Definition
LLMO (large language model optimization) is increasing the odds that large language models — ChatGPT, Gemini, Claude, Perplexity — know, correctly describe and recommend your brand when someone asks a question in your category. Where SEO wants a page higher in a list, LLMO wants the model to hold you in memory: when it talks about your subject, you are part of the conversation.
A language model knows a brand through two routes. The first is training data: everything the model read about you during training, on your own site and above all on other people's sites. The second is retrieval: for a question the assistant searches the web live and uses those pages as sources. The first route decides whether the model names you unprompted; the second whether you get cited. LLMO works on both but puts the emphasis on the first.
In practice LLMO overlaps almost entirely with GEO. People who say GEO usually think of citations in answers with sources; people who say LLMO think of what the model knows without searching. The tactics are the same: an unambiguous brand description, factual pages about what you do and for whom, mentions on sites that models trust, and content that answers the question directly. Which word you use is mostly a matter of who is speaking.
What demonstrably helps: using the same name, category and core facts everywhere; an about page that says in plain sentences what you are; mentions in comparisons, reviews, industry lists and press; and pages that give one clear answer per question. What does not help: hidden text, instructions to the model in your HTML or mass-generated pages without substance. Models filter that out, and it costs you trust with Google.
You measure with prompt sampling: a fixed set of questions you put to each model regularly, after which you count how often you are mentioned, whether the description is accurate and which competitors appear next to you. Because models ship in versions, your visibility changes in jumps; look at the trend per model rather than at one measurement.