DEV Community: Stefan Petschinka

The AI Visibility Evidence Model: What the Research Actually Supports

Stefan Petschinka — Fri, 31 Jul 2026 03:20:37 +0000

A reference model that orders the publisher-side factors behind AI visibility by strength of evidence. A markdown mirror is maintained on GitHub.

Status: July 31, 2026. Grades move as evidence accumulates.

01 Definition

What the AI Visibility Evidence Model is.

The AI Visibility Evidence Model is a reference model that orders the publisher-side factors behind AI visibility by strength of evidence. It defines five factors, Topical Relevance, Machine Access, Entity Consistency, Extractability and Independent Corroboration, and assigns each a documented evidence grade based on peer-reviewed research, controlled preprints and official platform documentation.

The model exists because the field still lacks a concise publisher-side reference that maps the main actionable factors to explicit evidence grades and primary sources. Every factor in the model carries its grade and its sources, so every statement on this page can be checked against the primary literature listed in the source register below.

02 Purpose and Boundary

A map of the evidence, not a methodology.

The AI Visibility Evidence Model maps what the evidence shows works. The AEO Mastery Framework describes how richresults.ai implements it. The model is descriptive: it reports the state of the research. The framework is prescriptive: it defines a working method. Neither replaces the other.

The model is not a ranking system, not a score and not a promise of results. AI visibility is a distribution across repeated, non-deterministic answers, and no factor in this model guarantees a citation. What the model provides is priority: it tells you which work is supported by evidence, which work is hygiene, and which work the evidence contradicts. That is less modest than it sounds. Knowing which factor rests on which evidence tells you what to start with and what can wait. A promise of results cannot tell you that.

One boundary is deliberate: the model orders the factors a publisher can work on. The retrieval stage itself, which engine selects which sources, how it ranks them and where it places them in the model context, is system-side. Controlled work shows this stage dominates citation outcomes [2], and platform documentation describes engine-specific retrieval decisions, including when a system grounds at all [13]. All five factors work toward that stage; none of them controls it.

03 The Evidence Scale

Four grades, defined before use.

Grade A: peer-reviewed and controlled, or independently replicated.
Grade B: controlled with limited transferability to open production systems, or an official platform statement.
Grade C: correlational or triangulated across independent datasets, without causal proof.
Grade D: unsupported or contradicted by evidence.

Each factor additionally carries its mechanism type. A gate is a binary precondition, a driver influences outcomes gradually, and hygiene reduces errors without creating an advantage. Mechanism type and evidence grade are separate dimensions: a factor can be a hard gate on weak empirical evidence, or a soft driver on strong evidence.

These four grades are this model's own scale, and it is deliberately conservative. A factor rated C is not unimportant; it means no one has isolated its causal effect in production, and anyone claiming otherwise is claiming more than the data supports. Grades move as evidence accumulates, and this page carries a visible update date for that reason.

04 The Five Factors

Factor 1: Topical Relevance

Content that directly addresses the actual question is the strongest documented content-side driver of citation. In the largest controlled citation study to date, 252,000 trials across six language models, topic match to the query and position in the model context were the dominant factors, and off-topic content was practically never cited first [1]. Controlled work confirms the same pattern from the model side: when weighing conflicting evidence, models rely heavily on a page's relevance to the query while largely ignoring stylistic authority signals such as scientific-looking references or neutral tone [14]. No entity work, no markup and no authority signal compensates for content that does not answer the question being asked.

Mechanism: driver. Evidence grade: A. Peer-reviewed, controlled, convergent across models and study designs [1, 2, 14].

Factor 2: Machine Access

A source that crawlers cannot reach cannot be retrieved, and a source that cannot be retrieved cannot be used for the content of an answer. Machine Access covers crawl permissions for the relevant bots, index presence, crawlable and renderable main content, and firewall configurations that do not silently block AI crawlers. Platform documentation is explicit on both sides of this gate. OpenAI requires OAI-SearchBot access for a site's content to be used in ChatGPT search answers; excluded pages can still appear as navigational links [12]. Google requires indexed, snippet-eligible pages and states that its AI features run on the same index and ranking systems as classical search [11].

Mechanism: gate. Evidence grade: B, official platform documentation. Missing access prevents the content of a page from being used as a rule; it creates no advantage, no guarantee of retrieval and no guarantee of citation [11, 12].

Factor 3: Entity Consistency

Consistent entity signals support attributing relevant content to the right source. Structured data, stable identifiers, canonical name strings and connected external profiles can reduce ambiguity in how systems resolve who is speaking. The documented effect is error reduction and disambiguation, not a visibility boost: Google states that no special markup is required for its AI features [11], and controlled work shows that knowledge-graph grounding reduces entity disambiguation errors in benchmark settings [15]. A direct causal effect on citation or mention rates in production answer engines has not been demonstrated; the transfer is inference.

Mechanism: hygiene. Evidence grade: C. Official platform statements and benchmark evidence for disambiguation; no causal proof as a citation driver [11, 15].

Factor 4: Extractability

A model can only cite what it can extract. The position of information in the model context changes outcomes causally [3, 4], and pages with concrete numbers, definitions, comparisons and procedures show substantially higher influence on generated answers than pages without them [7]. The boundary is equally documented: these effects apply after retrieval, the influence finding is descriptive and correlational [7], question-and-answer formatting alone does not help [7], and content rewriting tricks show no reliable effect and are frequently harmful under competition [2]. One transfer remains inference: a publisher does not control which passage a retriever selects or where that chunk lands in the model context, so answer-first page structure is a plausible way to serve extraction, not a demonstrated position lever.

Mechanism: driver. Evidence grades, split: B for position effects as a model phenomenon [3, 4] and for evidence density in fixed contexts [6]; C for answer-first page structure as a publisher tactic; correlational in production [7], bounded by [2] and [9].

Factor 5: Independent Corroboration

Mentions by third parties are the classic path into a model's parametric knowledge, and the evidence for this factor is best read as three separate statements. First, controlled research shows that the frequency and spread of an entity in training data causally determine what a model knows about it without searching; the study makes no claim about the independence of those documents [5]. Second, a preprint analysis reports a strong preference of several AI search systems for earned media over brand-owned content, based on the authors' own source classification [8]. Third, the claim that the independence or authenticity of a mention itself causes higher AI visibility is not demonstrated by either source. The practical principle stands on methodological grounds, not on [5]: self-published corroboration networks replicate one voice in many costumes, provide no independent confirmation, and can violate platform guidelines; for manufactured mentions no reliable evidence of a positive effect exists.

Mechanism: driver for the parametric layer. Evidence grades, split: B for training-data frequency as a causal factor in parametric knowledge [5]; C for the observed earned-media preference [8]; unsupported for independence itself as a causal visibility factor.

05 What the Evidence Does Not Support

Graded D, for different reasons.

llms.txt as a visibility lever. Google explicitly states that it does not use llms.txt for search or for generative search features [11]. Within the source register reviewed here, no platform documents the file as a ranking or visibility signal.

Content-level rewriting tricks. The broadest controlled benchmark found most conversational optimization methods ineffective and frequently harmful to citation ranking, while classical retrieval position dominated [2].

Schema as a citation switch. Structured data clarifies and disambiguates. As a direct citation driver for AI features it is officially not required [11], and no causal evidence exists for language-model citations outside search pipelines.

AI ranking positions as a metric. Answers vary heavily between identical runs, so a position from a single run is not a reliable rank [9]. Positions become meaningful only as distributions across repeated, paraphrased measurements with uncertainty intervals; visibility is a share, not a single rank.

Manufactured mentions. No reliable evidence exists for a positive effect of manufactured or self-produced mentions. In the literature the line between optimization and manipulation does not run along effectiveness anyway, but along truthfulness, verifiable evidence, the separation of content from model instructions, and disclosure of commercial intent [9]. What this model states about corroboration rests on training-data frequency [5] and an observed earned-media preference [8]; neither of those works examines self-produced mentions.

06 Visibility Is Not Citation Fidelity

Being cited and being cited correctly are different outcomes.

An independent audit of eight AI search engines found collectively incorrect answers to more than 60 percent of source-attribution queries, with premium systems frequently confidently wrong [10]. For any organization this cuts both ways: citation counts overstate control over what is actually said, and citation frequency alone says little about whether an organization is represented correctly and attributed to the original source. A measurement program should therefore track fidelity, whether a system's statements about the entity are accurate, separately from visibility.

07 Method Note

How this model was compiled.

Every claim in this model was verified against its source: the papers, official documentation and datasets listed in the register below. The synthesis was additionally checked for completeness and counter-arguments by prompting several AI systems with the same evidence question without access to this model. Convergence across systems is an editorial plausibility check, not independent scientific validation: systems share training data, sources and failure modes, and can converge on the same popular error.

Disclosure: This reference model is published by richresults.ai and authored by Stefan Petschinka, who also developed the linked AEO Mastery Framework. The source assessment, synthesis and evidence grades are the author's own work and have not undergone external peer review.

Three limits apply to everything on this page. Production systems are non-deterministic, so no single observation proves an effect. Models and retrieval methods change without notice, so findings carry dates. And no one, including the best published research, can causally isolate a single intervention inside a live answer engine. Where this page says the evidence ends, it ends.

08 Source Register

Numbered as cited above.

Vishwakarma, Kumar, Jamidar (2026). What Gets Cited: Competitive GEO in AI Answer Engines. SIGIR 2026. DOI:10.1145/3805712.3808445, arXiv:2605.25517.
Puerto et al. (2025). C-SEO Bench: Does Conversational SEO Work? NeurIPS 2025 Datasets and Benchmarks. arXiv:2506.11097.
Liu et al. (2024). Lost in the Middle: How Language Models Use Long Contexts. TACL 2024. arXiv:2307.03172.
Hsieh et al. (2024). Found in the Middle: Calibrating Positional Attention Bias. ACL 2024 Findings. arXiv:2406.16008.
Kandpal et al. (2023). Large Language Models Struggle to Learn Long-Tail Knowledge. ICML 2023. arXiv:2211.08411.
Aggarwal et al. (2024). GEO: Generative Engine Optimization. KDD 2024. arXiv:2311.09735. Effects conditional on fixed-context settings; see [2] and [9].
Zhang, He, Yao (2026). From Citation Selection to Citation Absorption. Preprint. arXiv:2604.25707.
Chen, Wang, Chen, Koudas (2025). Generative Engine Optimization: How to Dominate AI Search. Preprint, University of Toronto. arXiv:2509.08919.
Martinez (2026). Optimizing Visibility in Generative Engines: A Critical Survey. Preprint. arXiv:2607.14035.
Jaźwińska, Chandrasekar (2025). AI Search Has a Citation Problem. Tow Center, Columbia Journalism Review.
Google Search Central: AI features and your website; Optimizing your website for generative AI features on Google Search.
OpenAI: OAI-SearchBot documentation; Publishers and Developers FAQ.
Google AI for Developers: Grounding with Google Search.
Wan, Wallace, Klein (2024). What Evidence Do Language Models Find Convincing? ACL 2024. arXiv:2402.11782.
Pons, Bilalli, Queralt (2024). Knowledge Graphs for Enhancing Large Language Models in Entity Disambiguation. ISWC 2024. arXiv:2505.02737.

Related Resources

AEO Mastery Framework — the prescriptive counterpart: how richresults.ai implements what the evidence supports. → github.com/stefanpetschinka/aeo-mastery-framework

AI Citation Readiness Framework — a methodology for measuring whether an organization, expert or brand is understandable, verifiable and citable by AI systems. → github.com/stefanpetschinka/ai-citation-readiness-framework

Machine First: Why AEO Is Not SEO 2.0 — the feature article on the architecture the evidence points to. → Read it on DEV

The AI Citation Readiness Framework: Measure Before You Build

Stefan Petschinka — Tue, 02 Jun 2026 04:06:52 +0000

The AI Citation Readiness Framework is a methodology for measuring whether an organization, brand or expert is understandable, verifiable and citable by AI systems. It evaluates an entity across three dimensions and produces a single score from 0 to 100.

The problem: invisible to the machines that answer

Most organizations are invisible to AI systems. The cause is rarely a lack of expertise. The cause is entity signals that are incomplete, inconsistent or unverifiable. AI systems do not rank websites. They construct answers from entities they can understand, verify and trust. An organization without clear entity signals will be ignored, misrepresented or replaced by a competitor that AI systems can read more clearly.

The question is no longer whether you rank on Google. The question is whether ChatGPT can understand, cite and recommend you. The AI Citation Readiness Framework answers a question that comes before any AEO strategy: how do you measure where you are before you build?

What it measures: three dimensions, one score

Clarity

Is the entity unambiguously identifiable? AI systems must be able to determine who or what the entity is, what it does, and how it differs from similar entities, without guessing.

Consistency

Are the core claims about the entity identical across all sources? Conflicting names, descriptions or roles across websites, profiles and structured data create entity resolution failures.

Verifiability

Are there external, machine-readable anchor points that confirm the entity exists and is credible? GitHub, ORCID, Crunchbase, LinkedIn and structured data on the entity's own domain all function as verifiable signals.

The AI Citation Readiness Score

The three dimensions determine a single output: the AI Citation Readiness Score, a 0 to 100 measure of how ready an entity is to be understood, cited and recommended by AI systems. Four levels describe the result:

Score	Status	Meaning
0–25	Invisible	AI systems cannot identify or recommend the entity.
26–50	Recognizable	AI systems may find the entity but cannot reliably cite or recommend it.
51–75	Citable	AI systems can identify and cite the entity, but consistency and verifiability gaps remain.
76–100	Answer-Ready	AI systems can understand, cite and recommend the entity with confidence.

Most organizations that have never addressed their entity signals score below 40. The goal of AEO implementation is not a perfect score. It is a score high enough that AI systems consistently choose your entity over a competitor they can understand more clearly.

How to apply it: three steps from audit to architecture

Step 1 · Entity Audit

Map all existing signals for the entity: structured data, external profiles, sameAs references, published content, mentions and citations. Identify gaps, conflicts and missing anchor points.

Step 2 · Signal Assessment

Evaluate each signal against the three dimensions: Clarity, Consistency and Verifiability. Assign a score per dimension. The AI Citation Readiness Score is the weighted result.

Step 3 · Signal Architecture

Close the gaps. Structured data, external profile alignment, consistent claim formulation and verified anchor points are the primary tools. The goal is a coherent, machine-readable entity layer that AI systems can traverse without ambiguity.

A practical checklist for all three steps is included in the framework repository on GitHub.

Measurement meets strategy

The AI Citation Readiness Framework is a direct extension of the AEO Mastery Framework. The AEO Mastery Framework defines the strategic methodology: how organizations build entity signals, structured data and citation architecture to become visible in AI-generated answers. The AI Citation Readiness Framework answers a prior question: how do you measure where you are before you build?

Together they form a complete AEO methodology. The AEO Mastery Framework covers strategy and implementation. The AI Citation Readiness Framework covers measurement and diagnosis. Neither replaces the other. Measurement without strategy produces scores. Strategy without measurement produces assumptions.

Machine First: Why AEO Is Not SEO 2.0

Stefan Petschinka — Fri, 29 May 2026 23:08:34 +0000

Machine First is not a content strategy for machines. It is the structural condition under which answer engines, AI models and retrieval systems extract, verify and reuse information. This article explains why AEO requires a fundamentally different architecture than SEO and what that architecture consists of.

01 Core Thesis: AI does not rank. It reasons.

Search engines rank. Answer systems reason. This distinction is not semantic, it is architectural. A search engine returns a list of sources and leaves the decision to the user. An answer system forms a position and delivers it as a statement. The process that leads to that statement is not ranking. It is signal extraction, entity resolution and weighted inference. AEO is the discipline of structuring content, entities and data so that inference produces correct, citable, authoritative answers.

This architecture has two stages, and the first one still ranks. A retrieval layer selects which sources enter the model's context at all, and it works on classical relevance signals: topical match to the query, accessibility, index presence. Only the second stage, the synthesis, reasons over what was retrieved. AEO serves both stages: content that answers the actual question wins the retrieval, and clear entities with extractable statements win the synthesis. The evidence behind both stages is documented in the AI Visibility Evidence Model.

SEO optimizes for position. AEO optimizes for the answer itself. These are not the same problem. They require different methods, different architectures and a different understanding of what content is for.

02 Definition: AEO is not an SEO upgrade.

SEO is built on the premise that users search, results are ranked and clicks determine success. Every element of SEO, whether keyword density, backlink authority, crawl budget or page speed, serves that premise. The metric is position. The goal is to appear ahead of the competition.

AEO operates from a completely different premise. AI answer systems do not rank results. They construct answers. They do this by extracting entity signals from multiple sources, resolving identities, weighing corroboration and synthesizing a response. No click is involved. The question is not whether a source appears. The question is whether a source is understood well enough to be cited.

Machine First AEO is the structural approach of building content, entities and data so that answer systems can identify, extract, verify and reuse information with minimal ambiguity. It is not SEO with new vocabulary. It is a different discipline with different requirements and a different success metric: correct citation, not high ranking.

03 Architecture: How answer systems interpret content.

Entity Resolution

Before an answer system reads content, it asks a prior question: which entity does this content refer to, and is that entity known? Entity resolution is the process of mapping names, identifiers and signals to stable entries in a knowledge model. An organization without structured entity signals is not resolved. It is guessed. And guessing produces approximations, not recommendations.

Machine First consequence: Entity clarity decides whether relevant content is attributed to the right source. It does not replace topical relevance: a page that does not answer the question is not cited, however clear the entity. A well-written page about an unresolved entity feeds an approximation instead of a recommendation. The first layer of Machine First AEO is making the entity unambiguous.

Signal Extraction

Once an entity is resolved, the system extracts signals: what does this entity do, what does it know, what relationships does it have, what has it produced? Signal extraction is not keyword matching. It is structured inference from multiple content layers simultaneously: visible text, structured data, internal link architecture, external corroboration and authored content. Each layer reinforces or contradicts the others.

Machine First consequence: Content must be structured so that signals can be extracted without ambiguity. That means sentences that begin with the statement, not the context. Paragraphs that fully answer a question. Schema that mirrors what the visible content says, not decorates it.

Corroboration and Weighting

Answer systems do not cite individual sources. They weigh multiple sources against each other and produce answers with implicit confidence scores. A signal that appears on one page is a weak signal. A signal that appears consistently across the entity's own page, its structured data, its external profiles and authored content is a strong signal. Inconsistency lowers confidence and increases the likelihood of approximation.

Machine First consequence: Consistency is not a style question. It is a signal architecture requirement. The same name, the same role, the same organizational identifier must appear in every context where the entity is referenced. That is the Human Trust Layer, the signal architecture that tells a machine: this is verified, consistent and authoritative.

Answer Construction

The final step is the one users see: the system constructs an answer. That answer is not a reproduction of what one source says in full. It is a synthesis of extracted, weighted, corroborated signals. Sources structured to be extractable become the building blocks of that synthesis. Sources that require interpretation, context or extensive reading are deprioritized.

Machine First consequence: If the core statement of a page requires three paragraphs of context before it appears, the system will not wait. It will find the statement elsewhere or construct it from adjacent signals. Machine First content begins with the statement.

04 Framework: The four layers of a Machine First system.

Layer 1: Entity Layer. The foundation. Every entity that matters, whether person, organization, service or topic, must be declared with a stable identifier, a consistent set of attributes and verifiable external corroboration. The entity layer answers the question the machine asks before reading any content: what or who is this, and can it be verified?

Layer 2: Answer Layer. The content layer structured for extraction. Every page should answer a defined set of queries. Those answers must appear as isolatable passages, meaning paragraphs that begin with the direct answer, not the context. The answer layer is where most content fails: it delivers information but not answers. Information requires reading. Answers can be extracted.

Layer 3: Evidence Layer. The corroboration structure. Authored content, case documentation, external references and attributable proof that allow a retrieval system to assign confidence to a statement. An entity that says it is an expert is a weak signal. An entity described as an expert in authored articles, case documentation and external profiles is a strong signal.

Layer 4: Schema Layer. The machine-readable declaration layer. JSON-LD and Schema.org markup that exactly mirrors the visible content, not decorates it. The schema layer does not create signals; it clarifies and connects them. An Article schema that declares an author via a stable identifier reference connects that article to a Person entity to an Organization entity.

05 Entity Graph: The graph that closes the loop.

A Machine First system is not a collection of optimized pages. It is a graph. The graph connects entities through relationships that are declared, consistent and bidirectional. A Person entity connects to an Organization entity. An Organization entity connects to a Service entity. An Article entity connects back to the person who authored it, the organization that published it and the topics it addresses.

The strength of the graph is not in any single node, it is in the loop. When an answer system follows the signal from a person to an organization to an article to a topic and back to the person, it does not just find information. It builds confidence. Every traversal of the loop reinforces the same facts through a different surface. That is consistency at architectural level. It is the necessary structure, and it becomes corroboration when independent sources outside the graph confirm the same facts: the graph makes an entity verifiable, witnesses make it verified.

AI Answer Control, the ability to influence what AI systems say about an organization, is not achieved through better content alone. It is achieved by building a graph that minimizes the room for inference. When every node in the graph says the same thing about an entity, the answer system does not guess. It reasons. And it cites the source that gave it the clearest signal.

06 The Core Shift: Human reputation does not automatically become machine-readable evidence.

The most reputable organizations in the world are frequently invisible to AI systems. Not because they lack content. Not because they lack reputation. Because they lack the structural signals that answer systems need to resolve, verify and cite them.

A law firm with forty years of documented expertise and a 10,000-word website is invisible to an AI answer system if its entity signals are missing, inconsistent or unverifiable. A competitor with two years of history and a correctly structured entity graph gets recommended. That is not unfair. That is the architecture of the system. Machine First AEO is the discipline of working with that architecture, not against it.

The gap between human reputation and machine-readable evidence is the operational space of AEO. Closing it is not a marketing decision. It is an architecture decision.

Machine First. Four Principles.

1. Entity clarity takes priority over content production.

Before optimized content is created, the entity architecture is defined: the primary entity, its stable identifier, its consistent attributes such as name, role, organization and specialization, as well as the external profiles that confirm it. These are not content questions, they are identity architecture questions. Content produced before the entity is defined may reinforce the wrong signals. Entity clarity does not replace topical relevance; it makes sure that relevant content is credited to the right entity.

2. The schema layer mirrors the content layer, it does not decorate it.

Schema that contradicts the visible content is worse than no schema, because it introduces ambiguity into the signal architecture. The schema layer is a precise machine-readable mirror of what the page visibly states. If the page says a person is AEO Strategist, the schema says the same. If the page says the organization was founded in 2026, the schema says 2026. Decorative schema dilutes signals instead of strengthening them.

3. The graph loop is closed before publication.

A published page not connected to the entity graph remains an isolated node. It will be indexed, it will be read, but it will not be cited with confidence, because the answer system cannot confirm its claims by traversing related entities. The graph loop is not a technical nicety. It is the architecture that transforms isolated content into a confirmed signal.

4. Every paragraph begins with the extractable statement.

Answer systems extract, they do not read. A paragraph that builds to its conclusion over four sentences delivers only the first sentence to the extraction process, and if that first sentence is context rather than statement, it is a weak signal. Machine First content architecture requires every paragraph to begin with its central statement. That is not simplification, it is restructuring so that the most important information is also the most easily extractable.