Writing With AI Still Requires the Writer: Balancing Automation with Human Creativity

If you read my blogs you know how curious I am about artificial intelligence in the writing space, yet I’m not ready to hand it the keys to the classroom. That is probably the most honest way to say it. Call it Claude or call it ChatGPT, AI is here, and students, teachers, families, and schools are going to encounter it whether we are ready or not. At the same time, current AI writing tools are unreliable in important ways. They can sound confident when they are wrong, produce polished language without real understanding, and make weak thinking look stronger than it is. For writing instruction, that is a serious concern.

Of course AI can produce text; that was never really the question. The deeper issue is whether students are still doing the thinking that writing is supposed to develop, especially those with learning differences. Writing is not just getting words onto a page. Writing requires planning, decision-making, organization, evaluation, revision, and the ability to communicate meaning clearly to another person. If AI writing tools help students engage more deeply in that process, they may eventually have a useful instructional role. If it shortcuts the process, replaces student thinking, or hides misunderstanding behind smooth language, then we should be very careful.

The Current Research Is Interesting, But We Need to Be Careful

A recent article caught my attention because it takes a thoughtful, research-based approach to AI and writing without treating the technology as magic. In “Threshold concepts for writing with AI: Experimentation, expertise, agency”, Abram D. Anders and Emily Dux Speltz studied a university-level “AI and Writing” course with 38 undergraduate students from 22 majors. The authors identified three ideas that seemed important for productive AI use: writing with AI is an experimental process, writing with AI requires expertise and dialogue, and writing with AI should strengthen rhetorical agency.

That is useful language for higher education, but we have to be careful. This was not a study of elementary or middle school students. It does not prove that AI improves writing for younger learners. It does not show that students should use AI independently. And it does not settle the larger question of whether AI helps or harms writing development. What it does offer is a helpful way to think about the instructional problem.

For those of us working in K–12 writing instruction, the question is not, “Can we simplify this university-level work and hand it to children?” The question is, “What does this remind us about what younger writers still need?” To me, the answer is clear: students need strategies, teacher guidance, planning routines, self-monitoring, revision habits, and strong control over their own meaning. In other words, AI assistance does not make explicit writing instruction less important. If anything, it makes it more important.

What This Study Really Shows

What I like about this study is that it does not treat AI as a shortcut machine. The students in the course did not simply ask AI to write for them and call that learning. They had to experiment, reflect, revise their approach, and think about the relationship between their own expertise and the tool.

That matters because AI use often looks simple from the outside. A student types something in. The tool produces an answer. The page fills up. But the study points to something much more complicated: productive AI use requires the writer to make decisions before, during, and after the AI response.

That is an important distinction for educators. The question is not, “Can AI generate writing?” The answer is obviously yes. The better question is, “What kind of thinking did the student have to do to get there?”

If the student planned, questioned, checked, revised, and made decisions, there may be learning happening. If the student simply accepted fluent language from the tool, the writing may look stronger while the student’s actual development remains weak. That is the part schools cannot afford to miss.

AI Is Not One-and-Done

One of the strongest ideas from the article is that writing with AI should be treated as an experimental process, not a one-time exchange. That matters because many people approach AI like a vending machine: type a prompt, get a response, use the response, and move on. That may produce text, but it is not the same as writing.

For younger students, we would need to translate this concept into language they can actually use: try it, check it, and make it better. That sounds familiar because it is already part of the writing process. In Self-Regulated Strategy Development (SRSD), an evidence-based framework, students learn to plan, write, revise, evaluate, set goals, and monitor their progress. Those habits matter even more when AI enters the picture.

A student who cannot judge whether an AI response is accurate, appropriate, specific, or connected to the assignment is not being supported by AI. That student is being given a shortcut without the judgment needed to use it well. This is why I would be very cautious about any approach that lets young writers accept AI-generated language without questioning it.

The Tool Matters Less Than the Thinking Behind It

Teachers and students are not just hearing about “AI” in the abstract. They are hearing about specific tools like Claude and ChatGPT, and those names are becoming part of everyday conversations about writing. That makes it tempting to ask which tool is better. But for writing instruction, I think the more important question is not whether a student uses Claude or ChatGPT. The more important question is whether the student has enough strategy, guidance, and judgment to use any tool wisely.

A student can ask ChatGPT for a paragraph and get something that sounds polished in seconds. A student can ask Claude to revise a draft and receive smoother language almost immediately. A student can ask AI for ideas, outlines, examples, counterarguments, or feedback. All of that may look helpful at first. But none of it automatically means the student is becoming a stronger writer.

The instructional danger is that tools like ChatGPT and Claude can make the hard parts of writing less visible. If an AI tool supplies the idea, organizes the paragraph, improves the wording, and corrects the weaknesses before the student understands them, the final product may improve while the student’s writing ability stays almost exactly where it was. That is the central issue, not a small one.

So, I would not build classroom policy around the name of the tool. I would build it around the role of the writer. Whether students are using ChatGPT, another AI writing tool, or no AI at all, they still need to plan, explain, evaluate, revise, and make decisions. The tool should never become the writer. At best, it should become something the writer learns to question, direct, and control.

AI Needs the Student’s Thinking First

Another important idea from the article is that writing with AI requires expertise and dialogue. In plain language, AI needs the writer’s thinking first. That may be the biggest instructional issue for younger students.

If a student has not formed an idea, planned a response, or thought through the purpose of the writing, AI can fill the empty space very quickly, often without providing the necessary feedback for improvement. But that empty space is often where the learning is supposed to happen. The struggle to generate an idea, organize it, explain it, or revise it is not a flaw in writing instruction. It is part of how students become writers.

That does not mean we leave students stuck. Strong writing instruction gives students support without taking over the thinking. Teachers model. Students rehearse orally. Graphic organizers reduce cognitive load. Mentor texts show what strong writing can look like. Sentence frames help students begin. Feedback helps them revise. SRSD is built on this kind of scaffolded support. But the responsibility gradually moves back to the student.

AI becomes risky when it does not just support the student’s thinking but replaces it. A polished AI paragraph may look impressive, but that does not mean the student understands the content, owns the idea, or can reproduce the thinking independently. In fact, the polished quality may make the problem harder to see. Teachers can usually recognize a rough draft that needs development and feedback. It is harder to recognize a smooth draft that has bypassed the student’s actual learning.

The SRSD Question: Who Is Doing the Work?

This is where SRSD gives us a very practical way to think about AI. In strong SRSD instruction, the teacher does not remove the hard parts of writing. The teacher makes the hard parts visible, teaches students how to handle them, and gradually releases responsibility back to the student.

That same question should guide any future conversation about AI and writing: Who is doing the work?

Who is choosing the idea and deciding what belongs in the plan? Who is checking whether the evidence fits and the explanation is strong enough? Who is revising when the writing does not say what it needs to say?

If the answer is mostly “the student,” with support from the teacher and carefully limited tools, then we may be in safer territory. If the answer is mostly “the AI,” then we have a problem.

SRSD teaches students to become strategic writers. They learn that writing is not magic. It is a process they can understand, practice, monitor, and improve. AI should never take that away from them. At most, it should support the student in using the strategy more thoughtfully.

AI Should Support the Writer’s Voice, Not Replace It

The third idea from the article is that writing with AI should strengthen rhetorical agency. That phrase may be too abstract for younger students, but the classroom version is simple: students need to stay in charge of the meaning.

The writer chooses the idea. The writer decides the purpose. The writer checks whether the words match the message. The writer asks, “Does this still sound like me? Does this say what I meant? Did I make the important choices?” Those questions matter because writing belongs to the writer.

This is where SRSD has something important to offer. SRSD teaches writing as a strategic, self-regulated process. Students learn how to approach a writing task, use a plan, talk themselves through the hard parts, set goals, check their work, and build independence over time. Those are exactly the habits students will need in a world where tools can generate language instantly.

The more powerful the tool, the more important the strategy.

A Simple Classroom Test for AI

If schools eventually allow AI into writing instruction, I think we need a simple classroom test before using it with students.

Does this use of AI make the student think more, or think less?

That question cuts through a lot of noise. If AI helps a student compare two possible topic sentences, evaluate whether a reason is clear, generate revision questions, or check whether a paragraph matches the plan, it may support learning. The student still has to think, decide, and revise.

But if AI supplies the idea, builds the plan, writes the paragraph, improves the language, and fixes the weaknesses before the student understands them, then the tool has taken over too much. The student may end up with a better-looking product but a weaker writing process.

That distinction matters especially for students who already struggle with writing. These students do not need a tool that hides their thinking. They need instruction that makes their thinking stronger. They need models, routines, self-talk, feedback, practice, and opportunities to make real decisions as writers.

So my rule would be this: AI should not be allowed to do the thinking we are trying to teach.

Why This Belongs in the SRSD Conversation

I do not think the right answer is to pretend AI does not exist. Students are growing up in a world where AI writing tools will likely shape how people brainstorm, search, summarize, draft, revise, and communicate. Schools will have to respond to that reality.

But I also do not think the right answer is to treat AI as a writing teacher. The current tools are not trustworthy enough, and younger students are still developing the judgment needed to use them well. Some of the researchers I trust most are deeply skeptical of current AI tools in writing instruction, and I think that skepticism is healthy. We should not confuse curiosity with endorsement.

This is not an anti-AI argument; it is a pro-writing-development argument.

A responsible path forward would require careful translation for younger learners with constant feedback. AI-related writing instruction would need age-appropriate examples, clear student language, teacher-led routines, ethical boundaries, and much stronger research with children before anyone makes broad claims. Most importantly, any use of AI would have to protect the relationship between writing and thinking. Once that relationship breaks, the writing may look better on the surface, but the student may have learned less.

Curiosity With Caution

For now, my position is curiosity with caution. I want to understand what AI can do. I also want to protect what writing does for students.

If AI can help students think more clearly, plan more carefully, revise more thoughtfully, and stay in control of their meaning, then it may eventually have a place in carefully designed instruction. If AI does the thinking for them, we should name that honestly: it is not writing development.

The goal has never been simply better-looking text.

The goal is better writers.

Reference:

Anders, A. D., & Dux Speltz, E. (2026). Threshold concepts for writing with AI: Experimentation, expertise, agency. Computers and Composition, 81, Article 103008. https://doi.org/10.1016/j.compcom.2026.103008

---