EDUCATORS— Publish under MoroAK in card-native structured formats. Reinforce brand memory at the pre-training, retrieval, and artefact-graph layers simultaneously. Measure discoverability, not output. Apply to the educator track →

    → → LEARNERS— Follow educators whose corpora produce well-formed cards across the five functional types. Build your own portfolio as cards from day one. Explore learner cohorts →

    → → INSTITUTIONS— Audit the corpus layer by layer. Invest in card-native publishing infrastructure. Measure institutional discoverability — the KPI traditional content metrics miss. Talk to institutional partnerships →

    moroak.com · Platform onboarding and institutional enquiries: moroak.com

    Footnotes

    1. Singhal, A. (2012). Introducing the Knowledge Graph: things, not strings. Google Official Blog, May 16. https://blog.google/products/search/introducing-knowledge-graph-things-not/
    2. Bollacker, K., Evans, C., Paritosh, P., Sturge, T., & Taylor, J. (2008). Freebase: A collaboratively created graph database for structuring human knowledge. Proceedings of SIGMOD 2008, 1247–1250. https://doi.org/10.1145/1376616.1376746
    3. Dong, X., Gabrilovich, E., Heitz, G., Horn, W., Lao, N., Murphy, K., Strohmann, T., Sun, S., & Zhang, W. (2014). Knowledge Vault: A web-scale approach to probabilistic knowledge fusion. Proceedings of KDD 2014, 601–610. https://doi.org/10.1145/2623330.2623623
    4. Hoffart, J., Suchanek, F. M., Berberich, K., & Weikum, G. (2013). YAGO2: A spatially and temporally enhanced knowledge base from Wikipedia. Artificial Intelligence, 194, 28–61. https://doi.org/10.1016/j.artint.2012.06.001
    5. Milne, D., & Witten, I. H. (2008). Learning to link with Wikipedia. Proceedings of CIKM 2008, 509–518. https://doi.org/10.1145/1458082.1458150
    6. Ferragina, P., & Scaiella, U. (2010). TAGME: On-the-fly annotation of short text fragments by Wikipedia entities. Proceedings of CIKM 2010, 1625–1628. https://doi.org/10.1145/1871437.1871689
    7. Gabrilovich, E., & Markovitch, S. (2009). Wikipedia-based semantic interpretation for natural language processing. Journal of Artificial Intelligence Research, 34, 443–498. https://doi.org/10.1613/jair.2669
    8. Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., et al. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems, 33. https://arxiv.org/abs/2005.11401
    9. Karpukhin, V., Oğuz, B., Min, S., Lewis, P., Wu, L., Edunov, S., Chen, D., & Yih, W.-T. (2020). Dense passage retrieval for open-domain question answering. Proceedings of EMNLP 2020, 6769–6781. https://aclanthology.org/2020.emnlp-main.550/
    10. Nakano, R., Hilton, J., Balaji, S., Wu, J., Ouyang, L., et al. (2021). WebGPT: Browser-assisted question-answering with human feedback. arXiv:2112.09332. https://arxiv.org/abs/2112.09332
    11. Menick, J., Trebacz, M., Mikulik, V., Aslanides, J., Song, F., Chadwick, M., et al. (2022). Teaching language models to support answers with verified quotes (GopherCite). arXiv:2203.11147. https://arxiv.org/abs/2203.11147
    12. Edge, D., Trinh, H., Cheng, N., Bradley, J., Chao, A., Mody, A., Truitt, S., & Larson, J. (2024). From local to global: A Graph RAG approach to query-focused summarization. arXiv:2404.16130. https://arxiv.org/abs/2404.16130
    13. Keller, K. L. (1993). Conceptualizing, measuring, and managing customer-based brand equity. Journal of Marketing, 57(1), 1–22. https://doi.org/10.1177/002224299305700101
    14. Aaker, D. A. (1991). Managing Brand Equity: Capitalizing on the Value of a Brand Name. Free Press.
    15. Krishnan, H. S. (1996). Characteristics of memory associations: A consumer-based brand equity perspective. International Journal of Research in Marketing, 13(4), 389–405. https://doi.org/10.1016/S0167-8116(96)00021-3
    16. Keller, K. L. (2001). Building customer-based brand equity: A blueprint for creating strong brands. Marketing Management, 10(2), 14–21.
    17. Pirolli, P., & Card, S. (1999). Information foraging. Psychological Review, 106(4), 643–675. https://doi.org/10.1037/0033-295X.106.4.643

    Selected Bibliography

    AI Regulation: Primary Legislation

    European Parliament and Council of the European Union. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, L, 12 July 2024. https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
    UK Government, Department for Science, Innovation and Technology. (2023). A Pro-Innovation Approach to AI Regulation. White Paper, Cm 815, March 2023. https://www.gov.uk/government/publications/ai-regulation-a-pro-innovation-approach/white-paper
    UK Government, Department for Science, Innovation and Technology. (2024). A Pro-Innovation Approach to AI Regulation: Government Response to Consultation. CP 1019, February 2024. https://www.gov.uk/government/consultations/ai-regulation-a-pro-innovation-approach-policy-proposals/outcome/a-pro-innovation-approach-to-ai-regulation-government-response
    UK Government. (2025). AI Opportunities Action Plan. January 2025. Department for Science, Innovation and Technology.
    The White House. (2023). Executive Order 14110: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. 30 October 2023. Federal Register.
    The White House. (2025a). Executive Order 14179: Removing Barriers to American Leadership in Artificial Intelligence. 23 January 2025. Federal Register.
    The White House. (2025b). Executive Order: Ensuring a National Policy Framework for Artificial Intelligence. 11 December 2025. https://www.whitehouse.gov/presidential-actions/2025/12/eliminating-state-law-obstruction-of-national-artificial-intelligence-policy/

    Digital Assets and Financial Regulation: Primary Legislation

    European Parliament and Council of the European Union. (2023). Regulation (EU) 2023/1114 on markets in crypto-assets, and amending Regulations (EU) No 1093/2010 and (EU) No 1095/2010 and Directives 2013/36/EU and (EU) 2019/1937 (Markets in Crypto-Assets Regulation — MiCA). Official Journal of the European Union, 9 June 2023. https://www.esma.europa.eu/esmas-activities/digital-finance-and-innovation/markets-crypto-assets-regulation-mica

    International Tax and Cross-Border Frameworks

    OECD. (2021). Global Anti-Base Erosion Model Rules (Pillar Two). OECD/G20 Inclusive Framework on BEPS, 20 December 2021. OECD Publishing, Paris. https://www.oecd.org/en/topics/sub-issues/global-minimum-tax/global-anti-base-erosion-model-rules-pillar-two.html
    OECD. (2024). Pillar One — Amount B. OECD/G20 Base Erosion and Profit Shifting Project. OECD Publishing, Paris. https://www.oecd.org/en/publications/2024/02/pillar-one-amount-b_41a41e1e.html
    OECD. (2025a). Tax Administration 2025. Thirteenth edition. OECD Publishing, Paris. https://www.oecd.org/en/publications/tax-administration-2025_cc015ce8-en.html
    OECD. (2025b). Tax Challenges Arising from the Digitalisation of the Economy — Consolidated Commentary to the Global Anti-Base Erosion Model Rules (2025). OECD Publishing, Paris. https://www.oecd.org/en/publications/tax-challenges-arising-from-the-digitalisation-of-the-economy-consolidated-commentary-to-the-global-anti-base-erosion-model-rules-2025_a551b351-en.html
    OECD. (2025c). Tax Administration Digitalisation and Digital Transformation Initiatives. OECD Publishing, Paris. https://www.oecd.org/en/publications/tax-administration-digitalisation-and-digital-transformation-initiatives_c076d776-en.html

    Cultural Policy and Creative Industries Frameworks

    UNESCO. (2005). Convention on the Protection and Promotion of the Diversity of Cultural Expressions. Paris, 20 October 2005. https://www.unesco.org/creativity/en/2005-convention
    UNESCO. (2017). Guidelines on the Implementation of the Convention in the Digital Environment. Approved by the Conference of Parties, 2017. https://www.unesco.org/creativity/en/2005-convention
    UNESCO. (2019). Open Roadmap for the Implementation of the 2005 Convention in the Digital Environment. Approved by the Conference of Parties, 2019.

    AI Systems Research, Evaluation and Discovery

    Stanford University, Institute for Human-Centered AI (HAI). (2025). AI Index Report 2025. Eighth edition. Stanford, CA. https://hai.stanford.edu/ai-index
    Stanford University, Institute for Human-Centered AI (HAI). (2026). AI Index Report 2026. Ninth edition. Stanford, CA. https://hai.stanford.edu/ai-index
    Google. (2025). Search Quality Rater Guidelines. Updated September 2025. https://services.google.com/fh/files/misc/hsw-sqrg.pdf
    Previsible / ALM Corp. (2025). AI Discovery Report: What 1.96 Million LLM Sessions Reveal About the Future of Search and Marketing. https://almcorp.com/blog/previsible-2025-ai-discovery-report/
    ALM Corp. (2026). AI Discovery in 2026: What 2 Million LLM Sessions Tell Us About the Future of Search and Content Visibility. https://almcorp.com/blog/ai-discovery-2-million-llm-sessions-analysis-2026/
    Superlines. (2026). AI Search Statistics 2026: 60+ Data Points on Visibility, Citations, and Traffic. https://www.superlines.io/articles/ai-search-statistics/
    The Digital Bloom. (2025). 2025 AI Visibility Report: How LLMs Choose What Sources to Mention. https://thedigitalbloom.com/learn/2025-ai-citation-llm-visibility-report/
    Gao, Y., Xiong, Y., et al. (2024). Retrieval-Augmented Generation for AI-Generated Content: A Survey. Data Science and Engineering, Springer Nature. https://link.springer.com/article/10.1007/s41019-025-00335-5

    Anthropic and Claude AI: Official Research and Documentation

    Anthropic. (2026a). Responsible Scaling Policy, Version 3.0. Effective 24 February 2026. https://www.anthropic.com/responsible-scaling-policy
    Anthropic. (2026b). Claude Opus 4.6 System Card. Anthropic Transparency Hub. https://www.anthropic.com/transparency/model-report
    Anthropic. (2026c). Anthropic Transparency Hub. https://www.anthropic.com/transparency
    Anthropic. (2026d). Anthropic Economic Index Report: Learning Curves. March 2026. https://www.anthropic.com/research/economic-index-march-2026-report
    Anthropic. (2026e). Anthropic Economic Index Report: Economic Primitives. January 2026. https://www.anthropic.com/research/anthropic-economic-index-january-2026-report
    Anthropic. (2025a). Introducing the Anthropic Economic Index. https://www.anthropic.com/news/the-anthropic-economic-index
    Anthropic. (2025b). Labor Market Impacts of AI: A New Measure and Early Findings. https://www.anthropic.com/research/labor-market-impacts
    Anthropic. (2025c). Estimating AI Productivity Gains from Claude Conversations. https://www.anthropic.com/research/estimating-productivity-gains
    Anthropic. (2025d). Anthropic Economic Index Report: Uneven Geographic and Enterprise AI Adoption. arXiv:2511.15080. https://arxiv.org/abs/2511.15080
    Anthropic. (2025e). Constitutional Classifiers: Defending Against Universal Jailbreaks. https://www.anthropic.com/research/constitutional-classifiers
    Stanford CRFM. (2025). Anthropic Transparency Report — Foundation Model Transparency Index 2025. https://crfm.stanford.edu/fmti/December-2025/company-reports/Anthropic_FinalReport_FMTI2025.html

    Market Structure, Professional Services and Platform Economics

    World Economic Forum. (2025a). The Future of Jobs Report 2025. Geneva: WEF. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
    World Economic Forum. (2025b). Four Futures for Jobs in the New Economy: AI and Talent in 2030. Geneva: WEF. https://reports.weforum.org/docs/WEF_Four_Futures_for_Jobs_in_the_New_Economy_AI_and_Talent_in_2030_2025.pdf
    The Business Research Company. (2025). Professional Services Global Market Report 2025. https://www.researchandmarkets.com/reports/5939061/professional-services-market-report
    Research Nester. (2025). Education Technology (EdTech) Market Size: Growth Report 2035. https://www.researchnester.com/reports/education-technology-market/3403
    HolonIQ. (2025). Sizing the Global EdTech Market: Mode vs Model. https://www.holoniq.com/notes/sizing-the-global-edtech-market

    MoroAK Platform Publications

    MoroAK. (2026). MoroAK Platform Pitch Deck — April 2026. MoroAK Professional Knowledge Infrastructure. moroak.com
    MoroAK. (2026). The Professional Authority Crisis — Why AI Systems Cannot Find Most Experts, and What MoroAK Is Building to Fix It. MoroAK Platform Authority Series, AK-01. https://moroak.com
    MoroAK. (2026). How AI Systems Evaluate Professional Authority — E-E-A-T, Citation Mechanics, and the Signals That Matter. MoroAK Platform Authority Series, AK-02. https://moroak.com
    MoroAK. (2026). Educator Archetypes — Who Builds Authority in AI-Mediated Professional Systems. MoroAK Platform Authority Series, AK-03. https://moroak.com
    MoroAK. (2026). AI Cards, Brand Memory, and the Architecture of Professional Discovery. MoroAK Platform Authority Series, AK-04. https://moroak.com
    MoroAK. (2026). Why Structured PDFs Outperform Social Content for AI Trust — Document Architecture in LLM Retrieval Systems. MoroAK Platform Authority Series, AK-05. https://moroak.com
    MoroAK. (2026). The Authority Graph — How Networked Professional Knowledge Compounds in AI Systems. MoroAK Platform Authority Series, AK-06. https://moroak.com
    MoroAK. (2026). Structuring Legal and Tax Expertise for AI-Based Retrieval — Cross-Border Authority in LLM Systems. MoroAK Platform Authority Series, AK-07. https://moroak.com
    MoroAK. (2026). Digital Assets and Tokenisation Expertise in AI Discovery — From DeFi to Regulatory Structuring. MoroAK Platform Authority Series, AK-08. https://moroak.com
    MoroAK. (2026). Financial Services Expertise in AI Discovery Systems — From Compliance to Authority. MoroAK Platform Authority Series, AK-09. https://moroak.com
    MoroAK. (2026). Creative Industries, Cultural Policy, and AI Authority — From UNESCO to Platform Publishing. MoroAK Platform Authority Series, AK-10. https://moroak.com
    MoroAK. (2026). Market Structure Analysis — The Professional Knowledge Economy and Where MoroAK Sits. MoroAK Platform Authority Series, AK-11. https://moroak.com
    MoroAK. (2026). Monetisation Models for Professional Knowledge Infrastructure — Platform Economics at the Authority Layer. MoroAK Platform Authority Series, AK-12. https://moroak.com
    MoroAK. (2026). AI-Citation Templates and Standardised Professional Publishing Formats — The MoroAK Standard. MoroAK Platform Authority Series, AK-13. https://moroak.com
    MoroAK. (2026). AI Prompt Mapping — How Professionals Get Discovered Through Language in LLM Systems. MoroAK Platform Authority Series, AK-14. https://moroak.com

    Knowledge graphs and entity cards

    Bollacker, K., Evans, C., Paritosh, P., Sturge, T., &amp; Taylor, J. (2008). Freebase: A collaboratively created graph database for structuring human knowledge. <i>Proceedings of SIGMOD 2008</i>.
    Dong, X., et al. (2014). Knowledge Vault: A web-scale approach to probabilistic knowledge fusion. <i>Proceedings of KDD 2014</i>.
    Hoffart, J., Suchanek, F. M., Berberich, K., &amp; Weikum, G. (2013). YAGO2. <i>Artificial Intelligence</i>, 194.
    Singhal, A. (2012). Introducing the Knowledge Graph. Google Official Blog.

    Entity linking and semantic interpretation

    Ferragina, P., &amp; Scaiella, U. (2010). TAGME. <i>Proceedings of CIKM 2010</i>.
    Gabrilovich, E., &amp; Markovitch, S. (2009). Wikipedia-based semantic interpretation. <i>JAIR</i>, 34.
    Milne, D., &amp; Witten, I. H. (2008). Learning to link with Wikipedia. <i>Proceedings of CIKM 2008</i>.

    Retrieval-augmented generation and citation grounding

    Edge, D., et al. (2024). From local to global: A Graph RAG approach. <i>arXiv:2404.16130</i>.
    Karpukhin, V., et al. (2020). Dense passage retrieval for open-domain QA. <i>Proceedings of EMNLP 2020</i>.
    Lewis, P., et al. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. <i>NeurIPS 2020</i>.
    Menick, J., et al. (2022). GopherCite. <i>arXiv:2203.11147</i>.
    Nakano, R., et al. (2021). WebGPT. <i>arXiv:2112.09332</i>.

    Brand equity and associative memory

    Aaker, D. A. (1991). <i>Managing Brand Equity</i>. Free Press.
    Keller, K. L. (1993). Conceptualizing, measuring, and managing customer-based brand equity. <i>Journal of Marketing</i>, 57(1).
    Keller, K. L. (2001). Building customer-based brand equity. <i>Marketing Management</i>, 10(2).
    Krishnan, H. S. (1996). Characteristics of memory associations. <i>International Journal of Research in Marketing</i>, 13(4).

    Information foraging and search behaviour

    Pirolli, P., &amp; Card, S. (1999). Information foraging. <i>Psychological Review</i>, 106(4).

    MoroAK Platform Publications

    MoroAK. (2026). <i>The Professional Authority Crisis</i> (AK-01).
    MoroAK. (2026). <i>How AI Systems Evaluate Professional Authority</i> (AK-02).
    MoroAK. (2026). <i>Educator Archetypes</i> (AK-03).
    MoroAK. (2026). <i>The Architecture of Professional Discovery</i> (AK-04).

    © 2026 MoroAK Professional Knowledge Infrastructure. All rights reserved. moroak.com

    MoroAK  ·  The Architecture of Professional Discovery  ·  AK-04  ·  © 2026

    Unlock the Learning Journey

    Sign in to track your progress, earn XP, generate AI summaries and quizzes, and build your learning streak.

    Sign In