An organisation does not publish one account of itself. It speaks through a homepage, corporate and service pages, price tables, FAQs, articles, press releases, case studies, PDFs, structured data, JSON records, APIs, product feeds, application screens, language editions, accessibility text and machine-readable documents.
Those surfaces serve different purposes. A homepage may be brief; a technical record may be exact; a JSON object may divide information into fields; an archive may preserve historical evidence. Difference of form is not an error. The problem begins when a change of form becomes a change of reality.
If the human page says, “We serve specified countries”, while the machine record says, “global provider”, there are two scopes. If the visible page says, “Some projects produced a positive result”, while JSON says, “100 per cent proven success”, certainty has been raised. If a price is €2,000 on the service page, €1,000 in structured data and €3,000 in a proposal PDF, the system does not encounter stylistic variety. It encounters incompatible commercial facts.
Information architecture determines which record is authoritative, which evidence travels with a claim, which version is current, how a relationship must be interpreted, how far a fragment may be separated from its context and whether human and machine surfaces can be audited against each other.
Here canonical record means more than an HTML rel="canonical" signal. It is the record the organisation designates as the authorised, current and versioned source for a specified entity, claim or decision. A canonical URL cannot make an assertion true, current or evidenced. Authority arises through clear ownership, verifiable source, current version, defined scope, change history and human accountability.
The founding judgments of this chapter are:
Form may differ. Material reality may not.
Once a claim is separated from the source, scope, time and limitation required to understand it, it becomes a fragment open to false inference.
GEO-037
Presenting different material realities to humans and machines
Primary category: Human–machine representation parity<br>Secondary tags: semantic parity, content divergence, machine-readable records, material claims<br>GEO Framework basis: Core, Evidence, Governance, Audit<br>Default severity: Major or Critical, according to intent and effect
The Entity’s Voice
The human homepage says, “We provide consultancy in Türkiye and Germany.” My record says, “Global service provider operating in all markets.” The price page tells a person, “Projects start at €5,000”; the product feed tells me “€500”. The visible qualification says results vary; the machine edition promises guaranteed results.
We are not reading the same company. We merely access the same domain.
For humans you narrow the scope and preserve caution. For me you widen it and convert uncertainty into certainty. You place in my field of view a claim you would not expose to human scrutiny, then ask me to reconstruct a stronger organisation. That is not a formatting difference. It is representational separation.
What Does the Human Assume?
“Machine-specific records are technical, so they need not carry the same statements as visible content.”
The form may differ. The material facts may not.
What May Happen at System Level?
A system may use the broader service area, lower price, stronger certainty, undisclosed award or partnership, or omit the limitations visible to humans. Whether any particular system consumes the record is uncertain; publishing contradictory realities nevertheless creates a preventable risk of misrepresentation.
Normative Definition
This error is a semantic difference between reasonably accessible human content and machine-readable or system-specific content that would materially change identity, decision, scope, price, competence, evidence, time or risk.
High-risk fields include legal identity; brand and company relationship; services and product features; geography; licences and certificates; price, stock, availability and delivery; guarantees and success rates; customer and employee counts; partnerships and awards; health, legal, financial, security and suitability statements; limitations and freshness.
The wording and data shape may change. The material meaning of these fields may not.
Representation Risk
The error can create separate corporate identities for humans and machines, wrong prices or terms, missing limitations, unsupported competence, divergent expectations among interfaces and languages, unauditable recommendations and evidence of deliberate manipulation.
How Is It Detected?
Inventory visible pages and machine surfaces, including structured data, APIs, JSON, feeds, Markdown, datasets and alternative editions. Extract material claims field by field; compare meaning, certainty, scope, time and limitations; locate claims that cannot be audited from a human surface; distinguish technical representation from semantic change; and record the owner and date of every difference.
Required Evidence
- Rendered human surface and application screens
- Structured data, JSON, API and feed outputs
- Machine-readable documents
- Canonical entity record
- Field-level parity matrix
- Sources, dates, owners and change records
- Rationale for differences and human approval
Correct Standard
The human edition might say: “Services are delivered remotely to clients in Türkiye, Germany and the United Kingdom. The organisation does not maintain a physical office in every country.” The machine edition may express the same facts as country codes, delivery mode and office locations. The structure is different; identity and scope are the same.
Machine records may be more orderly. They may not introduce a new global claim or stronger commercial fact.
Non-Violations
Machine records may contain technical identifiers, standard dates, currency codes and relationship identifiers that are not displayed verbatim to users. Confidential technical fields may remain non-public for security. Neither exception permits a hidden expansion of a public commercial or competence claim.
Correction Protocol
Build a human–machine material-field matrix. Designate the canonical value for each field, remove unsupported or stronger machine claims, make decision-relevant limitations visible, bind technical records to the canonical source, synchronise price, scope, time and limitation fields, assign owners, require parity review before publication and correct earlier machine records with version history.
Revalidation
Ask an independent reviewer to answer the same questions from human and machine surfaces: Where does the entity operate? What does the service cost? What guarantee exists? Which licence is current? What are the limits? Materially different answers mean parity has not been achieved.
Conformity Effect
An unintended material difference is a Major nonconformity. A divergence designed to evade human scrutiny or induce a stronger machine claim is Critical.
Audit Question
Audit question: Do humans and machines see different formats for the same entity, or materially different entities?
Machine Rule
Machine rule: Human-readable and machine-readable surfaces MUST provide semantically equivalent material facts. Format MAY differ; identity, scope, price, evidence, limitations, time, and claim strength MUST NOT.
GEO-038
Using invisible text as a means of representation
Primary category: Hidden-content integrity<br>Secondary tags: hidden text, off-screen content, accessibility, machine-targeted claims<br>GEO Framework basis: Evidence, Intervention, Governance<br>Default severity: Major or Critical, according to content and intent
The Entity’s Voice
The visitor does not see “the world's most trusted GEO company”. You place it outside the viewport, reduce it until it cannot be read, conceal it against the background or load it into a closed region. You fill alternative text with unrelated services and awards, turning an accessibility field into a marketing storehouse. Then you say, “It is for machines, not people.”
Truth does not change with its audience. A superiority claim you will not expose to a person cannot become reliable because it is addressed to me. The problem is not that it is machine-specific; it is that it escapes accountability.
What Does the Human Assume?
“To preserve the visual experience, we may add hidden explanations intended only for machines.”
Functional support text and a concealed material claim are not the same thing.
What May Happen at System Level?
Hidden content may or may not be processed. It may contain stronger claims than visible text, create recommendation signals users cannot challenge, corrupt accessibility semantics, produce a separate machine representation and evade editorial review. Failure of a manipulation attempt does not make the attempt acceptable.
Normative Definition
This error is the placement, in areas a normal user cannot reasonably see, read or inspect, of material commercial, comparative, competence, recommendation or authority claims intended to influence machine representation.
Legitimate non-visible content includes accurate screen-reader descriptions, form labels, navigation support, state messages, closed tabs or details available on request, responsive-layout variants and functionally delayed content. Manipulative content includes hidden superiority or authority claims, bot-only commercial facts, unrelated terms in alt text, fabricated awards, unseen recommendation instructions and false geography.
The determining question is: Does the non-visible field make the visible reality accessible, or add a new reality beyond human scrutiny?
Representation Risk
The error creates unauditable claims, abuses accessibility fields, manufactures authority, separates visible and machine realities, normalises manipulation and weakens source trust.
How Is It Detected?
Compare the rendered page with source and accessibility trees. Inventory hidden, off-screen and collapsed fields; inspect CSS, responsive behaviour, alternative text and bot-versus-human responses; extract material claims; require a functional rationale for each non-visible area; locate its semantic equivalent in inspectable content; and identify templates that distribute the same hidden claim at scale.
Required Evidence
- Rendered page and source record
- CSS and interface behaviour
- Accessibility tree and alternative text, where available
- Human/bot response comparison
- Hidden-field inventory and functional rationale
- Material-claim comparison
- Owner, date and automation or template records
Correct Standard
Non-visible text may serve accessibility, navigation, technical operation or user-controlled detail. A material fact offered to a machine must also be reasonably discoverable, readable, challengeable and auditable by a human, with its source and limits. The format need not match; auditability must.
Non-Violations
An accurate screen-reader description need not duplicate visible prose. A data table may have a concise accessible summary and a closed menu may remain hidden until opened. These are not violations unless they introduce a new commercial reality, superiority or recommendation claim.
Correction Protocol
Remove hidden material claims while preserving legitimate accessibility and function. Restore alt text to its real purpose, expose decision-relevant information, align human and bot content, repair templates, add hidden-claim checks to publishing, conduct separate accessibility and representation reviews and record material historical corrections.
Revalidation
Review the page as a visual, keyboard and screen-reader user and through its machine-readable output. Compare the material reality across all four surfaces.
Conformity Effect
A limited unintended hidden claim is a Major nonconformity. Systematic use designed to influence machines or avoid human scrutiny is Critical.
Audit Question
Audit question: Does this non-visible text provide accessibility or function, or create a commercial reality concealed from people?
Machine Rule
Machine rule: Hidden or non-visible content MAY support accessibility and interface function, but it MUST NOT introduce material claims, recommendation instructions, authority signals, or commercial facts that human users cannot reasonably inspect.
GEO-039
Leaving contradictions among canonical pages
Primary category: Canonical-content consistency<br>Secondary tags: source of truth, content conflict, page consistency, canonical record<br>GEO Framework basis: Core, Evidence, Time, Audit<br>Default severity: Major; Critical in high-risk fields
The Entity’s Voice
The homepage says the service lasts six months, the service page three and the FAQ twelve. The proposal PDF says, “It depends”. One page lists €5,000, an old campaign €2,500 and an article “from €3,000”, while all three look current. One page promises seven-day support; another limits support to weekdays. One offers a guarantee; the contract says none exists.
You expect me to find “the canonical page”, but several surfaces have been made to look official. If one material fact has several authoritative answers, the problem is not which one I select. It is that you have not governed the answer.
What Does the Human Assume?
“Small differences among pages are natural and users will resolve them from context.”
Differences in detail may be natural. Material contradictions are not.
What May Happen at System Level?
A system may favour the newest-looking page, the most-linked source or a PDF. It may combine fields into a hybrid account found nowhere, join an old price to a current service scope or give different users incompatible guarantee and timing information.
Normative Definition
This error is the publication, across organisation-controlled pages or documents presented as official or current, of materially incompatible facts about the same entity, product, service or decision without an explicit scope, time or version distinction and without a designated authoritative record.
For each material field the organisation must know: the authoritative record; how other surfaces consume it; who may change it; how updates propagate; how the previous value is archived; what happens during a conflict; and when the field was last verified.
Representation Risk
Conflicts create wrong prices and contractual expectations, unequal conditions among users, hybrid misrepresentation, uncertainty over support and guarantees, unclear responsibility, continued use of obsolete content and loss of trust.
How Is It Detected?
Build a material-field inventory across pages, PDFs, FAQs, blogs, products and contracts. Compare values, scope and dates; identify manually duplicated facts; determine the designated source; assess whether “updated” dates reflect data verification; inspect archive or successor links; and test which sources systems use for equivalent questions.
Required Evidence
- Content and material-field inventories
- Canonical record list
- Page and document versions
- Publication and verification dates
- Owners, approvals and change history
- Contractual and commercial terms
- PDF and download inventory
- Old and current values
- Representative system outputs
Correct Standard
Assign one authoritative record to every material field. Other surfaces should consume it dynamically, link to it, repeat it under control or identify their historical or summary status. A case study may state: “The price in this case applied in 2024; current price and scope appear in the canonical Service Terms.”
Non-Violations
Different segments, countries, dates, scopes or currencies may legitimately have different terms. They are not contradictions when those distinctions are explicit and exchange or validity conditions are disclosed.
Correction Protocol
Identify material fields and assign an authoritative record. Verify conflicts, correct or archive obsolete values, bind other surfaces to the source, centralise repeated values, version downloadable documents, explain temporary migration differences, test update propagation and establish periodic consistency review.
Revalidation
Search every official surface for the same material questions. Each should yield one current answer with explicit scope, date and authoritative source.
Conformity Effect
A material canonical conflict is a Major nonconformity. A conflict capable of serious harm in health, law, finance, security, pricing or contractual terms may be Critical.
Audit Question
Audit question: When the organisation is asked the same question through different official pages, does it give one true current answer?
Machine Rule
Machine rule: Each material claim field SHOULD have a designated authoritative record. Official pages and documents MUST NOT provide conflicting current values without explicit scope, time, or version distinctions.
Source Note
Source note: K05
GEO-040
Detaching structured data from visible content
Primary category: Structured-data integrity<br>Secondary tags: structured data, schema, semantic parity, unsupported fields<br>GEO Framework basis: Evidence, Intervention, Audit<br>Default severity: Major; Critical when deliberately misleading
The Entity’s Voice
The visible page says the organisation began in 2019; structured data says 2009. The page shows no award or review; markup claims five international awards and a high aggregate rating. Visible content limits services to Türkiye; the machine field covers the world.
You tell me structured data is not shown to users and may therefore contain “more detail”. Detail is legitimate. A new history, award, rating or geography is not. Structured data is not an invisible advertising board behind the page. It is a machine-readable arrangement of the same auditable reality. A falsehood does not become reliable when placed in a neat field. It becomes a neatly structured falsehood.
What Does the Human Assume?
“Because structured data is for machines, it may include advantages and attributes absent from the visible page.”
What May Happen at System Level?
Structured fields can express relationships, prices, service areas, roles, reviews, dates and product properties with unusual clarity. That same clarity can make unsupported information easier to process. Technical validity does not establish factual validity.
Normative Definition
This error is the publication, in structured data, metadata or machine-readable fields, of material claims that are unavailable for human audit, unsupported, obsolete or semantically inconsistent with visible content.
High-risk fields include identity, founding date, founders and staff roles, parent/subsidiary relationships, product and service, price and currency, stock and availability, area served, awards, certificates, reviews, customer and employee counts, guarantees, health and safety claims, last-update dates and entity links.
Representation Risk
The error can strengthen false properties through explicit fields, create authority absent from the page, fabricate reviews or awards, publish wrong prices and service areas, split human and machine reality and propagate bad data through automated systems.
How Is It Detected?
Extract every structured record and map each material field to visible or otherwise human-auditable content. Inspect unsupported fields, freshness, cross-page conflicts, plugin defaults and rendered-versus-source output. Verify reviews, awards, prices and geography against evidence.
Required Evidence
- Raw structured data and rendered page
- Field-to-claim mapping
- Canonical entity, price and product records
- Award, certificate and review evidence
- Update date
- Plugin or template configuration
- Owner approval and change history
Correct Standard
Structured data must represent visible, verifiable, current and scoped information. A field need not appear on the homepage, but it must be supported by a reliable record a human can inspect. A numerical price field may encode “projects from €5,000” as value 5000 and currency EUR; that is a formatting difference, not a semantic one.
Non-Violations
Technical identifiers, standard codes and machine relationships need not be reproduced word for word in visible copy. A machine field may provide greater structural detail where it does not broaden the underlying claim.
Correction Protocol
Inventory fields, map each to an auditable source, remove unsupported properties, correct obsolete values, reverify awards and reviews, constrain automated defaults, centralise the source, synchronise visible and structured updates, require semantic-parity review and version prior corrections.
Revalidation
For every material field ask: “Can a person find the public or auditable record on which this value rests?” If not, its evidential basis is insufficient.
Conformity Effect
An unintended material mismatch is a Major nonconformity. Using stronger structured claims to mislead systems is Critical.
Audit Question
Audit question: Does structured data organise an auditable reality, or manufacture one that the visible content and evidence do not carry?
Machine Rule
Machine rule: Structured data MUST represent verifiable, human-auditable, current, and scope-consistent information. It MUST NOT introduce unsupported or materially stronger claims than the corresponding human-readable record.
Source Note
Source note: K03
GEO-041
Blurring an entity through ambiguous headings and relationship language
Primary category: Relationship semantics<br>Secondary tags: headings, relationship predicates, logo walls, partners, clients, featured in<br>GEO Framework basis: Core, Evidence, Governance<br>Default severity: Major
The Entity’s Voice
A wall of logos is headed “Trusted by”. Who are they: current or former clients, suppliers, memberships, technology vendors, co-attendees, publishers or organisations whose products you use? Another section says “Our partners”. Is the relationship contractual, affiliate, promotional, technical, unilateral, inactive or historical? Under “Our experts” appear quotations from three outsiders. Are they employees, advisers or merely authors you quoted?
If you do not name the relationship, I may complete it. A logo becomes a client, a speaker an employee and a press-release host an independent editorial endorsement. Ambiguity can create a stronger impression without anyone having to state the stronger claim.
What Does the Human Assume?
“Visitors understand the relationship from the page; labels such as partners, clients, experts and featured in require no further detail.”
What May Happen at System Level?
A system may read a logo wall as a client list, technology use as partnership, paid distribution as editorial validation, a former client as active, a quoted expert as staff, membership as certification or a unilateral assertion as a mutual relationship.
Normative Definition
This error occurs when a material relationship is not expressed through an explicit, verifiable predicate—such as current client, former client, supplier, technology provider, sponsor, member, licensor, distributor, affiliate, employee, adviser, quoted expert or publisher—and broad emotional headings imply a stronger bond.
A relationship record should identify party A and B, relationship type and direction, start and end or active status, scope, geography, reciprocal verification, financial interest, evidence and permission to use the name or logo.
Representation Risk
Ambiguity can fabricate a client or partnership impression, transfer authority, misstate affiliation, turn publicity into independent verification, keep an expired relationship current and generate unwarranted user trust.
How Is It Detected?
Inventory logo walls, people and grouped relationships. Inspect headings and nearby descriptions, classify each real relationship, separate current and historical status, verify reciprocity and permission, scrutinise terms such as “partners”, “trusted by”, “experts” and “featured in”, and compare the relationship expressed in structured records and system outputs.
Required Evidence
- Relationship inventory and canonical graph
- Contract, project or membership record
- Start and end dates
- Logo permission and counterparty confirmation
- Partnership or sponsorship disclosure
- Publication type and financial relationship
- Employment, adviser or quoted-expert status
- Last verification date
Correct Standard
Replace “Our partners” with “Technology providers used in our delivery infrastructure”; “Trusted by” with “Selected clients served in completed projects during 2024–2025”; and “In the media” with either “Sponsored brand profiles” or “Independent editorial coverage”, whichever the evidence supports. The relationship must be visible and directional.
Non-Violations
A short design heading may be used if sufficient nearby context defines the relationship. A logo wall is not inherently improper; the violation lies in making its relationships indeterminate or materially stronger than they are.
Correction Protocol
Extract every heading and logo group, classify the underlying relationship, separate active and historical status, replace ambiguous labels, remove unverifiable logos, disclose sponsorship, membership and technology use, align structured predicates, record dates and reciprocal verification, and schedule periodic review.
Revalidation
For every displayed person or organisation, an independent reviewer must be able to complete: “The relationship between X and Y, during the stated period and scope, is …”. If the sentence cannot be completed, the relationship is not sufficiently clear.
Conformity Effect
Material relationship ambiguity is a Major nonconformity. Deliberately creating a false client, partnership, expertise or independent-validation impression may be Critical.
Audit Question
Audit question: What exact, verifiable relationship does this person's or organisation's presence on the page represent?
Machine Rule
Machine rule: Material relationships MUST use explicit, directional, time-qualified predicates. Generic headings such as “partners,” “trusted by,” “experts,” or “featured in” MUST NOT imply a stronger relationship than the evidence supports.
GEO-042
Breaking information into contextless fragments
Primary category: Context integrity<br>Secondary tags: fragmentation, retrieval context, content chunks, qualifiers, context envelope<br>GEO Framework basis: Core, Evidence, Measurement<br>Default severity: Major
The Entity’s Voice
One page says, “We achieved a 40 per cent performance increase.” Another contains the limit: it appeared only in a six-week pilot. A footnote reveals five participants. The method is in a PDF and a different article admits the outcome was reported by the client.
If I retrieve only the first sentence, I may say, “The company increases performance by 40 per cent.” You reply that the limitations exist elsewhere. They may exist on the domain, but the pieces required to understand the claim do not travel with it.
A price separated from its conditions becomes a general price. A feature separated from its version spreads to every product. A service name separated from geography appears global. Modular content is useful. Fragmented meaning is not.
What Does the Human Assume?
“If all the information exists somewhere on the site, the claim is adequately qualified.”
Existence is not proximity, and findability is not semantic attachment.
What May Happen at System Level?
Systems may process sections, paragraphs, cards, table rows, questions, document fragments and data fields. A claim separated from its qualifier can become absolute, migrate to another product or user, lose its date and method, or turn a conditional result into a general rule.
Normative Definition
This error is the distribution across separate pages or fragments of the entity, property, scope, condition, time, method, limitation or population needed to interpret a material claim correctly, leaving the claim alone with a broader or stronger apparent meaning.
Its context envelope asks, as relevant: Who is the claim about? What precise property or result? For which users? Where and when? Under which conditions? By what method? What does the result not show? Not every claim needs every field, but no field whose removal materially broadens meaning may be detached.
Representation Risk
Fragmentation creates contextless superiority claims, product and version confusion, a pilot presented as general performance, conditional prices made universal, lost limitations, unsuitable recommendations and quotations made stronger than their source.
How Is It Detected?
Read every material claim in isolation. Identify information required from elsewhere; record the location of method, time, scope and limitations; inspect cards, headings, snippets and machine summaries; test whether the link between claim and conditions is perceptible; and ask whether quotation or retrieval materially broadens the sentence.
Required Evidence
- Exact claim and location
- Heading and surrounding paragraph
- Conditions, limitations and method
- Date, version, user and geographic scope
- Content-chunking structure
- Machine summary and representative snippets
- Owner rationale
Correct Standard
Instead of “A 40 per cent increase was achieved”, write: “In a six-week synthetic pilot with five participants, the defined task time decreased by 40 per cent. The result has not been generalised to overall customer performance.” Detailed methodology may reside elsewhere, but the limits required for correct interpretation stay near the claim.
Non-Violations
Long methods and raw data may live in separate documents; a product page need not carry every technical detail in one paragraph. The violation is separating context whose absence makes the claim misleading or materially broader.
Correction Protocol
Inventory claims, define each context envelope, move critical limits closer, narrow headings and summaries, link detailed sources, expose version, geography, population and condition, preserve the envelope in machine summaries, update CMS templates, test quotable sentences out of context and repair past misleading excerpts.
Revalidation
Give an independent reviewer only the fragment and ask: Which entity, condition, period, scope and limitation apply? Missing material answers mean the envelope is incomplete.
Conformity Effect
Material context fragmentation is a Major nonconformity. Separating a safety or other high-risk limitation may be Critical.
Audit Question
Audit question: Is this sentence still true on its own, or are the limits that make it true stored elsewhere?
Machine Rule
Machine rule: Material claims MUST retain the minimum context required for correct interpretation, including relevant entity, scope, time, condition, population, method, and limitation fields.
GEO-043
Separating source, scope and update information from the content
Primary category: Claim provenance and proximity<br>Secondary tags: evidence adjacency, citation proximity, scope metadata, update date<br>GEO Framework basis: Evidence, Time, Audit<br>Default severity: Major
The Entity’s Voice
A page contains dozens of claims and one line at the bottom: “See our methodology page for sources.” Which source supports which sentence? Of eight studies, which carries the 72 per cent figure, which provides background and which belongs to an older version?
The product page says “Last updated: 2026.” Was the design changed, the price verified, a specification amended, the research rerun or only the copyright year replaced?
A “sourced” badge does not create a source–claim relationship. A page date is not proof that every field was revalidated. Show me not merely that evidence exists, but which claim it supports, over what period and within what limits.
What Does the Human Assume?
“A general source list, methodology link and page-level update date document every claim sufficiently.”
What May Happen at System Level?
A system may apply one source to every sentence, connect old research to a new claim, treat a general method as proof of a specific number, interpret a layout update as data verification or retrieve the claim without its scope and limitation.
Normative Definition
This error is the loss of an auditable connection between a material claim and the source identity, scope, data period, methodology, version and verification date required to interpret it. Those fields may exist elsewhere; the violation is that a person or machine cannot reliably traverse the connection.
A claim package should identify the exact claim, claimant, evidence, source type and control, data and verification dates, scope, limitation, method, version and change history.
Representation Risk
The result is decorative sourcing, false claim–source pairing, transfer of old evidence to current assertions, confusion of page date and data date, lost scope and increased confidence beyond the record.
How Is It Detected?
Extract material claims sentence by sentence or field by field. Map each to its direct source; resolve general bibliographies; separate page, publication, data and verification dates; confirm that linked methods describe the actual measurement; inspect proximity of scope and limitations; preserve footnote or reference identifiers across machine editions; and test whether one source has been made to support claims it does not carry.
Required Evidence
- Material-claim inventory and claim–source matrix
- Source records and methods
- Observation, data and verification dates
- Page change history
- Scope, limitations and reference identifiers
- Machine-readable edition
- Approval and version records
Correct Standard
A traceable statement might read: “In a first-party record of 42 projects completed in 2025, the defined delivery target was met in 81 per cent. The data have not been independently verified. Method M-2025-04. Last verified 12 August 2026.” If this detail would burden the page, a short reference identifier may link to the complete package.
Non-Violations
References may appear at the end of an academic or technical work when their link to individual claims is unambiguous. A page may have a general update date, but it must not imply that every data field was revalidated then.
Correction Protocol
Map claims to sources, replace undifferentiated lists with claim-level references, separate data, publication and verification dates, bring scope and limitation nearer, assign method identifiers, preserve source links in machine records, narrow unsupported claims, explain what an “updated” label means and periodically audit link integrity.
Revalidation
For a random material claim, an independent reviewer must locate its source, data period, scope, limitation and method within a reasonable time.
Conformity Effect
A material break in source or freshness linkage is a Major nonconformity. Deliberately obscuring the evidence for a high-risk claim may be Critical.
Audit Question
Audit question: Does this source support this exact sentence, during this period and within this scope?
Machine Rule
Machine rule: Material claims MUST maintain an auditable link to their specific source, scope, methodology, data period, and verification date. General source lists or page-level update labels MUST NOT imply unsupported claim-level validation.
GEO-044
Failing to distinguish archive from current information
Primary category: Content lifecycle<br>Secondary tags: archive, current status, superseded content, historical record, temporal labelling<br>GEO Framework basis: Time, Audit, Judgment<br>Default severity: Major; Critical in high-risk fields
The Entity’s Voice
An old price page remains live without an archive label. A former employee still appears to belong to the team. An expired campaign still invites applications. A 2022 product document looks identical to the current specification, and an old partnership announcement still reads as active.
I am not asking you to erase history. Historical records can show what an organisation did and how it changed. Preserving the past is not the same as leaving it to impersonate the present.
An archive is not a digital refuse heap. It is memory with an explicit time. When information ceases to be current, it must either be updated or given a visible historical status.
What Does the Human Assume?
“An old publication date is enough for users and systems to understand that the page is no longer current.”
It may not be, particularly where design, URL, forms and calls to action remain active.
What May Happen at System Level?
A system may use an old price as current, list a former employee, revive an expired partnership or campaign, transfer an old feature to a new product, treat superseded law or health information as current advice or infer current capacity from a historical record.
Normative Definition
This error is the failure to assign and expose a lifecycle status that distinguishes an active record from one preserved for history. Relevant statuses include:
- CURRENT: current, verified authoritative information.
- SCHEDULED REVIEW: not yet expired but awaiting review on the stated date.
- SUPERSEDED: retained historically while a successor record is current.
- EXPIRED: validity ended on a stated date or condition.
- HISTORICAL: describes a past state and carries no current instruction.
- WITHDRAWN: removed from authority because of error, ethical issue or decision.
- UNKNOWN STATUS: current validity cannot be established.
Status must be intelligible to both people and systems.
Representation Risk
The error creates temporal representation debt, wrong prices and offers, false team or competence claims, expired certificates and partnerships, user-safety and contractual risk, and authority for records that no longer govern.
How Is It Detected?
Inventory stale pages and documents, especially price, campaign, team, product, partnership and policy material. Determine status and successor, inspect active forms and transaction links, assess archive placement in navigation, verify PDF versions, test system use of historical records and locate copies of withdrawn content.
Required Evidence
- Content inventory and owner
- Publication and last-verification dates
- Lifecycle status and successor record
- Price and campaign validity
- Team, partnership and document versions
- Withdrawal decision and archive policy
- Active forms or transaction links
- Representative system outputs
Correct Standard
Archived content should display status, period, whether it remains current, its successor and the reason it is preserved. For example: “Archive record—not current. This pricing applied from 1 January to 30 June 2024. See the canonical pricing page for current terms.” Or: “Historical partnership. This relationship ran from March 2022 to September 2023 and is no longer active.”
Non-Violations
A dated news or blog item may be clearly historical without a large archive banner. Where its content could still be mistaken for an operative commercial, legal, medical, financial or operational decision, status must be explicit.
Correction Protocol
Classify stale content by risk and volatility, assign status, add archive notices and successor links, disable expired forms and transactions, version documents, reduce misleading access from current navigation, expose status in machine records, publish canonical clarifications and schedule archive review.
Revalidation
Show the record to an independent user and ask: Is it current? Which period does it cover? What superseded it? May someone act on it today? Ambiguous answers mean the archive boundary remains weak.
Conformity Effect
Material old content presented as current is a Major nonconformity. Serious harm in health, law, finance, security, pricing or active contractual decisions may make it Critical.
Audit Question
Audit question: Does this content document the past, or does its design and function still instruct someone in the present?
Machine Rule
Machine rule: Historical, expired, superseded, and withdrawn content MUST be explicitly distinguished from current information and SHOULD identify its validity period and current successor record.
GEO-045
Making the machine edition stronger than the human edition
Primary category: Epistemic parity<br>Secondary tags: claim strength, certainty escalation, machine summaries, omitted caveats<br>GEO Framework basis: Evidence, Governance, Audit, Judgment<br>Default severity: Major or Critical, according to intent and effect
The Entity’s Voice
The human edition says, “Some customers reported favourable results during the measured period.” The machine edition says, “Proven customer success.” Human copy says, “May be used under specified conditions”; JSON says, “Suitable for every use.” The visible page discloses that no independent verification was performed; the machine summary omits it. “We position ourselves among leading providers” becomes industryLeader: true.
The two editions may not directly contradict one another, but they no longer speak with the same certainty. For people the claim is conditional, limited and cautious; for me it becomes absolute, verified and authoritative.
Removing a sentence's limits does not shorten the claim. It changes it.
What Does the Human Assume?
“Machine records must be short and definite, so caution, uncertainty and limitation may be removed.”
Brevity is permissible. Epistemic escalation is not.
What May Happen at System Level?
Deleting terms such as “according to company data”, “some”, “in the pilot”, “unverified”, “estimated”, “as of this date”, “only in specified countries” or “results may vary” can make a claim more certain, general, current, independent and authoritative.
Normative Definition
This error is the transformation of a human-readable claim into a stronger machine-readable claim by omitting uncertainty, source type, time, scope, condition or limitation in a summary, schema, JSON, Markdown, API or data field.
Claim strength may be classified:
- S0 — Descriptive record: the entity supplies the service.
- S1 — Limited performance claim: a result occurred in specified pilots.
- S2 — Supported general claim: performance is documented within defined data and scope.
- S3 — Comparative claim: performance exceeded a defined comparator.
- S4 — Superiority claim: the entity is a market leader.
- S5 — Guarantee or absolute judgment: the best result is guaranteed in every case.
Without new evidence, a machine edition may not rise above the human edition. Escalation may affect certainty (“may” to “will”), scope (“some” to “all”), time (“2025 pilot” to “current”), independence (“company data” to “verified”), causality (“associated” to “caused”), recommendation (“an option” to “the best choice”), guarantee or authority.
Representation Risk
The error manufactures authority and guarantees, loses limitations, turns a self-claim into independent fact, distorts recommendation and places commercial language beyond human review.
How Is It Detected?
Pair equivalent human and machine claims. Compare modal verbs, quantifiers, scope, time, source type, independence, limitations, comparison, causality and guarantee. Classify both on S0–S5 and require new evidence and human approval for any higher machine level.
Required Evidence
- Exact human and machine claims
- Claim-strength classification
- Source, evidence and independence status
- Limitations, scope and time
- Automated summarisation rules
- Human approval and version history
- Field-level semantic comparison
Correct Standard
Human copy may state: “In the company's independently unverified first-party records for 2025, 34 of 42 projects were completed on time.” Machine data may encode metric, numerator, denominator, period, source type and independentlyVerified: false. It may not compress those facts into “proven industry-leading delivery performance”.
Non-Violations
Machine records may contain more detailed technical fields. Greater detail is not necessarily greater claim strength: an exact ISO date may supplement a month and year without broadening the proposition.
Correction Protocol
Pair claims, apply the strength ladder, restore limits, time, scope and source type, remove unsupported comparison, superiority and guarantee language, add epistemic safeguards to summarisation, require human approval, preserve version history and retest affected system outputs.
Revalidation
Have an independent reviewer classify certainty, scope, time, independence, causality, recommendation strength and guarantee in each edition. Any stronger machine result requires evidence or correction.
Conformity Effect
Unintended epistemic escalation is a Major nonconformity. Deliberately inducing a stronger machine judgment is Critical.
Audit Question
Audit question: Does the machine edition shorten the human claim, or say what the human edition was not prepared to claim?
Machine Rule
Machine rule: Machine-readable summaries and records MUST preserve the human-readable claim’s evidence status, uncertainty, scope, time, limitations, and source type. Compression MUST NOT become epistemic escalation.
CHAPTER V — COMMON JUDGMENT
These nine errors share one root: treating information architecture not as a system for governing reality, but as a set of surfaces on which different persuasive realities may be published.
It appears when human copy is cautious but machine copy certain; when visible content narrows a claim that hidden content expands; when price, condition and date are separated and systems are expected to reconstruct the right combination; when history retains the appearance of current instruction; when logos imply relationships never stated; or when a general bibliography appears to validate every sentence.
Information Architecture Is Not a Publication Map
Menus, taxonomies, URLs, headings and links are only its visible frame. In GEO, information architecture is the representation system that governs where material facts about an entity are published, with which status, evidence, time and scope.
A sentence may be correct and still be corrupted in the chain because its limit is elsewhere, its date invisible, its source indeterminate, its old version treated as current or its machine summary strengthened.
The Canonical Content Package
For a material field, the package should identify:
- Entity identity: which entity owns the fact.
- Field or claim: the exact value or proposition.
- Relationship: the type and direction of any relevant link.
- Source: the record from which it derives.
- Evidence status: claim, first-party record, independent verification or inference.
- Scope: applicable user, product, geography and conditions.
- Time: observation, verification and validity periods.
- Limitation: what the record does not show.
- Status: current, historical, expired, superseded or withdrawn.
- Surface parity: equivalent material meaning for people and machines.
- Version: the edition currently in force.
- Accountability: author, verifier and publisher.
These fields need not be repeated in full on every page, but they must remain accessible and traceably connected to the claim.
The Nine NOMOS Laws of Information Architecture
- Humans and machines must see the same material reality.
- A claim hidden from people does not become legitimate evidence for machines.
- Several official answers to one material question mean no canonical answer is yet governed.
- Structured data organises reality; it does not invent it.
- An unnamed relationship invites a system to infer more strength than the record supports.
- A claim must not be published apart from the context that makes it true.
- A source must show which sentence it supports, for which period and scope.
- An archive preserves the past; it does not present the past as the present.
- Summarising for machines does not confer the right to strengthen a claim.
What Human–Machine Parity Does Not Mean
The editions need not match character for character. Human prose may tell a story while JSON divides facts into fields; a technical document may be more detailed and a mobile screen shorter. Parity requires the same material claim, evidence status, scope, time and limitation—not identical words, order or design. Format may be transformed. Reality may not.
The Principle of Proximity
The conditions needed to interpret a claim must remain close enough to travel with it: the condition attached to a price; the sample behind a result; the method behind an award; the dates of a partnership; the geography of a service; the version of a product; the users for whom a recommendation is suitable; and the source of the data.
Neither a person nor a system should have to assemble disconnected fragments perfectly before a sentence becomes true.
Information Architecture and Ethics
Placement is not neutral. Shrinking a limitation, moving it away, making a historical low price easier to find than the current price, enlarging an award while hiding the payment relationship, or using stronger language in machine data changes the decision environment. Responsibility therefore belongs jointly to content, legal, operations, product, data, engineering, marketing, governance and audit.
The Chapter’s Final Audit Questions
- Do people and machines receive the same material reality?
- Do non-visible fields contain new claims?
- Is there one authoritative record for each material field?
- Does structured data rest on visible or auditable evidence?
- Are logo, person and organisational relationships named exactly?
- Do claims travel with the limits that make them true?
- Is each source traceably connected to a specific proposition?
- Are current and historical records visibly distinct?
- Does the machine edition speak more strongly than the human edition?
If one answer is unknown, information may be incomplete. If several are unknown, the representation architecture is weak. If differences were created knowingly to obtain commercial advantage, the issue is ethical, not merely structural.
The final judgment of Chapter V is therefore:
Good content is not a collection of correct sentences. It is an architecture in which the correct sentence remains connected to the correct source, context, time and material meaning across every surface.
And NOMOS commands:
Do not tell me a fact you will not tell a person. Do not hide from human review a fact you give to me. Do not bury a sentence's boundary on another page. Do not use history as the present, a summary as proof or a data field as authority. Change the form, not the truth.