I use AI for writing almost every day. That sentence would have made me slightly nervous a few years ago. Mostly because I am a writer.
I write content professionally. I write fiction. I have spent an unreasonable amount of time changing a sentence, changing it back and then deciding that the first version was better. This is apparently part of the profession. And now I work with something that can produce twenty alternatives before I have finished wondering whether I need a second coffee. I use AI for research, SEO analysis, structure, localisation and editing. I ask it to find weaknesses in arguments and questions I haven’t considered. Sometimes I ask it why a paragraph doesn’t work. What I rarely ask is the most obvious thing: write this for me.
Not because AI can’t. It can. And that is precisely what has made the question more interesting. If a machine can produce a perfectly acceptable paragraph in ten seconds, what exactly am I doing during the twenty minutes in which I write one myself?
For a while, I thought the answer was probably something reassuring about human creativity. I’m no longer sure. I think the more useful answer has something to do with decisions.
How I use AI for writing
AI-assisted writing can mean almost anything by now. Brainstorming with ChatGPT counts. So does using AI for research, creating an outline, rewriting a paragraph, correcting grammar, translating an article or asking a model to produce the whole thing. Two people can therefore say that they use AI for writing while describing almost opposite ways of working. I've written separately about how I use ChatGPT as a writer, particularly for research, structure and editing without letting it take over the writing itself.
My own use starts long before the finished sentences. In my professional work, AI has become particularly useful for research, SEO analysis, localisation, testing structures and reviewing existing copy. I have written elsewhere about how this has changed my work more broadly, because the interesting part isn’t simply that some tasks have become faster. The relationship between the task and the person doing it changes as well.
Research is an obvious example. AI is remarkably useful for entering a subject quickly. It can identify terminology, surface competing arguments, suggest questions and point towards studies or sources I might otherwise have missed. But I don’t consider an AI answer a source. If a fact matters, I still want to see where it came from. If a study supports an argument, I want to read what the study actually says rather than what a language model tells me it says. AI makes the beginning of research much faster. It doesn’t remove the end.
The same is true for ideas. I can ask for twenty article ideas about AI and writing and receive them almost immediately. Most will be perfectly reasonable. That is also the problem. Anyone can receive them. I find AI much more useful once I already have an observation, a question or something that irritates me. Then I can ask what I am missing, what the strongest objection would be or which assumption I am quietly making without noticing it. Instead of asking AI to have the idea, I use it to make the idea work harder.
This is probably where AI has become most useful to me as a writer. I don’t need it to agree with an argument. I need it to find the hole in one. If I argue that AI will make judgment more important because producing language becomes cheaper, there are obvious questions to ask. Why wouldn’t judgment eventually be automated as well? Am I redefining writing around precisely the parts of my profession I would prefer machines not to take? Is the distinction between producing and deciding actually as clear as I want it to be? Those questions improve the article even when none of the resulting sentences comes from AI.
Structure works in a similar way. AI can create an outline for almost anything. Introduction, five sensible headings, challenges, best practices, conclusion. Nothing is technically wrong. Sometimes nothing is particularly alive either. I prefer to know roughly what I want to say before asking AI whether the structure actually helps me say it. Have I buried the answer? Do two sections make the same point? Does the argument move somewhere, or have I simply arranged information under headings? For SEO content, there is another useful question: have I actually answered what somebody searching for this subject came to find out? That matters. Search intent should influence what an article answers. It doesn’t have to decide how the article sounds.
Where I become more careful
The closer AI gets to the actual prose, the more cautious I become. Not because AI-generated prose is necessarily bad. That argument is becoming increasingly difficult to make. Sometimes it is clean, clear and better than the sentence I had. Sometimes annoyingly so.
But a first draft is doing more than producing language. At least mine is. I often discover the thought while trying to write it. An argument that seemed convincing in my head becomes suspicious as soon as I have to explain it. An example changes the direction of a section. A sentence leads somewhere I hadn’t planned to go. Occasionally I realise that the clever idea I started with wasn’t particularly clever after all. This is inconvenient, but useful.
If AI writes the first draft, I start somewhere else. I begin with its decisions about structure, emphasis, examples and language. Of course I can change them. But revising an existing answer is different from finding one. This is why I don’t think the useful question is simply whether writers should let AI create first drafts. It depends on what the draft is for. If I need functional copy, variants or something deliberately rough to react against, a generated draft can save time. If writing the draft is part of discovering what I think, saving that time may also remove part of the work I actually needed to do.
Editing is different. Here I use AI a lot, although I try not to ask it to “improve” a text. Improve it how? Shorter, clearer, stranger, warmer, more precise, more accessible? Better for whom? “Improve this” gives the model permission to make a series of decisions that I haven’t bothered to define.
I get more from narrower questions. Where does this argument become unclear? Which paragraph repeats something I have already said? Where does the tone change? Which claim needs evidence? Is there a sentence that sounds generic compared with the rest? Often I explicitly ask for criticism rather than a rewrite. That leaves the writing problem with me.
This distinction becomes even more important when people talk about preserving their voice while using AI. Voice is usually treated as something sitting on the surface of a text: sentence length, vocabulary, rhythm, perhaps a fondness for certain expressions. Give a model enough examples and it can reproduce some of that surprisingly well. But I don’t think voice begins with wording. It begins with what someone notices. What they consider important. Which detail they keep and which one they remove. What they find funny. Where they become uncertain. What they refuse to simplify.
AI can imitate the surface of those decisions. I am more careful about letting it make them.
The decision behind the sentence
This is the part of AI writing that interests me more the longer I use it. Every time I delegate a task, I may also be delegating a decision.
If AI creates the outline, it may decide what deserves emphasis. If it writes the first draft, it establishes a default direction for the argument. If it rewrites every sentence, it can gradually standardise the voice. If it does the research and I never open the sources, I have effectively allowed it to decide what I consider true. None of this means I should never use AI for those tasks. I do. But I want to know what else I am handing over with them. I developed the Human Decision Map to make that question more concrete. It explores where AI contributes to the writing process and where the decisions that shape the final work still belong to the writer.
That is also why the distinction between content writing and creative writing matters to me. I use AI in both areas, but not in exactly the same way. In professional content, efficiency is part of the job. If AI helps me compare search results, organise research, test a structure or find an unclear passage faster, good. There is no special artistic virtue in spending three hours on something that could sensibly take thirty minutes.
Creative writing changes the calculation. I also write fiction without AI. A story such as Gott würfelt nicht (DE) matters in this context not because writing without AI is somehow purer, but because it reminds me what the process feels like when there is no model waiting to solve the next sentence. In fiction, the strange sentence I didn’t know I was going to write may be the point. The detour may be useful. The inefficient part may be where something happens.
This is why I don’t find the usual opposition between “AI writing” and “human writing” particularly helpful. The more interesting question is where human judgment sits inside an AI-assisted process. Recent research into human-AI writing increasingly examines precisely these questions of agency, ownership and control rather than treating AI use as a simple yes-or-no category. For me, the practical version is simpler: before I delegate part of the writing process, I try to ask what decision I am delegating with it.
The important question isn’t whether AI touched the text. It is what the human stopped deciding.
From AI-assisted writing to the AI Writer
This is also why I use the term AI Writer somewhat differently from the way it is often used online. “AI writer” frequently means the software itself: a tool that generates copy. I use it to describe the person.
An AI Writer is a human writer who deliberately integrates AI into the writing process while retaining authorial judgment, creative direction and responsibility for the final work.
AI-assisted writing describes the process. AI Writer describes the human role within it.
That role isn’t defined by personally typing every word. If it were, editing, collaboration and half the history of professional publishing would become rather awkward to explain. What matters more to me is whether the writer is still making consequential decisions. Where does the idea come from? Who decides what the text is trying to do? Who evaluates the generated possibilities? Who rejects the plausible but boring answer? Who checks the claims? Who notices that something is technically correct but doesn’t belong in this particular text? And, finally, whose name is attached to it?
The last point matters. AI can produce language without having to stand behind it. I can’t. If I publish something under my name, “the model suggested it” is not a particularly convincing correction policy.
That may become one of the more important changes AI brings to writing. As producing acceptable language becomes easier, the value of the writer may shift towards observation, judgment, selection, direction and responsibility. Not because producing sentences suddenly becomes irrelevant, and not because machines are incapable of good prose, it’s mostly because generation itself is no longer scarce. This is also why I find the question of whether AI will replace writers more complicated than it first appears. Replacing parts of the writing process isn't necessarily the same as replacing the writer.
This doesn’t mean I think everyone who opens ChatGPT should start calling themselves an AI Writer. Using a calculator doesn’t make someone a mathematician either. The term is useful to me only if it describes a writer first: someone who has something to say, knows enough about the craft to judge what comes back and remains responsible for the work after AI has participated in making it.
When AI writing becomes bigger than writing
For me, this question became stranger when AI moved beyond text. I created Kuh Alma as a written character. Later I began experimenting with generative AI to see what happens when a character moves from words into images, music and video. I have written separately about that process and what generative AI may change for storytelling, because at that point the technology isn’t merely helping with a paragraph anymore. It can start producing possibilities for the world around it. I have written separately about that process and what generative AI may change for storytelling, because at that point the technology isn’t merely helping with a paragraph anymore.
An AI model can generate a new setting for Alma, another version of her appearance, a scene I never wrote or a direction I never intended. Technically, all of these things can exist within seconds. That doesn’t mean they belong to her world.
Someone still has to decide.
That is why I have started using another term, author-led storyworld, for a creative world in which an identifiable human author retains final creative authority over its canon, direction and meaning, even when AI contributes substantially to its development, expansion or production. I will explore that idea in more detail elsewhere, because it goes beyond AI for writing. But it grows from the same problem.
Generative AI gives writers an extraordinary amount of production capacity. Text, images, sound, video and eventually forms we haven’t found useful names for yet. The obvious question is what we can make with all of it.
I suspect the harder question will be deciding what should belong.
And perhaps that is the thread running through my entire AI writing process. I use AI because it is fast, useful and increasingly good at things that used to take me much longer. I don’t feel a particular need to make research slower for sentimental reasons. I don’t need spiritual growth while comparing headline variants. If a model finds the weak point in an argument in thirty seconds, excellent.
But efficiency is a strange measure for writing as a whole. Writing is sometimes production. Sometimes it is communication. Sometimes it is thinking. And sometimes it is spending twenty minutes on a paragraph that a machine could have produced in ten seconds.
The machine may even produce the better paragraph.
The question is whether producing the paragraph was the only thing I was doing.
I don’t think it was.
-
I think the most useful approach is to give AI specific jobs rather than asking it to take over the entire writing process. I use it to question arguments, examine structure, identify unclear passages and suggest alternatives. But I remain careful about letting it decide what matters in a text. Voice isn't just vocabulary or sentence rhythm. It also comes from what a writer notices, questions and chooses to leave out.
-
It can be. AI can help writers explore ideas, test possibilities, examine story structures or find problems in a draft. But creative writing isn't always about reaching a finished text as efficiently as possible. Sometimes the unexpected sentence, the detour or the difficulty of finding an idea is part of the creative process. I find AI useful for exploring possibilities, but I'm more cautious about letting it make the creative decisions.
-
That depends on what the draft is supposed to accomplish. For functional content, alternative versions or something rough to react to, an AI-generated draft can save time. But when I'm writing to discover what I actually think, I prefer to start myself. A first draft doesn't just put existing ideas into sentences. Sometimes it reveals that the idea I started with wasn't particularly good. I'd rather discover that before asking a machine to make it sound convincing.
About the Author
Carsten Jan Weichelt is a professional content writer and strategist who works with AI in research, SEO, content development and editing. He also writes fiction and experiments with generative AI in creative storytelling, including his project Kuh Alma.
His work explores how technology changes writing, creativity and the decisions behind both. He is particularly interested in what happens when AI makes producing content easier, but deciding what is worth producing remains a human responsibility.