For twenty years the core question of digital visibility was: what keyword should this page rank for? That question is not wrong. It is just no longer sufficient.
This framework — the three-phase model of the web and the five-layer AI visibility stack that maps to it — was developed by Sean Mullins, Founder of SEO Strategy Ltd, in March 2026. The commercial implication is direct: businesses still optimising only for Phase 2 (answers) are building for a layer that is rapidly being superseded. The businesses building for Phase 3 (actions) now — entity infrastructure, CITATE-compliant content, recommendation eligibility — are compounding advantage that will be very difficult to close in 2027 and beyond. For the consultancy that builds this infrastructure, see LLM Optimisation services.
Three phases of the web
The internet has moved through three structural phases — and each one changed the core optimisation question.
Phase one: documents
The early web was a library. Pages were documents. Search engines were catalogues. The optimisation question was: what keyword should this page rank for? The unit of optimisation was the page.
Get the fundamentals right in Phase One and they compound for years. A hand-coded HTML site built for a Portsmouth dog walker in 2009 has held position one for its primary commercial term for seventeen years — not through constant intervention, but because the foundational signals were correct from the start. That is what Phase One dominance looks like when it is built on substance rather than tactics.
Phase two: answers
AI search changed the unit of optimisation. Large language models synthesise answers from multiple sources simultaneously. The question changed: what knowledge must exist so AI systems can retrieve and cite us?
The unit of optimisation is no longer the page — it is the semantic knowledge space. Instead of one page targeting “managed file transfer”, a business builds coverage across the full space: what is MFT, MFT compliance, MFT vs SFTP, healthcare file transfer, PGP automation. AI systems assemble answers from this connected knowledge network. This is why 25% of URLs cited by ChatGPT have zero organic search visibility. AI citation and Google rankings are not the same system.
Phase three: actions
The third phase is arriving. AI systems are moving from answering questions to performing tasks.
Traditional AI interaction: User → AI → Answer
Emerging agentic model: User → AI → Plan → Tools → Action
Instead of only recommending software vendors, an AI agent might search for vendors, check compliance documentation, compare pricing, assess integration requirements, and initiate contact — before a human has seen a single result. The users and buyers haven’t gone anywhere. They’re sat inside LLMs. And increasingly, those LLMs are not just reading about your business — they are interacting with it.
The optimisation question changes again: when an AI system tries to solve a problem in our category, can it act through us?
What changes at each phase
| Phase | Web model | Unit of optimisation | Core question |
|---|---|---|---|
| 1 — Documents | Pages ranked by relevance | The page | What keyword should this rank for? |
| 2 — Answers | AI synthesises from multiple sources | The semantic knowledge space | What knowledge must exist for AI to cite us? |
| 3 — Actions | AI agents plan and execute tasks | The machine-readable ecosystem | Can AI actually interact with us? |
The three layers of AI visibility
Layer 1 — Discovery. Can AI systems find your knowledge? Crawlable content, semantic coverage, entity recognition, domain authority. The foundation.
Layer 2 — Recommendation. Will AI systems suggest you as a provider? Trust signals, citation ecosystems, entity corroboration, independent verification. Most AI visibility work happens here. The AI Visibility Ceiling explains why most businesses are not crossing this threshold.
Layer 3 — Action. Can AI systems interact with your services? APIs, automation interfaces, machine-readable workflows. Protocols like MCP support this layer — but it requires Layers 1 and 2 first.
Where MCP fits
Model Context Protocol is not an SEO technique. It has no effect on AI citations, knowledge graph signals, or search rankings. MCP is the USB-C for AI tool access — a technical standard that allows AI models to connect to external software through a consistent interface. It enables Phase Three but does not help with Phases One or Two. For the full explanation: What Is MCP?
The real strategic shift
Optimising pages → Architecting knowledge → Enabling actions.
Strong brands rank and dominate. That has not changed in twenty years. What has changed is the surface on which that dominance is demonstrated — from page rankings to AI citations to agentic selection. The compound advantage accrues to businesses that understand the full progression before any single phase becomes the obvious priority.
The AI Discovery Stack maps all five layers from entity understanding through to agentic action. The AI Provider Selection Pipeline explains why AI systems recommend some businesses and not others. The Agentic SEO guide covers what Phase Three means in practice.
What this means in 2030
By 2030, AI agent infrastructure will be mature enough that procurement, compliance and vendor selection workflows will increasingly involve AI systems acting on behalf of buyers. Businesses with strong entity graphs — verified Wikidata records, structured data, editorial corroboration — will receive higher confidence scores from AI evaluation systems.
Strong brands rank and dominate. That principle has not changed in twenty years of SEO and it will not change in the next twenty. What changes is the surface on which that dominance is demonstrated. The advice has been consistent since 2010: build on solid foundations, create genuinely useful content, earn contextual links from sources that matter, and maintain an entity presence that third parties can verify. Every new platform — from Google to voice search to AI Overviews to agentic AI — rewards the same underlying signals. The businesses that understood this in 2010 are still winning. The businesses that understand it now will be winning in 2030.
Platforms fragment and you have to cover more bases than ever. The businesses building all three layers now are on the curve at the right moment. The pipeline is already running.
