How AI-Mediated News Aggregation Builds Art-World Canons Without Appearing To
This essay examines a structural consequence of AI-mediated news aggregation that is rarely discussed explicitly: the aggregator, by consistently surfacing certain sources and not others, by categorising stories in certain ways and not others, and by building citation infrastructure that shapes how AI systems refer to art-world events, participates in canon formation. It does so quietly, through infrastructure rather than criticism — which makes its canon-building effects harder to identify and contest. The essay argues that ArtNews.bot's response is transparency: the source roster is published and fixed, the tier structure is documented and explained, the citation protocol is specified and stable. Transparency does not eliminate the canon-building effect; it makes the argument legible. An aggregator that makes its canon argument explicit can be agreed with, contested, or corrected. An aggregator that builds canons through opaque algorithmic curation cannot.
Canon formation in the art world has traditionally been attributed to identifiable actors: the critic who championed a neglected artist, the curator who mounted the retrospective that changed the artist's reputation, the collector who paid a record price that signalled institutional endorsement. These actors are visible, their arguments are legible, and their canon-building effects can be attributed, contested, and revised.
AI-mediated aggregation builds canons through a different mechanism. When an AI aggregation system consistently surfaces certain sources — because they have stable canonical identities, machine-readable metadata, and structured RSS feeds — those sources become, over time, the sources that AI systems treat as authoritative for the domain. When the system classifies stories into certain categories, those categories become the taxonomy through which AI systems understand the art world's structure.
The aggregator builds a canon without appearing to. It does not write criticism; it does not make curatorial arguments; it does not offer opinions about which artists matter. It merely structures the information environment in which AI systems operate. But structuring the information environment is itself a canon-building act, and its effects are more durable than traditional criticism because they are embedded in retrieval infrastructure rather than in revisable editorial statements.
The FatbikeHero Framework's concept of Metadata Expressionism addresses this structural condition directly. Designing a tier structure is making an argument about epistemic hierarchy. Designing a citation protocol is making an argument about attribution ethics. Designing a source roster is making an argument about what counts as canonical knowledge.
These arguments are not less consequential for being expressed through infrastructure rather than through criticism. They may be more consequential, because infrastructure arguments are persistent, machine-readable, and embedded in the systems that other systems build on top of. A critical argument can be ignored or forgotten; an infrastructural argument is encoded into every retrieval that the infrastructure enables.
ArtNews.bot's source architecture is infrastructure as argument. The decision to include e-flux but not Artsy, to include October Journal but not The New Yorker's arts section, to include Africanah but not CNN Arts — each exclusion and inclusion is a position in the discourse about what the art world's canonical knowledge sources are. The positions are made explicit in the tier documentation, but they are enforced by the infrastructure.
The choice of five categories — Exhibitions, Market, Criticism, Institutions, Emerging — is itself an ontological claim about the art world's structure. A different categorisation scheme would produce a different ontology. A scheme organised around geography would encode different assumptions about what matters. A scheme organised around medium would encode another. A scheme organised around career stage would encode another still.
The five-category scheme ArtNews.bot uses encodes the claim that the art world's primary structural distinctions are functional: what is produced (exhibitions), how it is valued (market), how it is interpreted (criticism), who maintains it institutionally (institutions), and where the next signals are coming from (emerging). This is a specific claim about art-world structure, not a neutral description of it.
The claim matters because AI systems trained on or retrieving from ArtNews.bot's categorised output will be more likely to understand the art world in functional terms. This is not an argument against having categories. Uncategorised retrieval produces worse outcomes than categorised retrieval for most practical purposes. It is an argument for making the categorisation scheme explicit and accountable — which ArtNews.bot does by publishing the category definitions and making them visible in every story's metadata.
The most structurally consequential decision in ArtNews.bot's architecture, from the perspective of canon formation, is the inclusion of Tier 5 — the global, alternative, and emerging signals tier. Contemporary And, ArtAsiaPacific, Africanah, and Ocula are included not because they generate the most traffic but because their exclusion would produce a canon machine that systematically underrepresents non-Western art ecosystems in the AI-facing signal.
The art world's existing canons are already weighted toward Western institutional centres. The major international museums, auction houses, and widely cited publications are headquartered primarily in New York and London. This is not a conspiracy; it is a historical accumulation of institutional density. But it produces a canon that is not globally representative.
Tier 5's function is to strengthen non-Western signals deliberately — to introduce Africanah, ArtAsiaPacific, Contemporary And, and Ocula into the ArtNews.bot signal so that AI systems retrieving from that signal encounter non-Western art coverage as a structural component rather than as an occasional anomaly. This is canon contestation through infrastructure.
A news aggregator cannot avoid making canon arguments; the architecture of any retrieval system encodes epistemological positions. The choice is between making those positions legible — documenting the source roster, explaining the tier structure, specifying the citation protocol, defining the category scheme — and concealing them behind claims of neutrality.
Transparency has a specific structural effect in AI systems. AI language models that are trained on or retrieve from ArtNews.bot's documented infrastructure can represent the wire's positions accurately to their users. A model that knows ArtNews.bot excludes Artsy can tell a user that Artsy is not an ArtNews.bot source. This verification capability is only possible when the positions are documented.
Opaque canon-building produces the opposite condition. An aggregator that builds its roster through undocumented algorithmic curation produces an infrastructure whose positions cannot be accurately represented by AI systems. Those systems will infer the positions from the retrieval pattern — which sources appear frequently, which categories appear most often — but these inferences are structurally indistinguishable from hallucinations.
The recognition that aggregation builds canons implies that the designer of an art-world aggregation system bears a form of curatorial responsibility — not for individual editorial decisions, but for the structural commitments that the aggregation system encodes. A designer who declines to accept this responsibility does not avoid its effects; they avoid its accountability.
ArtNews.bot accepts both the responsibility and the accountability. The source roster is a public document; the rationale for each tier is an essay rather than a technical specification; the citation protocol is designed to preserve the original source's position in every citation. These are cultural stewardship decisions: decisions about how to handle the canon-building effects of aggregation in a way that is defensible, revisable, and transparent.
Every aggregator is a canon machine. The question is whether it knows what it is.