Generative AI is artificial intelligence that produces new content, such as text, images, audio or code, in response to a request, instead of only sorting or scoring content that already exists. Ask it to summarise a contract and it writes a summary; ask an older system the same thing and, at best, it labels a document you already have. It is one branch of artificial intelligence, and what sets it apart is that it makes something that was not there before.

How it works, without the maths

Most of today's text systems are large language models built on an architecture called the transformer, introduced in a 2017 research paper titled "Attention Is All You Need". You do not need the internals, only the idea behind them.

A large language model generates text by repeatedly predicting the next piece of writing, given what came before. Type "the capital of France is" and it answers "Paris" because that continuation is overwhelmingly likely in the text it learned from. Do that one small step at a time, at enormous scale, and the same mechanism can hold a conversation, rewrite a paragraph or produce working code. Generators for images, audio and video are also generative AI, but they are built on different methods, such as diffusion models, so generative AI is a family of approaches rather than a single architecture.

Foundation models, and why one tool does many jobs

A decade ago, each task tended to need its own model: one for translation, another for spam, another for sentiment. The shift that made generative AI feel general is that a single large model, trained once on broad data, can be pointed at many different tasks without being rebuilt. Researchers at Stanford named these "foundation models" in a 2021 report, because so much gets built on top of one base.

In practice, that is why the same assistant can draft a job ad, explain a spreadsheet formula and translate an email. One model, many uses, which is also why choosing between ChatGPT, Claude and Gemini is a real decision rather than a cosmetic one.

Generative AI versus predictive AI

The older, quieter kind of AI classifies or predicts. A fraud model answers "is this transaction suspicious". A credit model answers "how likely is this person to repay". Both sort the world that already exists into categories or scores.

Generative AI answers a different kind of request: "write the customer email explaining the hold". Same company, same data, but one system makes a decision and the other produces a draft. Knowing which kind you are dealing with tells you how to use it and how it can fail. A misclassification and an invented fact are very different problems.

Where it helps in a company

Generative AI pays off fastest on work that starts with a blank page or a long document, and the same patterns hold up from one team to the next:

  • Drafting. First versions of emails, job descriptions, proposals and reports, which a person then edits.
  • Summarising. Turning a long thread, meeting or contract into the few lines that matter.
  • Search over your own documents. Asking a question and getting an answer drawn from internal files, instead of scrolling.
  • Translation and tone. Moving text between languages or adjusting it for a different audience.
  • Code. First-draft scripts and functions that a developer reviews, which also underpins AI agents.

In each case the output is a starting point someone improves, never a finished product that ships unchecked.

The catch

Because a language model predicts plausible text rather than verified text, it can produce a confident answer that is simply wrong, including invented facts, quotes and sources. That failure has a name, the hallucination, and it is the single most important limit to understand before you rely on the output.

There is a second catch that is organisational rather than technical. Depending on the tool and its settings, what you type may be stored or used to improve the service, so sensitive data needs clear rules about which tools it can go into. This is a question of AI governance, and it is far easier to settle before adoption spreads than after.

Frequently asked questions

What is the difference between generative AI and AI?
Generative AI is one branch of artificial intelligence. AI is the broad field of systems that infer how to produce an output. Generative AI is the part that creates new content, such as text, images, audio or code, on request; older, predictive AI only labels, scores or ranks content that already exists.

Is ChatGPT generative AI?
Yes. ChatGPT is built on a large language model, which is a type of generative AI. Claude and Gemini are the same kind of system from other providers. All three generate text one piece at a time based on the request you give them.

Can generative AI be wrong?
Yes, and often with full confidence. A generative model predicts plausible content, not verified facts, so it can invent a statistic, a quote or a source that does not exist. This failure is called a hallucination, and it is why every important output needs a human check.

Sources