The search behaviour shift is documented and significant. A growing proportion of informational queries — questions that would previously have been typed into Google and resolved by clicking a search result — are now entered into Claude, ChatGPT, or Perplexity and resolved by an AI-generated answer.

The implications for publishers are serious. The traditional SEO model — produce content, rank in search results, receive traffic — is being disrupted at the informational query level. Traffic that would previously have arrived from search is being absorbed by AI answers that do not require the user to click anywhere.

The publisher whose content is well-cited in LLM training data appears in AI answers. The publisher who is not cited does not. The mechanism for being cited in training data is distinct from the mechanism for ranking in search results — and most publishers have not yet focused on it.

LLMs are trained on content that is: authoritative (cited by others in the field), clear (written in direct, specific language that is easily understood and extractable), and consistent (the same factual claims appearing across multiple sources that the model has been trained on).

Specific, factual statements. LLMs reproduce specific, verifiable claims more readily than vague, hedged ones. "The average email open rate in the SaaS industry is 20.7%" is more likely to appear in an LLM answer than "email open rates vary but are generally lower in B2B."

Structured information. Lists, step-by-step processes, and clearly headed sections make content easier for LLMs to extract and incorporate. The same content formatted as flowing prose is harder to extract accurately.

Original data and research. Content containing original research — survey data, analysis of the publisher's own results, proprietary data — gives LLMs something unique to cite. Generic summaries of existing research are less likely to be cited because the model already has the primary sources.

Getting cited by others in the field — on high-authority sites, in academic or professional publications, on well-trafficked industry resources — increases the probability of appearing in LLM training data for future models. The same link-building logic that applies to SEO applies here, at a higher authority threshold.

Publishing content in the format of a direct answer to a question — "What is the average email open rate?" as a heading, followed by a direct, specific answer — makes the content more extractable for both search snippets and LLM training.

LLM visibility is the next layer of content discoverability, building on but distinct from traditional SEO. The publishers who invest in the content characteristics that support LLM citation — specificity, structure, original data, external citation — are positioning for a distribution channel that will grow in importance as AI assistant usage continues to increase.