Large language model

What it is

An LLM is a transformer based model with many parameters, trained first on broad text to learn language patterns and then tuned with instruction data and human feedback so it responds helpfully. It works in tokens, holds a limited context window, and can be extended with retrieval, tools and function calling. Different models vary in size, speed, cost and how well they follow instructions.

Why it matters

LLMs are the layer that now reads and rewrites your content before a person sees it, whether inside an AI answer, a chat assistant or a research agent. Visibility depends on whether a model can find your pages, parse them cleanly and quote them accurately, which is a different job from ranking a blue link. Understanding how the model handles context also explains common failure modes such as confident errors and outdated claims.

How it works

Practitioners use LLMs through APIs or interfaces, shaping behaviour with prompts, system instructions, examples and retrieval of trusted source documents. Marketing teams use them for drafting, clustering queries, classifying content, extracting entities and testing how assistants describe their brand. Content teams optimise for them by writing clear, extractable answers, keeping facts on the page and maintaining consistent entity information across the web.

When it applies

It applies any time content is consumed or produced through an AI assistant, AI search feature or automated content pipeline, which now covers most discovery and research journeys.

Examples

• A team prompts an LLM with a product page and asks what a buyer would still not know, then fills those gaps.
• A B2B brand tests several assistants monthly to see which competitors get named for its core category question.
• An agency uses an LLM to cluster twelve months of search queries into intent groups before rebuilding a content hub.

How it is measured

• Share of assistant answers that mention or cite your brand for target prompts
• Citation accuracy, measured as correctly attributed claims versus misquotes
• Crawl and fetch success rates for AI user agents on key pages
• Referral sessions and assisted conversions from AI assistant sources