AI search combines language generation with methods developed over decades of search. Understanding that history helps explain why crawling, indexing, links, and clear page content still matter.
Directories, then links
Early web directories organized sites into categories maintained by editors. Crawler-based search engines collected pages automatically and matched queries against indexed text. As the web grew, search engines needed ways to assess relevance beyond the presence of a word.
Brin and Page’s 1998 paper described a search engine that used link information alongside text. PageRank calculated page importance from the web’s link structure. The paper helps explain why links became central to SEO. Backlinks and AI Search examines their relationship with AI visibility."The Anatomy of a Large-Scale Hypertextual Web Search Engine", the original Stanford paper describing Google and PageRank.
Answering before LLMs
Search engines began answering some questions directly before generative AI became common. Google’s universal search combined web results with other formats. Featured snippets displayed extracts from pages. A person could sometimes obtain an answer without opening the source.
Google introduced the Knowledge Graph in 2012. Its announcement described more than 500 million entities and 3.5 billion facts and relationships. Entity information helps distinguish different subjects with the same name. This is separate from proving that a particular entity record affects a generated recommendation.Google's 2012 announcement, "Introducing the Knowledge Graph: things, not strings", with the entity and fact counts as stated at launch.
Google also developed methods for interpreting queries. Its BERT announcement in 2019 described using language context to improve understanding, initially for about one in ten English searches in the US. These developments extended search beyond literal keyword matching.Google's 2019 BERT announcement, which states the one-in-ten figure and the focus on longer, conversational queries.
By the time conversational AI products reached a broad audience, search already included direct answers, entity information, and learned language representations.
The generative turn
ChatGPT launched in November 2022. It gave users a conversational interface for asking questions and requesting generated text. Its initial launch should be distinguished from the later addition of web search.
Google added generated answers to Search through SGE and then AI Overviews. ChatGPT added search in 2024. Other assistants also introduced web retrieval and citations. The resulting products combine search and generation in different ways. See AI Search Products and Their Sources.OpenAI's announcement, "Introducing ChatGPT search", describing the third-party search providers and publisher partnerships behind it.
What this history changes in practice
Existing SEO skills remain relevant wherever an AI product retrieves pages through a search index. An accessible, useful page can support both ordinary search results and a generated answer. Neither indexing nor a strong search position guarantees citation.
The additional task is to measure the generated answer itself. Separate what the model produces without retrieval from what it produces after receiving sources. Model Memory and Live Retrieval explains how to make that comparison.