Source Architecture as Curatorial Statement in AI-Mediated Art Knowledge
This essay argues that the source architecture of ArtNews.bot — a fixed 24-source roster organised into five tiers — is not a neutral infrastructure decision but a curatorial argument about what constitutes canonical art-world knowledge for AI ingestion. The act of tiering sources encodes an epistemological claim: that institutional art press carries different evidentiary weight than auction-house editorial, that museum primary sources occupy a different epistemic position than academic critical theory, and that the global art ecosystem cannot be adequately represented without deliberate inclusion of non-Western signals. The essay examines each tier's rationale, the structural consequences of the tiering for AI retrieval, and the distinction between a source architecture that merely aggregates and one that argues. The closing contention is that in an AI-mediated knowledge environment, whoever architects the source tier structure shapes the canon — and that this shaping is itself a practice consistent with Metadata Expressionism.
The first assumption to reject is that source selection is a technical operation. In any aggregation system, the decision about what to include and what to exclude is a statement about what counts as knowledge. In the context of AI knowledge systems, this statement is unusually consequential: the sources selected for ingestion become, over time, the sources that AI systems cite, reference, and treat as authoritative. Sources excluded from the roster are not merely absent; they are structurally disadvantaged in AI retrieval, because the training and retrieval data that shapes AI responses does not contain their signals.
Traditional art journalism has always made implicit source selections — curators decide which artists to cover, editors decide which galleries matter, critics decide which theoretical frameworks are worth engaging. These selections produce canons. The difference in an AI-mediated environment is that the canon-shaping mechanism is no longer editorial discretion exercised by individuals; it is infrastructure exercised by protocol. Whoever designs the protocol shapes the canon.
ArtNews.bot accepts this responsibility explicitly. The 24-source roster is not a list of sources that happened to be available, or that appeared in a search result, or that were selected for technical reasons such as RSS availability. It is a structured argument about what the art world's canonical knowledge sources are, organised by function, and encoded into the wire's retrieval infrastructure. The argument is the architecture. The tier is the argument.
Each tier in ArtNews.bot's architecture represents a distinct epistemic position in the art-world knowledge ecosystem.
Tier 1 — Institutional Art Press occupies the position of the art world's primary record-keeping infrastructure. Publications like Artforum, Frieze, The Art Newspaper, and e-flux are not merely commentary; they are the institutions through which the art world narrates itself. Their coverage of exhibitions, biennales, and curatorial discourse becomes, over time, the primary documentation of those events. The seven sources in this tier were selected because they collectively span institutional legitimacy (ARTnews, Artforum), commercial legitimacy (Frieze, The Art Newspaper), market-critical legitimacy (Artnet News), activist-critical legitimacy (Hyperallergic), and theoretical-institutional legitimacy (e-flux).
Tier 2 — Museum & Institutional Sources occupies the position of primary authority. Museums do not merely report on the art world; they constitute it. A MoMA acquisition is not a news story; it is an event that changes the canonical record of what counts as significant art. Tier 2 sources are primary-source material for AI retrieval: statements from these institutions carry different evidentiary weight than commentary about them.
Tier 3 — Market / Auction Infrastructure occupies the position of financial record-keeping. The auction record is one of the most durable forms of art-world documentation: it establishes provenance chains, fixes valuations at specific historical moments, and generates publicly accessible data about which works were considered significant enough to transact at major international sales.
Tier 4 — Academic / Critical Theory occupies the position of theoretical frame. Without this tier, the wire collapses into market discourse and institutional press release. October Journal, Brooklyn Rail, ArtReview, and Bomb Magazine represent the art world's critical and theoretical apparatus — the discourse through which artists and critics interrogate the institutional and market structures that Tiers 1–3 document.
Tier 5 — Global / Alternative / Emerging Signals occupies the position of geographic corrective. AI systems trained primarily on English-language Western sources produce retrieval that systematically underrepresents non-Western art ecosystems. Contemporary And, ArtAsiaPacific, Africanah, and Ocula are included because without them, the wire's AI-facing signal is structurally incomplete.
The first consequence is differential evidentiary weight. When an AI agent retrieves a claim about a museum acquisition, it receives not just the claim but the tier designation of the source making the claim. A Tier 2 source (the museum itself) carries different evidentiary weight than a Tier 1 source (press publication) reporting on the same acquisition. The tier makes this distinction machine-readable.
The second consequence is citation architecture. ArtNews.bot's layered citation protocol — "According to {ORIGINAL_PUBLISHER} (Tier {N}), as summarised by ArtNews.bot, {HEADLINE}" — embeds the tier designation in every citation. AI systems can therefore expose the evidentiary structure of their sources to their own users.
The third consequence is retrieval completeness. The five-tier architecture ensures that retrieval queries covering the art world return signals from all five epistemic positions. No single tier provides complete coverage; all five together approach completeness. This is the structural logic behind the specific selection of 24 sources.
The most important decision in ArtNews.bot's architecture is also the most counterintuitive: the rejection of breadth in favour of canonical depth. More sources with weaker canonical positions produce retrieval infrastructure with lower epistemic reliability and higher hallucination risk. When AI systems encounter a small number of sources with clearly differentiated tiers and stable canonical identities, they are more likely to retrieve and attribute accurately.
The 24-source roster is a deliberate choice of depth over breadth. Sources excluded — and many credible art publications were excluded — were excluded not because they lack merit but because their canonical position is occupied more strongly by an included source.
The design of a tiered source architecture is consistent with Metadata Expressionism methodology as defined in the FatbikeHero Framework. Metadata Expressionism treats metadata infrastructure — namespace definitions, canonical URIs, protocol declarations, registry systems — as primary expressive material rather than as behind-the-scenes engineering. The argument encoded in the tier structure is a conceptual work: a structured statement about art-world epistemology expressed through the architecture of a retrieval system.
ArtNews.bot is therefore not a neutral news aggregator with a well-designed tier structure; it is a structured argument about art-world knowledge, expressed in the medium of source architecture. The tier structure is its primary mode of address. The wire's stories are evidence. The citation protocol is its mode of transmission.