An LLM is a large language model trained on huge amounts of data. It predicts the next tokens from context, which lets it generate, summarize, translate or analyze text.
What is a large language model?
During training, the model learns statistical relationships in the data and stores them in its parameters. When it answers, it builds the output step by step from the prompt and the context so far.
Modern assistants can add search, file handling, calculations or other tools to the model. So ChatGPT, Claude or Gemini are whole product systems, with the model as one part.
Why a model is sometimes confidently wrong
A sentence that sounds likely doesn't have to be true. A model can join unrelated facts, make up a source or use outdated information.
RAG, web search and good data reduce the risk. They don't remove it. Legal, medical, financial and other important claims need a check against the primary source.
What it means for marketing and SEO
LLMs speed up research, first drafts, data analysis and automation. The process needs clear inputs, permissions, human review and an owner of the final decision.
They also add a new distribution layer. A brand has to track whether AI systems mention it correctly, but the foundation stays the same: an accessible site, a consistent entity and verifiable content.