LLM (Large Language Model)
A type of AI trained on vast amounts of text data to understand and generate human-like language (e.g., GPT-4, Claude 3).
What is LLM?
A large language model is a neural network trained on a very large corpus of text to predict the next token in a sequence. That simple objective, applied at scale, produces a system that can answer questions, summarise, translate, write code and hold a conversation. Models such as OpenAI's GPT series, Anthropic's Claude, Google's Gemini, Meta's Llama and xAI's Grok are all LLMs, and they power the answer engines that GEO is concerned with.
Two properties of LLMs shape AI search. First, a model's built-in knowledge is frozen at its training cutoff; anything after that date is unknown unless the system retrieves it. Second, models do not store facts as a lookup table, so what they "know" about a brand is a fuzzy impression assembled from every mention in the training data. Consistent, widely repeated facts survive that process; rare or contradictory ones do not.
Answer engines wrap an LLM in additional machinery: a retrieval step that searches the web, a ranking step that picks which pages to read, and instructions that tell the model to cite its sources. So a brand's visibility depends on both what the underlying model learned in training and what the retrieval layer fetches at answer time.
Why it matters for AI search
Understanding how an LLM forms its impression of a brand explains most GEO tactics. Training-time visibility rewards consistent facts repeated across many sources; answer-time visibility rewards crawlable, well-structured pages the retrieval layer can find. Knowing which crawlers feed which layer (GPTBot, CCBot and Google-Extended for training; OAI-SearchBot, ChatGPT-User, PerplexityBot and ClaudeBot for retrieval) lets teams make deliberate robots.txt decisions instead of blocking everything or nothing.
Related terms
RAG (Retrieval-Augmented Generation)
A technique that allows LLMs to pull in real-time information from external sources (like a website) before generating an answer.
Hallucination
When an AI model generates factually incorrect information confidently. Monitoring hallucinations is critical for brand safety.
Prompt Engineering
The practice of crafting specific inputs (prompts) to get the most accurate and useful output from an AI model.
GEO (Generative Engine Optimization)
Similar to AEO, the technical and strategic process of improving visibility within generative AI models.
Training Data
The text data used to train AI models. Being in training data affects how AI understands your brand.
Frequently asked questions
What is a training cutoff?+
The date after which a model has no built-in knowledge. Anything newer must be supplied through retrieval at answer time. This is why recent product launches or rebrands can be invisible in plain chat but visible in search-enabled modes.
Do all AI search engines use the same LLM?+
No. ChatGPT uses OpenAI models, Gemini and AI Overviews use Google models, Copilot uses OpenAI models with Microsoft's retrieval, Claude uses Anthropic models, and Perplexity routes across several. Their retrieval sources and citation behaviour also differ, so visibility has to be checked per platform.