]> AI Visibility Architecture Group Limited

Auckland New Zealand bernard@aivisibilityarchitects.com https://aivisibilityarchitects.com

This document describes the 11-Stage AI Visibility Lifecycle, a proposed analytical model of how AI systems discover, understand, trust and show websites to people. The eleven stages fall into three related phases -- AI Comprehension (Stages 1-5), Trust Establishment (Stages 6-8), and Human Visibility (Stages 9-11). The lifecycle is non-linear: Stages 1-2 come first for any page; Stages 3-11 are weighed together, and a domain can be progressing in several at once. Its figures are illustrative examples (analytical estimates), and its descriptions of how AI systems work are inferred from observed behaviour, not documented internals. Its contribution is a common vocabulary and a dependency map for telling crawlability from visibility and for measuring each element, with the claims that rest on outside research cited to it in the deposited specifications. Its figures and some of its mechanisms are provisional and remain to be tested. Canonical Source Notice This Internet-Draft is NOT the canonical source for the AI Visibility Lifecycle framework. The framework is deposited at Zenodo under the concept DOI , which resolves to the latest deposited version. This Internet-Draft mirrors the specification for IETF community accessibility and reflects framework version 0.7.1 (October 2026). Where this Internet-Draft and an earlier deposited version differ, the later version governs.

Introduction The AI Visibility Lifecycle (version 0.7.1) provides a structural model for understanding how AI systems discover, evaluate, trust, and surface websites to human users. This framework is observational and analytical, not prescriptive. This document does not propose a standard, protocol, or recommendation for implementation. Every figure in the framework -- timelines, success rates and conditions met, trust thresholds, exposure levels and classification bands -- is an illustrative example (analytical estimate) drawn from observation, not established fact, and remains to be tested. Descriptions of how AI systems work are the framework's model, inferred from observed behaviour, not documented internals. In the framework specification and its stage specifications, a statement that carries a citation rests on the sources cited; a statement without one is the framework's model, inferred from observation and still to be tested. This document summarises them; the citations are in the deposited specifications. The lifecycle does not claim that any AI platform runs eleven internal stages: systems may combine, reorder, repeat or leave out what the stages describe. A domain's standing can differ by page, topic, query, country, platform and time, so a stage given for a whole domain is an aggregate. 'Lifecycle' describes how a domain's visibility progresses, not a sequence of processing steps; the instruments for measuring each stage are kept separately. This document mirrors the framework specification, version 0.7.1; earlier versions are deposited at Zenodo . Companion papers on ambiguity elimination and website visibility reporting provide additional context. The framework is also being developed collaboratively through the W3C AI Visibility Lifecycle Framework Community Group .

Framework Overview A stage is a condition in a dependency map, not a step in a strict sequence. Stages 1-2 come first for any page: a page must be discovered and ingested before it can be used. Stages 3-11 are weighed together as parallel dimensions of evidence, and a domain can be progressing in several at once; total journey time equals the slowest-resolving dimension, not the sum of all stages. There are no shortcuts: weakness in any of Stages 3-11 lowers the chance of human-facing visibility. The work sits in two layers: the ranking systems, which decide which pages and sources are candidates, and the language model, which reads them and writes the answer. Visibility has two surfaces: placements in search results, and, in AI answers, a share of appearances around a stable core of sources. The eleven stages are organised into three related phases:

Phase 1: AI Comprehension (Stages 1-5)

AI systems discover, interpret, classify, harmonise, and check the domain's knowledge structure for corroboration with established knowledge from authoritative sources.

Phase 2: Trust Establishment (Stages 6-8)

AI systems build trust, formally accept the domain as reliable, and shortlist it for answers (candidate surfacing).

Phase 3: Human Visibility (Stages 9-11)

AI systems observe the domain's first user interaction signals, establish baseline visibility, and scale visibility based on positive performance.

Memory and Retrieval Google describes the first step for its Gemini models in its own words: "The model analyzes the prompt and determines if a Google Search can improve the answer" . The understanding the lifecycle takes from this is that memory steers retrieval. Memory is made of the web the model has already crawled, processed and learned to trust, and it is redrawn and updated as models are retrained. It decides question by question whether a search runs, and it shapes the searches the model writes and the candidates it picks. Every AI answer that draws on the web therefore passes through the lifecycle at some point: at answer time, through retrieval, or earlier, when the web the model learned from was crawled and processed. The lifecycle describes how a domain reaches AI answers. It is measured on the retrieval route, and the same stages bear on what models learn and carry in memory.

Stage Definitions

Stage 1: AI Crawling Discovery: crawling, fetching and rendering. AI systems discover the domain through URL submissions, sitemaps, inter-domain signals, or autonomous exploration, and pages are fetched and prepared for analysis. How often a page is crawled can follow how much search traffic it already receives. No trust exists yet.

Stage 2: AI Ingestion (in search, indexing) Extracted content is converted into internal semantic representations that can later be reasoned over, compared, and synthesised, and the domain's concept structure is recorded. For AI answers built on a search engine's index, this work is done by the search engine at indexing.

Stage 3: AI Classification (Purpose and Identity Assignment) AI systems determine what kind of website a domain is: non-commercial, commercial, or hybrid, by its site purpose. This classification bears on every other stage, including the strictness of evaluation and the trust threshold weighed in Stage 7. Purpose clarity is essential.

Stage 4: AI Harmony (Internal Consistency) Formerly: AI Harmony Checks. AI systems check whether the website is internally coherent: consistent structure, definitions, purpose, and schema across all pages. Pages must agree with each other conceptually and structurally. Chaotic, contradictory, or low-coherence domains are weighted down early.

Stage 5: AI Knowledge Corroboration (Corroboration with Established Knowledge) Formerly: AI Cross-Correlation. AI systems check whether the site's content is corroborated by established knowledge from authoritative sources, such as government databases, foundational references, scientific repositories and occupational frameworks. AI systems do not always weigh conflicting sources reliably.

Stage 6: AI Trust Building (Accumulating Evidence) AI systems gather evidence of reliability across stability, accuracy, consistency, neutrality, structural integrity, and purpose transparency. Trust (in industry terms, source reliability) is iterative, not binary.

Stage 7: AI Trust Acceptance (Eligibility for Use in Answers) Once trust signals meet the trust threshold, the domain is weighted as a reliable reference and becomes eligible for use in answer synthesis, citations, and multi-source reasoning. Trust alone does not make it visible to humans.

Stage 8: Candidate Surfacing (Shortlisting for Answers) In industry terms, retrieval. AI systems assemble a shortlist of viable contributors for a given informational need, weighing query relevance and competing sources. This is not final ranking.

Stage 9: Early Human Visibility (Early User Interaction Signals) Formerly: Early Human Visibility Testing. The domain is exposed to a small fraction of human-facing results -- AI answers and search results -- and its first user interaction signals build up: which results users choose, whether they quickly return to the results, and how they respond to answers that draw on the domain.

Stage 10: Baseline Human Visibility (First Stable Presence in Search and AI Answers) The domain holds placements in search results that do not fluctuate wildly and are no longer removed immediately or automatically, and, in AI answers, a share of appearances around a stable core of sources. This stage establishes the first reliable human traffic baseline.

Stage 11: Growth Visibility Formerly: Growth Visibility and Human Traffic Acceleration. If baseline performance is strong, AI systems expand visibility across regions, query families, device types, and tail depths.

Key Principles

• Stages 1-2 come first for any page; Stages 3-11 are weighed together as parallel dimensions of evidence.
• Each stage is reached as fast as the system gathers enough evidence: architectural quality sets how quickly evidence builds, and site purpose (non-commercial, commercial, hybrid) sets how much is needed, through the trust threshold weighed in Stage 7.
• Crawlability (access and fetching, Stage 1) does not equal visibility (Stages 9-11).
• AI search does not operate exactly like traditional search engine optimisation, nor does it replace it: for AI answers built on a search engine's index, that engine's own ranking factors apply.
• Framework versions are released as Zenodo DOI deposits under the concept DOI; this document reflects version 0.7.1.

Canonical Reference This Internet-Draft is NOT the canonical source. The framework is deposited at Zenodo. Citations are best made against the concept DOI, which always resolves to the latest deposited version: Concept DOI: Framework version 0.7.1, reflected in this document, is to be deposited under the same concept DOI. Where this document and an earlier deposited version differ, the later version governs. Companion specifications are deposited separately, each with its own concept DOI: ambiguity elimination and website visibility reporting . The eleven stage specifications corresponding to Sections 3.1 through 3.11 are deposited individually and indexed at the publisher's site listed below. The framework is developed collaboratively through the W3C AI Visibility Lifecycle Framework Community Group . The group was proposed on 10 February 2026 and launched on 24 February 2026 under a Community Group charter (version 1.0, 25 February 2026). It is chaired by the author of this document. The group operates under the W3C Community Contributor License Agreement. Its deliverables are not W3C Standards and are not on the W3C Standards Track.

Security Considerations This document describes an observational framework and does not define any protocols, data formats, or executable specifications. There are no security considerations directly applicable to this document.

IANA Considerations This document has no IANA actions.

References Normative References Informative References Google AI for Developers

Changes from -02

• Updated to framework version 0.7.1 (October 2026): the framework's specifications now cite the published research behind each evidenced statement; this document states what the framework contributes and what is provisional, what it does not claim, and how memory and retrieval relate to the lifecycle (Section 2.1).
• Stage names and subtitles: Stage 4 AI Harmony (formerly AI Harmony Checks); Stage 5 AI Knowledge Corroboration (formerly AI Cross-Correlation); Stage 6 AI Trust Building (Accumulating Evidence); Stage 9 Early Human Visibility (formerly Early Human Visibility Testing); Stage 10 Baseline Human Visibility; Stage 11 Growth Visibility (formerly Growth Visibility and Human Traffic Acceleration).
• The lifecycle is described as non-linear, in three related phases: Stages 1-2 come first; Stages 3-11 are weighed together. Each stage is reached as fast as the system gathers enough evidence: architectural quality sets how quickly it builds, and site purpose sets how much is needed.
• Figures are stated as illustrative examples (analytical estimates); descriptions of AI systems are stated as the framework's model, inferred from observed behaviour.
• The relation to search engine optimisation is corrected: for AI answers built on a search engine's index, that engine's own ranking factors apply.
• The canonical source notice states which framework version this document reflects.