How Artificial Intelligence Works

A large language model begins with training. During training, the model processes enormous collections of text — books, articles, websites, and other publicly available material — and adjusts billions of internal parameters so that it can estimate which words are likely to follow others in a given context. The result is not a database of facts that can be looked up. It is a statistical model of language, and the knowledge it appears to hold is stored as patterns in those parameters rather than as retrievable records.

When a question arrives, the model performs inference: it processes the prompt and generates a response one token at a time, each choice influenced by everything that came before it. A token is a small unit of text — often part of a word — and the model assigns probabilities to candidate tokens, selecting the sequence that best fits the patterns it learned. That is why AI answers can read fluently while still being wrong: fluency and factual accuracy are different things, and a model can produce a confident, well-formed statement that no source supports.

For businesses, this matters because an AI-generated answer about a company is shaped by two different things: what the model learned during training, and what it retrieves at the moment of answering. Training data may be months or years old, may contain errors, and may confuse one company with another that shares a name or a similar description. The retrieval step — covered in How AI Retrieves Information — is how newer, external information can enter the answer. Understanding both steps is essential to understanding why a business’s public information sometimes appears incomplete, outdated, or mixed up in AI answers, and what kind of work can make it clearer. Nothing in this section, or anywhere on this site, guarantees that any particular information will appear in an AI-generated answer.

What this section covers

Questions this section answers

Readers who work through this section will be able to answer questions such as: Where does an AI answer come from? Why can two platforms give different answers to the same question? Why might an AI system describe a business inaccurately even when the business’s own website is correct? And what is the difference between information a model learned in training and information it retrieved just now? The companion section on retrieval takes up that last question in detail, and Understanding AI Citations explains how the sources behind an answer are chosen and shown.

How it connects to the other sections

This section is the foundation of the Knowledge Center. How AI Retrieves Information builds directly on it, explaining how retrieval-augmented generation supplements a model’s training with external sources. Entities, Knowledge Graphs and Semantic SEO explains how machines form a structured picture of a business, which is part of what makes retrieved information usable. Building AI Visibility and Authority then turns to the practical question of how businesses can make their public information clearer and more referenceable, and AI Visibility Measurement and Research explains how that work can be tracked over time. The full overview is on the Knowledge Center page.

What comes next

This introduction is the starting point, not the complete guide. Articles in this section will explore each topic above in greater depth, with references to primary documentation and published research where it is available. If you notice an error here, or can point to a source that improves an explanation, please use our contact and corrections form.

Latest in this pillar