Slobodan Manic published Machine-First Architecture in Search Engine Journal on 26 May 2026. It is a four-pillar build framework (Identity → Structure → Content → Interaction) for designing websites that AI systems can identify, read, cite, and transact with. The architectural sequence and the dependency chain in his piece map closely to the Four-Floor Model that has been the spine of this site’s writing for six months, and to Mike King’s Beyond RAG decomposition published at iPullRank earlier in May. Three senior practitioners independently arriving at near-identical architectural sequences is meaningful directional evidence the underlying pattern is real. It is not evidence that any one framework’s specific implementation is correct. This piece is about what happens after the architecture is built — what happens to a 6/6 page on day 500, when the architecture is still right but the pages have quietly drifted.
The convergence is directional. Don’t overclaim from it.
The architectural sequence Slobodan describes is recognisable. Identity before content. Content before interaction. Each pillar depending on the one before it. That is the same logic the Four-Floor Model uses (Entity Foundation → Content Extractability → Trust and Selection → Agentic Execution), and the same logic Mike King’s pipeline-to-floors mapping uses across the five stages of the Beyond RAG decomposition.
Three honest reads.
First, when multiple senior practitioners independently identify the same architectural sequence with the same dependency chain, that is directional evidence the pattern reflects something real about how AI systems actually work. It is not evidence that any one framework’s specific implementation, scoring model, or terminology is correct. Convergence at the architectural level does not extend automatically down to operational specifics.
Second, no individual practitioner is going to own the agentic layer. OpenAI, Google, Anthropic, Microsoft, and Perplexity are all building it simultaneously. The W3C and the Linux Foundation are shipping the protocols (MCP, A2A, WebMCP, the Agentic Commerce Protocol, the Universal Commerce Protocol, the Visa Trusted Agent Protocol). The architectural territory is being defined by the platforms. What individual practitioners can do is help businesses operationalise the work inside those protocols. That is a narrower position than category ownership. It is also a stronger one.
Third, the genuinely important shift in Slobodan’s piece is not the four-pillar framework itself. It is this sentence buried inside the Structure pillar:
The rendered page is one possible output of the data model. AI search results, voice answers, agent tool calls, and chat citations are other outputs the same data model has to serve.
That is the architectural inversion most agencies still do not understand. Page first, crawler second, AI layer third is the old sequence. Knowledge model first, machine extraction second, interfaces as presentation layers is the new one. That inversion is the part of the piece worth amplifying loudest.
Descriptive frameworks and measurement frameworks do different work
Reading Machine-First Architecture next to the Four-Floor Model next to CITATE clarifies that these are not competing frameworks. They are different layers of the same discipline.
| Framework type | What it does | What it cannot do alone |
|---|---|---|
| Descriptive (MFA, Four-Floor Model, Beyond RAG decomposition) | Names the architectural sequence, explains the dependency chain, makes the order non-negotiable | Tell you whether a specific page in production today passes the criteria the architecture implies |
| Measurement (CITATE, scoring models, pass/fail diagnostics) | Operationalises the architecture, generates pass/fail per page, makes degradation visible | Tell you how to architect the system if you are starting from scratch |
| Build (AI Page Anatomy, schema implementation guides, content structuring patterns) | Provides the patterns for individual pages and components | Tell you whether the page held up six months after publication |
A descriptive framework tells you what to build and in what order. A measurement framework tells you whether what was built is still working. A build framework tells you how to construct the components. The three operate at different layers and need each other.
The strategic question is never which framework is best. It is which layer the specific commercial problem in front of you actually lives at.
Designing a new site for a B2B SaaS that does not yet exist? Descriptive frameworks earn their keep. They tell you what to build and in what sequence. Operating a 150-page legal services site that was built in 2019 and has been edited by twelve different people since? You do not need a description of what should have been built. You need to know which specific pages are still cited, which ones are drifting, and which ones have already failed.
That is the brownfield problem.
Where descriptive frameworks stop being enough
Machine-First Architecture is a greenfield framework. So is the Four-Floor Model when read as architectural guidance. Both assume you are designing the system.
The Identity pillar assumes you get to write the canonical definition before the platforms have already extracted twelve incompatible versions of your brand from twelve different sources. The Structure pillar assumes you get to define the data model before the page is designed. The Content pillar assumes the author entity is being established at the same time the content is being written.
Most commercial SEO work does not happen in those conditions.
What it happens in is this. An estate of 30 to 200 pages, built over five to ten years by different people, with different priorities, at different points in the platform’s development. Some pages have proper schema; some do not. Some have named authors; some have generic bylines. Some have inline statistics with named sources; some have orphaned claims that got pasted in during a 2022 update and never got re-attributed. Some have machine-readable opening paragraphs; some have marketing intros that were strong in 2018 and unreadable to a retrieval system today.
The architecture, where it exists, may still be right. The pages have drifted.
This is the operational territory where measurement frameworks earn their keep.
The 6/6 to 3/6 drift problem
Specific drift patterns recur across every brownfield estate I have worked on.
A page published in March 2024 with a named author, an inline statistic, and a specific attribution gets a routine content refresh in October 2025. The refresh updates the body copy and the call to action. The named author is replaced with “Our editorial team” because the original author has left. The inline statistic — “30-40% citation rate for pages with definitions in the opening paragraph (Princeton GEO-Bench, 2024)” — gets generalised to “research shows higher citation rates for pages with definitions in the opening paragraph” because the editor felt the specific number aged the piece. The dated claim is dropped. The page now scores 3/6 against CITATE where it scored 6/6 at publication.
Nobody flags it. The CMS does not surface the change in CITATE terms. The original author is no longer there to defend the structure. The marketing manager who oversaw the refresh was working from a content brief that did not mention citation criteria. Six months later, the page that used to surface in AI Overviews has been replaced in that slot by a competitor whose page scored 5/6 at publication and has held.
This pattern recurs because content editing systems were designed to optimise for human readability, not machine-readable structure. The named author was an editorial preference, not a structural element. The inline statistic was a stylistic decision, not a citation requirement. The specific attribution was a reference, not a Floor 3 trust signal. The brownfield problem is what happens when content editing happens without measurement underneath it.
A measurement framework catches this. Every release, every page, the score is recalculated. Pages that drop are flagged. Pages that drift get remediated. The score is not a one-time audit. It is continuous governance.
What this looks like in production
The seostrategy.co.uk site runs CITATE scoring on every page on every release. 115 pages, 9 case studies, 29 posts. Each release produces a per-page score and a delta from the previous release. Pages that drop are flagged in the admin bar. Drift is visible at the page level before it shows up in citation data.
This is not strategy. It is governance.
Most SEO work is sold as strategy and delivered as one-time work. The reason measurement frameworks matter is that AI citation is not a one-time outcome. It is the result of an estate of pages each independently meeting criteria at a specific point in time, and continuing to meet them as the estate evolves.
The bigger shift the convergence points at
The conversation Slobodan’s piece is part of is bigger than any individual framework.
What is actually happening across SEO, GEO, AEO, AAO, and the various successor disciplines being named monthly is that the discourse is moving up-stack. Historically these were separate concerns. SEO for retrieval. Schema for enhancement. CRO for conversion. UX for humans. APIs for engineering. Accessibility for compliance. Each had its own practitioners, its own vocabulary, its own deliverables.
What Machine-First Architecture, the Four-Floor Model, the AI Discovery Stack, and Mike King’s pipeline-to-floors mapping are all doing is collapsing those separate concerns into a single machine-consumption continuum. The page is one output of the data model. The AI citation is another output. The agent action is another. They all read from the same underlying knowledge structure.
The agencies that internalise this (knowledge model first, interfaces as presentation layers) will keep up with where the discipline is going. The ones that keep building page-by-page with the AI layer treated as an afterthought will not. That is the directional consensus that matters.
What multiple senior practitioners converging actually tells you
Multiple senior practitioners converging on similar architectural arguments tells you the architecture is real. It does not tell you any one practitioner’s framework is the right framework. The territory is being defined by the platforms — by what Google’s, OpenAI’s, Anthropic’s, Microsoft’s, and Perplexity’s systems actually read, cite, and act on — not by any individual consultant’s nomenclature.
The work that earns its keep is operational. Build the architecture. Then measure whether the estate continues to express it. Descriptive frameworks like Machine-First Architecture name the sequence. Measurement frameworks like CITATE catch the drift. Both layers are necessary. Neither is sufficient on its own.
Descriptive frameworks build the site. Measurement frameworks keep it cited. The brownfield drift problem is what measurement frameworks are for. — Sean Mullins, SEO Strategy Ltd, 2026
The page that scored 6/6 at publication and now scores 3/6 is the operational problem this discipline still has to solve. It will not be solved by any single framework. It will be solved by the businesses that decide to measure. See the SEO Strategy Frameworks register for the full set of named frameworks behind the measurement work, or the AI Visibility Audit for the commercial engagement this discipline lives inside.
