AI visibility is the umbrella discipline. The AI Visibility Audit diagnoses your specific position across all three floors. This page defines the discipline, maps the infrastructure it requires, and explains why the stakes are higher for enterprise and regulated-sector businesses than most frameworks currently acknowledge.
What AI Visibility Is — and What It Is Not
Search engine optimisation determines where a page ranks when a user types a query and receives a list of links. That mechanism is not going away. But it is no longer the only mechanism that matters, and for a growing proportion of buyer journeys it is no longer the primary one.
AI visibility operates differently. When a user asks ChatGPT to recommend a managed file transfer solution for a healthcare IT environment, when a procurement team uses Microsoft Copilot to identify legal representation in their sector, when a founder asks Perplexity which SEO agencies specialise in SaaS — the AI system does not return a ranked list of links and invite the user to browse. It generates a response that either names specific providers or does not. There is no position three. There is no second page. There is present or absent, and the infrastructure that determines which outcome a business gets is not its ranking. It is whether it has built the three floors that AI selection requires.
The Three Floors
The Four-Floor Model maps the sequential dependency layers that determine AI visibility. Each floor is a prerequisite for the one above it. A business that fails Floor 1 cannot benefit from Floor 2 work. A business that fails Floor 2 cannot benefit from Floor 3 investment. Floor 4 is the horizon layer that becomes commercially significant as agentic infrastructure matures.
Floor 1 — Retrieval Eligibility
Can AI systems find the business? Google AI Overviews retrieves from the Google index. ChatGPT Search and Microsoft Copilot retrieve from the Bing index. Perplexity uses its own crawler, PerplexityBot, which must be explicitly permitted in robots.txt. Claude uses ClaudeBot. Each has separate access requirements. A single legacy NOARCHIVE directive can remove all Copilot visibility. Absence from the Bing index removes ChatGPT Search and Copilot eligibility regardless of Google ranking. Floor 1 failures are invisible in Google Analytics and Google Search Console — they produce no signal until you check the specific platforms where they are occurring.
Floor 2 — Content Extractability
Can AI systems cite the business? AI systems retrieve at paragraph and section level, extracting fragments out of sequence from pages they have decided to consult. Content written for human readers reading linearly — context-dependent paragraphs, unnamed entities, qualitative claims without numbers — is retrieved but not cited. The CITATE framework defines the six structural criteria that determine whether content crosses the extraction threshold. Research from Princeton, Georgia Tech and IIT Delhi found structured content interventions consistent with CITATE criteria improved citation rates by 30–40% in controlled testing.
Floor 3 — Recommendation Eligibility
Will AI systems name the business? Muck Rack’s analysis of over one million AI response links found 82% of all AI citations come from earned media — independent editorial, not brand-owned content. AI systems do not select businesses based on what those businesses say about themselves. They select based on what independent sources say: editorial coverage, review platform profiles, structured entity databases. The selection mechanism is corroboration, not assertion. A business that has only ever declared its own expertise is invisible at Floor 3 regardless of how well-structured its website is.
The Selection Layer
Floor 3 is governed by what Sean Mullins, SEO Strategy Ltd terms the Selection Layer — the stage of AI information processing in which the system determines which businesses to name as recommended providers. The selection layer is distinct from retrieval (which sources can the AI access?) and extraction (which sources can the AI cite?). It is the commercially decisive stage, and it is the stage most AI visibility frameworks fail to address directly.
A business that passes retrieval and extraction but fails the selection layer is used as an anonymous source. The AI cites its content without attributing it. The user receives the information but has no reason to contact the business. This is the most common failure mode in AI visibility work — and it is invisible in standard analytics, which cannot distinguish between traffic that arrived having been recommended by name and traffic that arrived having been sent to a page the AI cited without naming the provider.
AI systems do not choose the best business. They choose the business they can most confidently justify selecting. That is a structural distinction. Trust signals, corroboration, and attribution consistency outweigh quality claims, content volume, and keyword coverage. Absence at this layer does not mean ranked lower. It means not considered.
The selection layer is governed by three signals: entity consistency (does the AI encounter the same business identity across multiple independent sources?), editorial corroboration (do independent publications reference this business in contexts relevant to the buyer query?), and trust infrastructure (review platforms, professional directories, structured databases, named credentials). Building these signals is Floor 3 work. It takes months, not weeks. The businesses that start now will hold compounding advantages when the competitive field wakes up to the distinction.
A useful demonstration of how selection actually operates: DigiMarCon, a conference series with a few hundred attendees per city, consistently surfaces in AI answers as the “world’s number one digital marketing conference series” — ahead of events with audiences ten times larger. This does not happen because the claim is factually accurate. It happens because the claim has been repeated across multiple independent surfaces, referenced by attendees, reinforced in emails and event listings, and associated with real interaction patterns. AI systems select what is most corroborated and most reinforced — not what is most true. That is the Selection Layer in operation.
One signal worth naming explicitly: independent experts and digital PR practitioners now confirm that unlinked brand mentions carry similar weight to followed links in AI selection processes. The mechanism is the same — both reinforce entity association in a system that processes co-occurrence rather than hyperlink topology. The practical implication is that editorial mentions without links are not worthless. They are Floor 3 signals and should be treated as such.
Why Enterprise and Regulated Sectors Reach Floor 4 First
For most businesses, Floors 1–3 of the Four-Floor Model describe the AI visibility challenge that determines commercial outcomes today: can AI find you, extract from you, and trust you enough to name you. Floor 4 — agentic execution — is the horizon layer for most. For enterprise businesses and those operating in regulated categories (healthcare IT, legal, financial services, managed file transfer, compliance-driven software), Floor 4 arrives earlier than it does for the rest of the market. The reason is mechanical, not strategic.
Agentic AI is transitioning from answering questions to completing tasks. Google has already deployed agentic restaurant booking globally — UK included — in which AI agents scan multiple platforms simultaneously and complete a reservation without the user navigating any website. Google CEO Sundar Pichai has described search evolving into an “agent manager” by 2027. That transition does not arrive uniformly. It arrives first in low-stakes, high-frequency decisions (restaurant bookings, travel, consumer shopping) and progresses toward higher-stakes, lower-frequency decisions (vendor shortlisting, supplier evaluation, procurement). The enterprise buying cycle is not immune. It is later in the sequence.
When AI agents begin performing or heavily pre-structuring vendor evaluation — querying capability signals, comparing against requirements, generating shortlists for human approval — the businesses that benefit are not necessarily those with the best marketing. They are those with the most structured, machine-readable, attributable capability data. A vendor that publishes its capabilities in prose marketing copy is opaque to an agent performing an automated evaluation. A vendor whose capabilities are published in a structured, scored, versioned format that an agent can query and compare is legible. That distinction will determine shortlist inclusion before a human buyer ever makes a decision.
This is the commercial case for the OARCAS framework beyond its original application in managed file transfer evaluation. OARCAS defines five capability dimensions — Orchestration, Automation, Reliability, Control, Security — each scored 1–5 with published rubrics. These are not arbitrary dimensions. They map directly to the constraints an autonomous system must evaluate when executing a regulated workflow: what does this system orchestrate, how deeply is it automated, how resilient is it, how auditable, how secure? It was designed as a human procurement tool. The five dimensions it scores are precisely the capability signals an autonomous agent needs when evaluating a vendor for a regulated, automated workflow — the questions enterprise buyers ask today, and that enterprise agents will ask on those buyers’ behalf tomorrow.
Research commissioned by FT Longitude and published in 2025 found that only 2% of AI agents currently deployed are fully governed. The organisations deploying ungoverned agents are the same enterprise buyers whose procurement decisions vendors are trying to influence. The governance gap in those agents creates a different kind of risk: an agent without governance boundaries may shortlist vendors based on incomplete or manipulable signals. The OARCAS model — published, versioned, independently applicable — is designed to be the kind of structured evaluation input that governs, rather than exploits, that gap.
From Content to Capability
Traditional digital visibility is based on content — pages, articles, explanations. AI visibility introduces a second requirement: capability representation. Content explains what a business does. Capability data allows a system to evaluate whether it can be used. In agentic environments, evaluation precedes selection, and selection precedes execution. Businesses that cannot express capability in structured, comparable terms are not evaluated and therefore cannot be selected. This is the transition most businesses have not yet made — from publishing content about what they do, to publishing structured signals about what they can reliably deliver.
AI Visibility vs SEO, AEO, and GEO
These terms are used interchangeably in most industry commentary. They are not the same, and conflating them produces misdirected investment.
SEO optimises for rankings in search engine results pages. The outcome is traffic. The mechanism is matching content to query signals that ranking algorithms evaluate. SEO is Floor 1 infrastructure for AI visibility — it provides the technical foundation that AI retrieval builds on — but it does not produce Floor 2 or Floor 3 outcomes automatically.
AEO (Answer Engine Optimisation) optimises content to be extracted and used as direct answers by AI systems. AEO is Floor 2 work. It addresses content extractability: standalone opening answers, explicit definitions, statistics with named sources, attributable claims. The CITATE framework is the scored, reproducible standard for AEO implementation.
GEO (Generative Engine Optimisation) optimises a brand’s presence across generative AI platforms. GEO is the combination of Floor 2 extractability and Floor 3 recommendation eligibility — the complete programme rather than one layer of it.
AI visibility is the umbrella discipline that encompasses all three surfaces, maps them to the infrastructure that each requires, and adds the agentic procurement layer that AEO, GEO, and SEO frameworks have not yet addressed. The AI Discovery Stack maps the five technical layers that govern AI visibility from retrieval through to agentic execution.
The Commercial Consequence of Getting This Right Early
Ahrefs analysis of 863,000 keywords in 2026 found that 62% of AI Overview citations come from pages outside the organic top 10. The citation layer and the ranking layer are diverging. The businesses capturing early citation positions are building a compounding advantage: cited sources become training data, training data reinforces future citation, citation frequency produces the independent corroboration signals that govern Floor 3 selection. This is not a slow-accumulating advantage that can be closed by a better-funded competitor arriving later. It is a structural position that becomes progressively harder to displace.
Seer Interactive found that traffic arriving via AI citation converts at 14.2% versus 2.8% for standard organic search. That gap will widen as AI-mediated discovery moves from informational queries toward commercial ones. The businesses visible in AI at the informational stage — when buyers are still in research mode — are those that will be named at the recommendation stage, when buyers are ready to act.
If you are not named, you are not contacted. The path from invisible to named provider follows a fixed dependency sequence — retrieval, extraction, selection — and each layer requires the one before it. The AI Visibility Engagement Model maps this sequence in full: what each stage costs, what it changes, and why the order is non-negotiable.
