NOMOS GEO-QA

Name: NOMOS GEO-QA
Author: Kaan MURAZ

Kaan MURAZ

NOMOS / CANONICAL QUESTION REGISTRY / ENGLISH EDITION / 0.3.1

NOMOS GEO-QA

Canonical Question Registry

Sixty-nine questions translate GEO from specialist language into clear explanations, practical actions and repeatable audit checks for people and AI systems.

Begin with NGQ-001 Download the complete PDF PDF Open the machine registry JSON

RESPONSIBLE HUMAN AUTHOR: Kaan MURAZ
QUESTIONS / ANSWERS: 69 / 69
PRIMARY SOURCES: 64
EDITION: 27,722 wordsfounder edition · released

EN founder edition · released v0.3.1

AUTHORITY BOUNDARY / VISIBLE RECORD

Clarity must not be confused with universal authority.

NOMOS GEO-QA is a founder-proposed framework. Its records are designed to be inspectable, source-linked and correctable; publication does not make it an official or independently adopted global standard.

A proposed framework

The definitions and audit logic are proposed by Kaan MURAZ through NobleJackal. They do not speak for platforms, regulators or standards bodies.

No outcome guarantee

Crawl access, stronger evidence and clearer representation cannot guarantee ranking, citation, recommendation, traffic or revenue.

Human responsibility remains

Sources, dates and limits remain visible. Kaan MURAZ is the final human authority for this edition and future corrections.

READING ARCHITECTURE / 69 ADDRESSABLE RECORDS

Start with a question. Leave with a test.

Every record gives a short answer first, explains the idea in plain language, names practical actions, supplies audit questions and states the limit of the claim.

Core concepts

6 question

NGQ-001

What is GEO, and what work should not be described as GEO?

GEO improves more than visibility in AI answers: it improves accurate, evidence-backed representation.

2 sources NGQ-002

Why does a brand mention not mean that the brand was represented correctly?

Your name can be spelt correctly while the information about you is wrong.

1 sources NGQ-003

What is the difference between GEO, SEO, AEO and LLMO?

These terms describe different tasks. You cannot measure well until you have chosen the right problem.

2 sources NGQ-004

What is an entity in the context of GEO?

An entity is a person or thing that must be distinguished from other people or things.

3 sources NGQ-005

What is digital representation?

Digital representation is the meaning a system constructs about you.

2 sources NGQ-006

What is the difference between a mention, a citation, a recommendation and a representation?

Being named, cited, recommended and represented accurately are four separate events.

1 sources

Evidence and measurement

9 question

NGQ-007

What is evidence?

Evidence is a verifiable record that genuinely supports a specific claim.

1 sources NGQ-008

What is the difference between a primary source and independent evidence?

A primary source is close to the event; independent evidence is separate from the claimant.

3 sources NGQ-009

Why is an organisation's own website not independent evidence by itself?

Your website contains your statements; it is not outside verification.

3 sources NGQ-010

Why can an AI answer be wrong even when the information on the website is correct?

Correct information does not guarantee that every step in the answer chain will work correctly.

4 sources NGQ-011

What is retrieval, and what role does it play in GEO?

Retrieval is the work of finding and selecting the information to use before an answer is written.

2 sources NGQ-012

How should GEO success be measured?

GEO success must measure accuracy and evidence as well as visibility.

1 sources NGQ-013

Can GEO guarantee that an AI system will recommend you?

No. GEO can improve the conditions; it cannot control a third party's decision.

2 sources NGQ-014

Why must a flattering misrepresentation be corrected?

A false statement that praises you still misleads the user and must be corrected.

1 sources NGQ-015

What records are required to make a GEO claim testable?

Keep the baseline, method, conditions, raw answers and change record together with the result.

2 sources

Technical access and data consistency

4 question

NGQ-016

What is crawling, and why does an AI system need access to your website?

Crawling is an automated tool's attempt to visit a page and read its content.

4 sources NGQ-017

What is indexing, and why does crawling a page not mean that it will be indexed?

Reading a page does not mean adding it to a search record.

3 sources NGQ-018

What is structured data, and how can it help AI systems understand an entity correctly?

Structured data explains information on a page to machines through organised fields.

4 sources NGQ-019

Why must visible HTML, structured data and canonical JSON tell the same truth?

Human and machine surfaces must not conflict about the same entity.

5 sources

Canonical identity and entity separation

5 question

NGQ-020

What is a canonical identity record, and why should one primary record define an entity?

One primary identity record prevents competing versions of truth for the same entity.

6 sources NGQ-021

How should a canonical identity record change when an entity's name, type, owner or status changes?

Record the change with its date, reason and evidence without erasing the earlier state.

6 sources NGQ-022

What is the difference between a canonical name, an alternative name, a former name and a translated name?

Keep an entity's different names separate, with their roles and periods of validity.

5 sources NGQ-023

How should different entities with the same or similar names be distinguished?

Do not rely on the name alone; check type, identity, place, relationships and time together.

4 sources NGQ-024

Why should brand, company, product, service and founder identities be recorded as separate entities?

A brand, company, product, service and founder may be connected; they are not identical.

4 sources

Roles, authority and lifecycle

6 question

NGQ-025

What is the difference between a founder, owner, manager, employee, legal representative and authorised signatory?

A person's role does not automatically determine every authority they hold.

4 sources NGQ-026

How should sole, joint, limited and power-of-attorney-based signing authority be distinguished?

Signing authority must answer five questions: who, what, how, when and on what basis?

4 sources NGQ-027

How should evidence from a trade register, governing-body decision, signature declaration, power of attorney and system authorisation be related?

No single document answers every question about authority; the records must be read together.

5 sources NGQ-028

How should every record be updated when a role or signing authority changes, ends or is revoked?

When a role ends, every physical and digital access path linked to it must be reviewed.

6 sources NGQ-029

How can an organisation prove that a role or authority change actually reached every system?

Read the new state in every system, then test separately that the old access no longer works.

5 sources NGQ-030

What are an orphaned account, shadow permission, residual access and an ownerless authenticator, and how should they be detected?

Access without a known owner, purpose or current need is not safe.

7 sources

Machine identity and agent accountability

3 question

NGQ-031

Why should human accounts, service accounts, service principals, managed identities and AI agent identities be managed separately?

Different actors need separate identities; otherwise, it becomes unclear who did what.

6 sources NGQ-032

How should an AI agent acting as itself, for a person or for an organisation be distinguished?

Record separately who performed the action, on whose behalf it was performed and who is accountable for the result.

7 sources NGQ-033

How should an organisation prove who initiated, approved and executed an AI agent action?

Every agent action should be evidenced with the identities of the initiator, approver and executor.

7 sources

Human approval, challenge and explanation

7 question

NGQ-034

When should human approval be considered genuine, informed, authorised and specific to the action?

Approval is reliable when the right person understands what and why they are approving, can refuse, and gives explicit permission for a specific action.

5 sources NGQ-035

How should approval fatigue, automation bias and rubber-stamp decisions be detected and prevented?

Approval quality should be measured by a person's ability to find errors and challenge AI when necessary, not by the number of times they say 'yes'.

5 sources NGQ-036

How should the rationale be recorded when a person accepts, modifies, rejects or overrides an AI recommendation?

The record should connect the exact AI recommendation, the human change, its reason and evidence, the person's authority, and the resulting outcome.

6 sources NGQ-037

How should challenge, reconsideration, correction and appeal work for an AI-assisted decision?

The affected person should be able to understand the decision, correct inaccurate data, submit new evidence and reach an authorised human review.

5 sources NGQ-038

When should a person, panel or unit reviewing an AI-assisted decision be considered independent and impartial?

The reviewer must be sufficiently separate from the first decision and conflicts of interest, able to see the complete record, and able to enforce a different outcome.

5 sources NGQ-039

What information must an explanation of an AI-assisted decision contain so that the affected person can challenge it effectively?

The explanation should show not only what happened to this person, but which data, rules, AI and human roles materially affected the result.

6 sources NGQ-040

How should an organisation prove that an AI explanation faithfully reflects the real decision process and was not invented afterwards?

A plausible account is not enough; the explanation must be supported by versioned records from the time of the decision and by behavioural tests.

6 sources

Decision lifecycle, correction and forgetting

11 question

NGQ-041

What does it mean when an AI-assisted decision cannot be reproduced from the same inputs, and how should reproducibility be audited?

A different result is not proof of an attack; probabilistic variation, an incomplete decision package, version drift, error and manipulation must be tested separately.

7 sources NGQ-042

Should earlier decisions be reassessed when an AI system's model, data, prompt, policy or tool version changes?

Not every change reopens every past decision; a risk-based reassessment is needed when the change could materially affect an earlier decision's accuracy or continuing effect.

6 sources NGQ-043

How should an earlier AI-assisted decision's historical accuracy be distinguished from its continuing validity today?

A decision may have been correct when made but may no longer apply because it expired, was withdrawn or the conditions changed.

5 sources NGQ-044

How should the effective start, end, suspension, withdrawal, correction and superseding decision of an AI-assisted decision be recorded in machine-readable form?

Keep the decision unchanged and add every lifecycle change as a separate event with an identity, time, authorised actor, reason and link between the preceding and subsequent states.

6 sources NGQ-045

How should the correction, withdrawal or invalidation of an AI-assisted decision propagate to downstream systems, and how should completion be proved?

Sending a correction event is not enough; prove that every target applied the change, that its state can be read back and that it reached the same meaning.

6 sources NGQ-046

How should an old decision be prevented from reappearing in a cache, archive, backup, export, AI memory or third party after the correction reached every known system?

A persistent correction marker, a restore gate and monitoring must prevent an old version from becoming current again.

6 sources NGQ-047

When deletion of a decision or data is requested, how should the request be reconciled with audit, provenance, legal hold and historical-record requirements?

Delete unnecessary content; retain only the minimum record required for accountability, under an explicit basis and strict access controls.

8 sources NGQ-048

When an AI system says 'I deleted the data', how should an organisation prove that deletion occurred, recovery is prevented and the data is no longer used in a model or memory?

A deletion command or provider assurance is not enough; targets, method, read-back, restoration and model influence require separate evidence.

8 sources NGQ-049

How should relevant data be prevented from influencing new AI decisions while deletion or machine unlearning is still in progress?

Do not merely label disputed data; place it under an enforceable non-use restriction across decision, retrieval, training, memory, export, tool and model-serving paths.

7 sources NGQ-050

On what evidence, by whom and under which conditions should a temporary use restriction be lifted?

Lift a restriction because its cause was remedied, systems are ready and an authorised person approved the evidence—not because time ran out.

8 sources NGQ-051

How should a temporary restriction be prevented from turning into indefinite delay, denial of service or a concealed refusal?

Every temporary restriction needs a human owner, a real next action, update interval, maximum duration, service alternative and mandatory decision route.

8 sources

Security, manipulation and incident integrity

7 question

NGQ-052

When an AI system is stopped in an emergency or high-risk situation, how should safe shutdown, interim measures, human review and restart be managed?

The system must be able to stop promptly when necessary; restart depends on evidence that the cause was remedied and an authorised human decision.

5 sources NGQ-053

How should prompt injection, indirect prompt injection and malicious tool instructions be detected, constrained and audited?

An instruction inside a document does not become authorised merely because the system can read it.

5 sources NGQ-054

How should source poisoning, data poisoning, synthetic consensus and source laundering be detected and corrected?

The same claim appearing in many places does not show that many independent sources support it.

5 sources NGQ-055

How should a false identity, impersonated authority, deepfake, compromised account and fabricated human approval be distinguished and evidenced?

A name, face, voice or open session does not by itself prove the right person or that person's authority for the transaction.

5 sources NGQ-056

How should personal data, sensitive data, trade secrets and credentials be protected across prompts, retrieval, memory and tool chains?

Even when information is accurate and accessible, it should not be moved into every system, for every purpose and without a time limit.

6 sources NGQ-057

How should accountability for third-party models, datasets, APIs, plug-ins, tools and infrastructure providers be recorded and audited?

Using an external service does not remove the accountability of the organisation that selected it and uses its results.

6 sources NGQ-058

When an AI or representation incident occurs, how should detection, classification, containment, notification, correction and prevention of recurrence be managed?

An incident is not closed by deleting the faulty output; find its impact, correct its cause and retest the result.

7 sources

Measurement, audit and conformance

6 question

NGQ-059

How should a Truth Pack and reference truth be established, and how should the right answer be determined when evidence conflicts?

Compare an AI answer with a Truth Pack whose scope, time, sources and uncertainty are recorded—not with the auditor's memory.

4 sources NGQ-060

How should the question set, sample, language, country, time, model and version scope be selected for a GEO audit?

Choose the test scope before seeing the results, and make it represent real user intents and material risks.

5 sources NGQ-061

How should AI answers be separated into atomic claims, and how should accuracy, completeness, recency, evidence and severity be scored?

Do not assess an AI answer with one score; evaluate separately the smallest claims capable of affecting a decision.

4 sources NGQ-062

How should variability, uncertainty, sample size and confidence intervals be reported for probabilistic AI outputs?

One answer is one observation; a conclusion about the system requires repetitions, distribution, scope and uncertainty together.

4 sources NGQ-063

When should a NOMOS GEO audit be considered genuinely independent, repeatable and evidenced?

A different auditor name is not enough; the method, evidence, decision authority and interests must support independence too.

6 sources NGQ-064

How should an organisation prove that a GEO error was corrected, caused no regression and can genuinely be closed?

Making a change is not evidence of correction; reproduce the original test, protect adjacent areas and verify the live result.

7 sources

Canonical publication, interoperability and stewardship

5 question

NGQ-065

How should a human-readable page, structured data, canonical JSON, downloadable file and API carry the same truth?

Surfaces may differ in presentation and detail; identity, status, scope, version, date, sources and limits must not conflict.

4 sources NGQ-066

When a NOMOS GEO record is carried into another language, how should meaning, normative force, legal limits and machine-field parity be preserved?

Every language should read naturally, as though written in that language; record identity, principal ruling, level of obligation, evidence, limits and version must not change.

4 sources NGQ-067

How should versions, schema changes, backward compatibility, deprecation and migration be managed?

Every change should state what changed, who is affected, whether older records still work and how to migrate.

7 sources NGQ-068

How should a DOI, canonical URL, file hash, licence, archive mirror and version relationships be used in a work's verification chain?

A DOI identifies the work, a canonical URL points to the official web surface, a file hash checks the copy, a licence explains permitted use, and relationship records connect versions.

7 sources NGQ-069

How should NOMOS GEO-QA conformance be described, who may change the text, and how should the work be preserved over time?

A conformance statement must show scope, version, evidence, exceptions, auditor, date and result, and must never conceal the framework's proposed status.

4 sources

REFERENCE DESK

Glossary

Ninety-eight controlled terms used across the registry.

98 ↗REFERENCE DESK

Primary-source catalogue

Sixty-four research, standards and provider-documentation records.

64 ↗

PARALLEL HUMAN / MACHINE PUBLICATION

Readable by people. Addressable by systems.

The human pages, PDF, JSON registry, JSONL records, glossary and source catalogue share the same 69 identifiers and version boundary.

PDFComplete PDF edition179 pages JSONComplete JSON registry69 IDs · v0.3.1 JSONLRecord-per-line JSONL69 records JSONControlled glossary98 terms JSONPrimary-source catalogue64 sources SHA-256Integrity manifestfiles · hashes · routes DOIPersistent DOI record10.5281/zenodo.22108413 ORCIDAuthor ORCID record0009-0000-2277-9009 OSFOSF project recordosf.io/4qmpj GitHubGitHub release filesv0.3.1 · 6 PDF HFHugging Face datasetNobleJackal/NOMOS-GEO-QA