Large Language Model
A language model at sufficient scale — billions of parameters, trained on web-scale text — to exhibit emergent capabilities like reasoning, instruction following, and open-ended generation. Gemini, GPT, Claude, and Llama are all LLMs.
What a large language model is
A large language model (LLM) is a language model trained at a scale large enough that qualitatively new capabilities emerge: coherent long-form reasoning, instruction following, translation between languages, code generation, and open-ended question answering. There is no precise threshold, but modern LLMs typically have billions of parameters and are trained on trillions of tokens of text from the open web, books, code, and other sources.
The models that power AI search and AI assistants — Google Gemini, OpenAI GPT, Anthropic Claude, Meta Llama — are all LLMs. So are the models underlying AI Mode, AI Overviews, and every product discussed on this site.
Why scale changes things
Smaller language models can classify and retrieve well. LLMs can reason, synthesise, and generate. The difference is emergent: capabilities that do not appear at smaller scales appear discontinuously as models cross certain parameter and data thresholds. This makes LLMs useful for tasks no smaller model handles well — writing coherent answers to complex questions, following nuanced instructions, understanding implicit intent — and also makes their internal behaviour harder to predict or audit.
Architecture
Almost all modern LLMs are decoder-only transformer architectures (GPT style) trained with a next-token prediction objective. They may use Mixture of Experts layers to increase effective capacity without proportionally increasing compute per token, or multimodal extensions to process images and audio alongside text.
AI SEO implications
LLMs are now the interface layer between search queries and answers. Grounding gives them access to fresh web content, but their primary bias — what they believe before retrieval — shapes which sources they select and how they represent brands. Understanding LLM behaviour, not just search algorithm behaviour, is the central challenge of modern AI SEO.
Related concepts
Mentioned in
-
Google Search changes almost half of Gemini's brand recommendations
We asked Gemini 4 Hydrogen 10,701 brand questions from memory and again with Google Search. Search supplied 45% of the brands in each answer, and in 1 in 5 of its reasoning summaries the model wrote about a "mental check" or "simulated search" while searching live.
-
Gemini writes better answers when it knows the user is a recognised expert
Tell Gemini you're a known expert and its answers get better. Across 15 topics, blind-rated answers scored 7.8 out of 10 for experts and 5.8 with no name.
-
Gemini 4 Argon - One Step Closer to 'Model as an Employee' Paradigm
Google's Gemini 4 Argon introduces a 1-million-token output limit to support long-horizon autonomous tasks, enterprise benchmarks, and complex code migrations.
-
AI Influence With Confidence Threshold Manipulation
A study reveals that language model abstention is governed by a single internal confidence signal that outranks question difficulty and fact retrievability.
-
How did the 1955 plan for AI go? Surprisingly well!
This page compares the 1955 Dartmouth artificial intelligence proposal with the public record of AI capabilities and world models as of 2026.
-
Relax. Google won't penalize all AI generated content.
Google evaluates content based on substance and user behavior signals rather than penalizing pages simply because AI generation is detected.
-
Ox Alpha is GLM
Prompt injection and gzip-NCD compression analysis reveal that OX Alpha, a mysterious LLM on OpenRouter, is GLM developed by Z.ai.
-
AI Models Encode Brand Data but Fail to Recall a Third of It
Google Research found frontier AI models encode 95-98% of brand facts in parametric memory but fail to recall 26-34% of them. The bottleneck is retrieval, not storage.
-
Introducing eCommerce Optimization Engine
The eCommerce Optimization Engine uses Google's multimodal embedding model to align product text with images to improve search retrieval and SEO performance.
-
Why deep learning works.
An excerpt from François Chollet’s Deep Learning with Python exploring the manifold hypothesis and how structured information enables deep learning to work.
