What is a Large Language Model?
Before you can optimise for AI, you need to understand what you're optimising for. Large Language Models are the technology behind ChatGPT, Claude, Gemini, and every AI assistant your customers are using.
In this guide
- What LLMs are and how they work
- Why they matter for your brand
- Key concepts: tokens, context, and parameters
- The difference between models
The Simple Explanation
A Large Language Model (LLM) is a type of artificial intelligence that has been trained on massive amounts of text to understand and generate human language. Think of it as a very sophisticated autocomplete system, one that has read billions of web pages, books, and documents.
When you ask ChatGPT a question, it's not searching the internet (usually). Instead, it's predicting what words should come next based on patterns it learned during training. This is a crucial distinction that affects how you should think about AI optimisation.
LLMs predict the most likely response based on training data patterns
Key Concepts You Need to Know
Tokens
LLMs don't read words. They read tokens. A token is roughly 4 characters or 3/4 of a word. "ChatGPT" might be 2-3 tokens. Why does this matter? Because LLMs have token limits on how much they can process at once, and your content needs to efficiently convey information within these limits.
Context Window
The context window is the maximum amount of text an LLM can consider at once. GPT-4 can handle about 128,000 tokens (~96,000 words). Claude can handle up to 200,000 tokens. This affects how much of your content an AI can "see" when generating a response.
Parameters
When you hear "GPT-4 has over 1 trillion parameters," think of parameters as the model's "knowledge connections." More parameters generally mean the model can understand more nuanced patterns, but also requires more computing power.
Training Data
LLMs learn from the text they're trained on. If your brand is well-represented in high-quality content across the web, the model is more likely to have learned about you. This is the foundation of AI optimisation.
Crucially, training data is a snapshot in time. It's collected once and then remains fixed, often for a year or more until the next model version. If your brand wasn't visible in authoritative sources when that snapshot was taken, the AI simply won't know about you. You need to be in the game before the whistle blows.
Why This Matters for Your Brand
Understanding LLMs reveals why traditional SEO isn't enough anymore:
- LLMs have memory. Unlike search engines that crawl and index in real-time, LLMs "remember" what they learned during training. Getting into training data matters.
- LLMs synthesize information. They don't just link to sources. They generate answers by combining knowledge. Your brand needs to be part of that knowledge.
- LLMs have preferences. Clear, well-structured content is easier for LLMs to understand and reference accurately.
Key Takeaway
The key insight: LLMs don't search. They recall and generate.
Your content needs to be present in training data, structured for easy comprehension, and associated with clear entity relationships. This is fundamentally different from optimising for search engine rankings.
Technical Implementation
The way your content is technically structured affects how well LLMs understand it. Proper schema markup helps AI associate your brand with the right concepts.
Structured Data for AIThe Major LLMs You Should Know
| Model | Provider | Knowledge Cutoff | Used In |
|---|---|---|---|
| GPT-4.1 / GPT-4o | OpenAI | Jun 2024 | ChatGPT, Microsoft Copilot |
| Claude 4 Sonnet/Opus | Anthropic | Mar 2025 | Claude.ai, Amazon Bedrock |
| Gemini 2.5 Pro | Jan 2025 | AI Overviews, Gemini App | |
| Sonar / Sonar Pro | Perplexity | Real-time | Perplexity.ai |
| Mistral Large 3 | Mistral AI | ~2025 | Le Chat, various apps |
| DeepSeek-V3 / R1 | DeepSeek | Dec 2024 | DeepSeek Chat |
| Llama 4 | Meta | Aug 2024 | Various applications |
LLMs in Search vs LLMs doing Search
There's an important distinction between how LLMs appear in your customers' experience:
LLMs Powering Search Results
Google's AI Overviews use Gemini to summarise search results. The LLM synthesises what Google has already indexed, so your SEO directly impacts what appears.
LLMs With Web Search Tools
ChatGPT, Claude, and Perplexity can actively search the web when answering questions. The LLM decides when to search and which results to cite, combining training knowledge with live data.
Why this matters: For AI Overviews, traditional SEO is key. For LLMs with search, both training data presence and SEO matter, along with the LLM's judgment on which sources to trust.
We cover each of these in detail in the AI Models section, including their specific capabilities and optimisation considerations.
What's Next
Now that you understand what LLMs are, the next step is understanding how they learn. This will help you grasp why training data matters and how your content can become part of an AI's knowledge base.