There are now more than a hundred guides and articles on this site covering SEO and AI visibility. Published one at a time, they read as a stream. Read together, they are a single argument that developed in a particular order, and the order is the useful part.
This page is the map. It is not new analysis and it is not a sales pitch. It is a record of how the thinking developed, what each phase concluded, and where to go deeper. If you are new here, it is also the reading sequence: start at the phase that matches your current question rather than the top.
One pattern runs through all of it. The work moved steadily from describing AI visibility toward measuring it. The early pieces map the territory: what AI search is, how each platform behaves, which layer your problem sits in. The later pieces stop describing and start instrumenting: scoring pages, logging what AI systems actually do, and testing the frameworks against first-party data instead of asserting them. That shift, from map to instrument, is the spine of everything below.
Here is the same year in prose, walked forward from the foundations.
2025: foundations and the unbroken thread
The starting position was deliberately unfashionable: AI search is not a new game, it is the same discipline expanding. Before any of the AI-era material, the foundations went down first. What SEO actually is in 2026, semantic SEO and entity architecture, structured data, site architecture, Wikidata. These are the load-bearing pieces the later work assumes.
The thesis that ties them together is that every new acronym, AEO, GEO, AIO, AAO, describes an execution shift on top of the same compounding-authority discipline that worked in 2010. The businesses that understood that a decade ago hold the same structural advantage now. The medium changed; the mechanism did not.
Go deeper: Entity SEO and Site Architecture.
Late 2025 to early 2026: the platforms, and a model to organise them
The next phase was practical and platform by platform: how to rank in ChatGPT, Perplexity, Gemini, Copilot and Claude, each a separate retrieval system with a different source pool and a different speed of response. Useful individually, but a pile of platform guides is not a strategy.
The model that turned the pile into a system was the AI Discovery Stack: five layers, Understanding, Retrieval, Selection, Recommendation, Action, that every AI system moves through. Once the Stack existed, every platform guide and every later framework could be placed inside a specific layer rather than floating as its own idea. The lesson: platform tactics without a layer model produce activity, not direction.
Go deeper: The AI Discovery Stack and the AI Recommendation Pipeline.
Spring 2026: the AI-era trust architecture
This is the densest and most original phase, a four-keystone sequence, each building on the last. Schema Architecture argued the web is moving from machine-readable to machine-verifiable. Footprint vs Fingerprint made the editorial-record question measurable: a footprint is content a competitor could publish a near-identical version of tomorrow with the same prompt; a fingerprint is content only one entity could have produced. Editorial Selection defined the per-event mechanism by which AI-era trust accumulates, an editor or system choosing to include you through independent judgement rather than paid placement. Retrieval Gravity completed the set, explaining how AI systems accumulate preference for entities they have already validated.
Together these four are the trust half of the discipline: not how a page describes itself to machines, but whether independent sources corroborate it and whether that corroboration compounds. The lesson: AI-era trust is earned per event and compounds over time. It cannot be bought in a quarter.
Go deeper: Schema Architecture for the AI Era, Footprint vs Fingerprint, Editorial Selection and Retrieval Gravity.
2026: the measurement turn
This is the pivot, where the work stops describing and starts instrumenting. CITATE defined a six-criterion, pass or fail threshold for whether a page is extractable, evidenced and attributable enough to be cited, a measurement standard rather than another concept. The Observed Outcomes Register began logging documented AI retrieval, citation and feature-extraction behaviour from real client work, evidence rather than opinion. The thesis was then stated outright: descriptive frameworks build the site, measurement frameworks keep it cited. The self-directed version turned the entity-corroboration model on its author and documented where it found him short.
A field full of people making claims about AI visibility has very few systematically documenting what AI systems actually do, and fewer still willing to run their own framework against themselves in public. The lesson: a framework you can measure beats one you can only assert, and the honest version of that is publishing the measurement when it does not flatter you.
Go deeper: The CITATE Framework and the Observed Outcomes Register.
2026: the agent layer
As the trust and measurement work matured, the frontier moved to execution: AI that acts rather than answers. The arc ran from MCP through WebMCP and the practical guide to making a web form agent-actionable, to Google’s Agentic Browsing audit in Chrome and the proof that this site passes it. The throughline is the Four-Floor Model: entity foundation, extractability, trust, and now agentic execution as the top floor.
The lesson: agentic readiness is a real early edge, but it is the top floor. Building it before the foundation, extractability and trust floors are solid optimises a layer the buyer never reaches.
Go deeper: WebMCP, How to Make a Web Form Agent-Actionable and the Agentic Browsing Audit.
Mid 2026: the editorial and disclosure thread
The most recent phase applies the trust architecture to a live problem: AI Overviews citing vendor-written best-of lists as if they were independent editorial consensus. The cluster runs from the Best-Of problem and the Listicle Question to the synthesis pillar, which adds the legal and regulatory layer, the FTC, ASA and CMA, that the others leave out. This is Editorial Selection and Footprint vs Fingerprint applied at close range, with the disclosure law the rest of the discourse tends to skip.
The lesson: the line between earned editorial inclusion and self-promotion dressed as consensus is now both a retrieval question and a regulatory one. Most of the market is only treating it as the first.
Go deeper: The Self-Promotional Listicle, Demystified, The Listicle Question and The Best-Of Problem.
The architecture the year produced
The year did not just produce articles. It climbed a structure. The work moved from making an entity findable, to making its content extractable, to making it trusted enough to be selected, to making it actionable by agents. That progression is the Four-Floor Model, and it is the map the year resolves into. Where the timeline above travels through time, the model below shows what depends on what. Most businesses are not failing for lack of content; they are failing at Floor 2 or Floor 3, where content is not structured for selection and trust signals are not independently corroborated.
How to use this
If you know something is broken but not what, start at the AI Discovery Stack and find your failing layer. If you are deciding whether to scale AI-assisted content, start with Footprint vs Fingerprint. If you are weighing a paid-visibility opportunity, start with Editorial Selection. If you want to know whether any of this is measured rather than asserted, start with CITATE and the Observed Outcomes Register, which is where the next phase of this work is heading: less describing, more measuring.
