What "authority" means to a language model
In traditional SEO, authority is a proxy metric derived from inbound link patterns — PageRank and its descendants. In GEO, authority means something deeper: whether the language model has encountered your brand, your claims, and your expertise across enough sources, with enough consistency, that it treats you as a reliable reference. This kind of authority is built over time across the whole web, not just on your own domain. It is the reason that domain authority — the accumulated signal of how much of the web references you — outweighs schema markup as a citation driver by a significant margin.
Cold recall: your brand in the model's training data
Cold recall is the ability of an AI model to correctly identify and describe your brand when asked, drawing only from its training data — with no live web retrieval involved. It is a test of whether you exist in the model's world knowledge. A brand with strong cold recall will be mentioned and cited even in contexts where the model is not actively searching; a brand with weak cold recall may be excluded from answers even when it is the most relevant option. Building cold recall means building a presence across authoritative, widely-crawled sources: Wikipedia and Wikidata, industry publications, news coverage, analyst reports, and reference databases.
Wikipedia and Wikidata: the knowledge graph anchor
Wikipedia and Wikidata occupy a special position in AI training pipelines. Both are extensively crawled, regularly updated, and treated as high-authority structured references. A Wikipedia article about your brand is one of the most reliable ways to establish a knowledge graph entity — a node in the machine-readable model of the world that AI systems draw on. Wikidata, the structured sibling, enables sameAs linking: connecting your brand's Wikipedia entity to your domain, your social profiles, and other reference databases so AI systems can unambiguously resolve which entity they are talking about. Entity disambiguation — the model knowing which "Apple" or "Nike" you are — is foundational to consistent citation.
The compounding effect and why to start early
Cold recall and entity authority compound slowly. Training data has a lag of 12 to 24 months between when content is published and when a new model includes it. Wikidata and Wikipedia entries take time to accumulate the edit history and reference network that makes them authoritative. Brand mentions in high-authority publications take time to build. This is not a reason to delay — it is a reason to start immediately. The brands building cold recall and entity presence today will have a compounding advantage when the next model generation trains. The brands that wait for a short-term return will start 12 months behind.