A large language model (LLM) is a neural network trained on a very large body of text to predict the next piece of text in a sequence. That simple objective, at scale, produces a system that can answer questions, summarize, write and reason over text. GPT models behind ChatGPT, Claude, Gemini and Llama are all LLMs. In search, LLMs are the part of an AI answer engine that reads sources and writes the response.
How LLMs work, briefly
Modern LLMs use the transformer architecture, introduced by Google researchers in 2017. Text is split into tokens, and the model learns statistical patterns across them during training. After pre-training, models are usually fine-tuned with human feedback and other methods so they follow instructions and answer helpfully.
An LLM's built-in knowledge is fixed at its training cutoff and is not a database. It can be wrong, out of date or confidently mistaken, which is why AI search products pair LLMs with live retrieval.
Two ways an LLM can know about your brand
The first is training data: if your company was written about widely before the model's cutoff, the model may have learned facts and associations about it. You cannot edit this directly, and it changes only when a new model is trained.
The second is retrieval: when an AI tool searches the web at answer time and feeds pages into the model. This is the path you can influence quickly, through crawlable, clear and current pages and through third-party coverage that search can find.
What this means for SaaS marketing
Buyers use LLM-powered assistants to research categories, build shortlists and compare vendors. To be included, your product needs to be well represented in both paths: discussed accurately across the web over time, and easy to retrieve for specific questions now. Checking how the major assistants describe your product today is a sensible starting point, because they may repeat outdated pricing or positioning.
Example
A fictional rebrand
Formsy, a made-up form builder, repositioned last year as a workflow automation tool. Assistants answering from training data still call it a form builder. When web search is used, answers are more current because they retrieve the new homepage. The team updates its review site profiles and directory listings so that retrieved third-party sources match the new positioning too.
Common mistakes
Treating an LLM like a search index
Without retrieval, an LLM generates plausible text from patterns. It does not look up your site.
Testing in one tool only
Different LLMs and products give different answers. Check the assistants your buyers actually use.
Questions
Not the trained model directly. You can update your own pages and third-party sources, which retrieval-based answers can use, and which may feed future training.
AI search tools use LLMs to write answers, combined with search and ranking systems to find sources.
