LLM SEO: optimizing content for LLMs.

There are two paths, and optimising means serving both. The first is training: everything the model learned about you from the web it was built on. You cannot edit a trained model, but you shape what the next one learns by being described clearly and consistently across the sources it will read.

The second is retrieval. When an engine answers a live query, a retrieval system, often the same kind of vector search behind any RAG pipeline, fetches candidate passages, ranks them, and hands the best to the model to answer from and cite. This is the path you influence most directly, and it rewards content built to be retrieved.

Because retrieval works on passages, not whole pages, LLM SEO is partly an information-architecture problem:

• Self-contained passages. Each key answer complete in one place, so a retrieved chunk makes sense without the rest of the page around it.
• Clear semantics and structure. Honest headings, question-and-answer blocks and schema, so both the retriever and the model parse meaning correctly.
• Named sources and specifics. Figures, dates and citations. Adding sources and statistics measurably lifts how often engines quote a passage.
• A clear entity. One consistent identity the model can bind facts to, so a retrieved passage is confidently attributed to you.
• Freshness and crawlability. If a retriever cannot fetch or trust the page, none of the above is reachable.

These are the same foundations behind generative engine optimization services and answer engine optimization, expressed in the language of how the models work.