What is generative AI, in plain English?

Generative AI is the technology that produces things. Not predictions, not decisions, not recommendations, but actual content: a paragraph of writing, a photograph, a piece of music, a synthetic voice, a short video. You give it a prompt and it creates something that did not exist before.

That distinction matters because most AI before 2022 was focused on classification and prediction. It was good at looking at something that already existed and making a decision about it. A spam filter decides whether an email is junk. A credit scoring algorithm decides whether a loan application is likely to default. A recommendation engine decides which video to show you next. All useful, but none of it creates anything new. The output is always a decision: yes or no, likely or unlikely, this category or that one.

Generative AI does something different. Instead of classifying inputs, it produces outputs. The scale of what it can produce has expanded rapidly. In 2022 it was mostly text and images. By 2024 it had extended convincingly into audio and video. By 2026 the distinction between AI-generated content and human-produced content has become genuinely difficult to spot in many contexts.

The reason 2022 became such a turning point was the public release of ChatGPT in late November. The underlying technology was not entirely new; large language models had existed in research contexts for years. But ChatGPT put a conversational interface on top of that technology and made it accessible to anyone with a browser and an email address. Within two months it had over 100 million users. That rate of adoption was faster than any consumer technology in history. It forced the rest of the industry to respond.

Large language models are at the heart of text-based generative AI. The basic idea is that these systems have been trained on enormous amounts of written text, from books to websites to scientific papers, and have learned to predict what word or phrase should come next given what has already been written. That sounds simple, but when done at sufficient scale the results are remarkable. The model develops something that looks a great deal like understanding, though researchers argue at length about whether it constitutes understanding in any meaningful sense. What it definitely develops is fluency and breadth: the ability to produce coherent, contextually appropriate text across an enormous range of topics.

Images work on a related principle but through different mechanisms. Diffusion models, which power tools like Midjourney and Adobe Firefly, work by learning to remove noise from images until they arrive at a coherent picture. You give the model a text description and it generates an image that matches. The quality achievable in 2026 would have seemed impossible in 2020. A professional-quality photograph from a twelve-word prompt is now routine.

Where generative AI gets genuinely complicated is at the question of what it is actually doing when it produces these outputs. The honest answer is that it is doing a sophisticated form of pattern completion. It has seen so much human-produced content that it has internalised the patterns that make text feel coherent, images look realistic, music sound structured. It is not thinking in the way a human thinks. It does not have intent or curiosity or understanding of what it is making. But the outputs it produces can be extraordinarily useful regardless.

This is where the earlier AI versus generative AI distinction becomes practically important. Older predictive AI was reliable within its domain but brittle outside it. A model trained to detect tumours in X-rays would fail completely if you asked it to do anything else. Generative AI models are flexible in a way that earlier systems simply were not. A large language model can draft a legal letter, explain a concept to a ten-year-old, write code in Python, translate text, summarise a long document, and suggest birthday present ideas, all within the same session. That general-purpose flexibility is what makes it genuinely useful rather than just impressive.

The creative and business implications of that flexibility are still working themselves out. Generative AI has already changed how marketing copy gets written, how code gets drafted, how customer service gets delivered, and how content gets produced. It is changing more slowly but no less certainly in professional services, education, and healthcare. The pattern is consistent: tasks that involve producing text or analysing documents or communicating information are the ones getting affected first.

For most people the relevant question is not how it works under the hood but what it can do for them. At its most basic level it is a tool that converts a description of what you want into something close to what you asked for, immediately, at no marginal cost per use. That is a genuinely new capability. No professional writer, no stock photo library, no voice actor, no video production company works that way. The economics and the speed of content production have changed permanently.

The risks that come with it are equally real. Generative AI can produce convincing misinformation at scale. It can be used to impersonate real people. It produces output that sounds authoritative even when it is factually wrong. The same fluency that makes it useful also makes its mistakes harder to spot. None of these risks are reasons to avoid the technology, but they are reasons to understand how it works and to bring appropriate scepticism to what it produces.

Understanding that generative AI is fundamentally a pattern-completion system that produces plausible outputs rather than verified facts changes how you should use it. It is at its best when the outputs are things you can check and when the creative flexibility is the point. It is at its worst when you treat its outputs as reliable facts about the world without verification.

The technology is moving fast enough that some of what is written about it today will be dated within a year. But the underlying principle, that this is a system for producing content rather than making decisions about content that already exists, is unlikely to change. That distinction is worth holding onto.