What is an LLM?
An LLM is an AI model that understands and generates language; the abbreviation stands for Large Language Model. An LLM is trained on massive amounts of text and calculates the most likely continuation of an input, word by word. That is how it answers questions, summarizes texts, and translates. LLMs are the technology behind ChatGPT, Google Gemini, and Claude, and they write the answers of AI search.
How does an LLM work?
An LLM breaks text into tokens, small word and character units, and calculates from the text so far which token is most likely to follow. The model repeats this prediction until a complete answer has formed. The technical foundation is the transformer architecture, introduced by Google researchers in 2017: its attention mechanism weighs which parts of the input matter most for the next word.
A large language model acquires its knowledge in two phases:
- Pretraining: The model processes billions of text pages from the web, from books, and from databases and learns the statistical patterns of language from them.
- Fine-tuning: Human feedback trains the model to follow instructions and answer helpfully (Reinforcement Learning from Human Feedback).
The size of an LLM is measured in parameters, the adjustable weights of the neural network. Modern models reach hundreds of billions of parameters; that is exactly what the “large” in the name refers to.
Which LLMs exist?
The best-known LLM families come from five providers:
- GPT (OpenAI): The model series behind ChatGPT and ChatGPT search.
- Gemini (Google): Powers the Gemini app and writes the answers of AI Overviews and AI Mode in Google Search.
- Claude (Anthropic): Model family with a focus on long contexts and safe behavior.
- Llama (Meta): Models with open weights that companies can run themselves.
- Mistral (Mistral AI): European provider from France, also offering open models.
What do LLMs have to do with SEO and AI search?
LLMs write a growing share of all search answers. AI Overviews, AI Mode, ChatGPT search, and Perplexity answer queries directly, and in each of these systems a large language model writes the answer text. Your content reaches these answers in two ways:
- Through training data: What an AI crawler collected during training, the model can later reproduce as knowledge.
- Through live retrieval: AI search systems pull current web pages into the answer at runtime, a technique called Retrieval-Augmented Generation (RAG). Only this path produces source links and, with them, visitors.
Preparing content so that LLMs understand and cite it is a discipline of its own: Generative Engine Optimization (GEO), also called LLMO. Citable definitions, clear entities, and machine-readable facts are the core of our optimization for AI visibility.
What are the limits of an LLM?
An LLM does not know truth, only probabilities. Three limits follow from that:
- Hallucinations: The model writes fluently even when the facts are missing, inventing plausible-sounding claims along the way.
- Knowledge cutoff: An LLM’s knowledge ends at the cutoff date of its training data. Anything later, the model only knows through a connected search.
- Missing sources: On its own, an LLM cannot prove where a statement comes from. Only the connection to a search index delivers verifiable source links.
Exactly these limits make well-sourced web content valuable: AI search systems check their answers against current, trustworthy pages and cite the sources that state facts most clearly.
Frequently asked questions about LLMs
Is ChatGPT an LLM?
Not quite: ChatGPT is the product, the underlying LLM is a model from OpenAI’s GPT series. ChatGPT adds a chat interface, a web search, and further tools on top of the language model. What that means for your visibility is covered in our article on ChatGPT SEO.
What does the “large” in Large Language Model mean?
The “large” refers to the size of the model and its training data: modern LLMs have hundreds of billions of parameters and were trained on text collections spanning trillions of tokens.
What is the difference between an LLM and AI?
AI is the umbrella term for all systems that perform human-like cognitive tasks. An LLM is a specific type of AI that specializes in language. Image recognition or recommendation algorithms are AI, but not LLMs.
Can an LLM access my website?
Not directly. Content reaches an LLM in two ways: AI crawlers collect texts for training, and AI search systems fetch pages at runtime through a connected web search. You control both paths via robots.txt.
