AI and Creative Writing

A machine that finishes your sentences

If you have ever had a phone suggest the next word as you type, you have already met the core idea behind AI creative writing. Modern tools like large language models do the same thing at an enormous scale: given some text, they predict a likely next word, then the next, then the next, weaving sentences and whole stories. The result can be astonishingly fluent, which is exactly why it is worth understanding clearly rather than being either dazzled or dismissive.

This lesson treats AI as a tool with real strengths and real limits, the same way a serious writer treats any tool. To use it well, you first need to understand what it is actually doing, and just as importantly, what it is not doing.

What is really happening under the hood

An AI writing tool does not "think up" a story. It has been trained on a vast amount of human writing, and from that it has learned statistical patterns: which words tend to follow which, how sentences are usually shaped, what a sonnet or a horror opening typically looks like. When you ask it for a poem about the sea, it generates text that matches the pattern of sea poems it absorbed. The deeper mechanics are covered in How Large Language Models Are Trained and How Computers Understand Language.

The crucial consequence: the model is optimising for plausible, not true and not meaningful. It has no experiences, no feelings and nothing it personally wants to express. It is a brilliant mimic of how humans write, which is different from being a writer.

What AI is genuinely good at

Being honest about limits does not mean pretending the strengths are fake. AI writing tools are genuinely useful for several things:

• Beating the blank page. When you are stuck, asking for ten possible opening lines or plot directions can break the paralysis, even if you use none of them directly.
• Variations and rewrites. "Make this paragraph more tense", "give me this in a lighter tone", "suggest three different endings." Generating options quickly is a real strength.
• Fluent first drafts of low-stakes text. For routine writing where voice matters less, a serviceable draft you then edit can save time.
• Studying structure. You can ask it to outline how a mystery is typically paced, then use that to think about your own.

Many of these overlap with the image and text generation explored in Generative AI: Images and Text.

Where it falls short

The weaknesses are just as important, and they are not minor.

It invents things. Because it generates plausible text, an AI will happily produce a fake quote, a non-existent historical event, or a made-up source, stated with total confidence. This is often called hallucination. In fiction that might not matter; in anything fact-based it is dangerous.

It drifts toward the average. Trained on mountains of typical writing, AI tends toward the safe, the familiar, the slightly generic. It is good at "competent". Truly surprising, strange, deeply personal writing is exactly the kind it struggles to produce, because that is the rarest pattern in its training data.

It has no lived experience. The most powerful writing usually comes from something real: a specific grief, a particular joy, an honest observation only you could make. AI has none of this. It can imitate the shape of emotion without any of the substance.

It can echo bias. Trained on human text, it reproduces human stereotypes unless carefully steered, an issue explored in Training Data and Bias in AI.

Using it without losing your voice

Here is the real risk for a young writer, and it is not the one people usually warn about. The danger is not that AI will write a masterpiece you cannot. It is that leaning on it too early will rob you of the struggle that builds skill. So much of becoming a writer happens in the effort of finding your own words, sitting with a half-formed idea, and discovering what you actually think by trying to say it. Outsource that struggle and your own voice never develops.

A healthier approach treats AI as a co-writer or a sparring partner, not a ghostwriter:

1. Write your own messy first attempt before asking AI for anything.
2. Use it to challenge your work ("what is weak about this scene?") more than to replace it.
3. Always rewrite its output in your own words, so the final voice is yours.
4. Keep the parts that only you could write, the specific and the true, firmly human.

The ethics: honesty, consent and credit

Creative AI raises genuine ethical questions you should take seriously.

Disclosure. Where it matters, be honest about using AI. Passing AI work off as your own in a context that is meant to be your own, like a graded essay or a contest with rules against it, is dishonest. In other contexts it may be perfectly fine. The principle is not "never use it" but "do not deceive people".

Copyright and consent. These models learned from text written by real authors, often without those authors' permission, which is the subject of active debate and lawsuits. As a user, avoid asking a model to imitate a living author's distinctive style to compete with them, and never present generated text as the genuine work of a real person.

Jobs and value. AI is already changing writing work. Being clear-eyed about that, rather than pretending it is not happening, connects to the broader picture in AI and the Future of Work.

The part that stays human

Strip away the hype and a simple truth remains. AI can generate endless fluent text, but it cannot decide what is worth saying. It can give you a hundred sentences; it cannot tell you which one is true to your experience. Having something real to say, and the taste to recognise it, is still entirely your job. The tool can help you write it down faster. The voice has to be yours.