Thoughts on Using AI During Drafting

Writing is thinking made visible, some say. For me, writing is thinking made possible. I’m not alone in saying I write to find out what I think. That’s why I’ve spent my entire life writing day in and day out. To think, I write.

From my first interaction with GPT until today, I have consistently concluded that GPT cannot write. I have no problem conceding that AI can simulate texts, but it’s beyond reason in my mind to say it can write. This isn’t a pose or an argument. It’s a judgment I can’t refute.

I do, however, use AI all the time to draft. Sometimes I hand write a lengthy prompt describing the ideas and the specifics I want to discuss. Then I revise it as a type it in and ask the LLM to draft an example for a particular audience with a particular tone. I’ll critique the simulated text and discuss it with the bot. I’ll decide at that point if I really want to write about it.

Experimenting in this fashion has made me a better writer, I think. I’m much more sensitive to performative voice, and I’ve tried out genres I never considered before. But I still don’t consider the machine to be a writer.

When I consider AI's role in drafting in the writing classroom, I wrestle with this fundamental question: does AI drafting enhance or undermine the cognitive processes that make us better writers and thinkers? So far it seems the answer depends entirely on timing, context, and developmental readiness.

The Cognitive Cost of Skipping the Struggle

The blank page isn't an obstacle to get around. It’s a crucible. Research in writing cognition suggests that facing that emptiness engages what researchers call "generative processing": retrieving relevant knowledge, making novel connections, and organizing inchoate thoughts into linear expression (Bereiter & Scardamalia, 1987). This struggle isn't a bug in the writing process; it's the feature that builds intellectual capacity.

When we draft, several critical processes occur. We discover what we actually think as opposed to what we thought we thought. We identify gaps in our understanding when we can't express ideas clearly. We find unexpected connections as we search for the right words.

AI drafting risks bypassing this generative struggle, providing ready-made expression for thoughts we haven't fully formed. If the writer hasn’t fully formed the ideas expressed in the text, the text was not written at all but simulated.

The Embodied Nature of Writing

Neuroscience research reveals writing as an embodied cognitive process. The physical act of forming words creates neural pathways linking motor control, language processing, and conceptual thinking (James & Engelhardt, 2012). When we write, we're not just recording thoughts; we're creating them through the act of expression.

Having spent the past two years or so pondering my habit of drafting in spiral bound journals, shuttling between reading other writers, doing searches with Perplexity to find other sources to read, and using a bot to do things like “give me ten word pairs that might fill these two blanks in this sentence” etc., I’ve learned that the original ideas become the core of my own writing enhanced by resources.

AI drafting severs this embodied connection. I’ve seen it over and over again. For example, a while ago I used AI to write emails for me to send to music venues seeking gigs for my music duo. My problem was I had no idea of what to say. Because I started this task with minimalist prompts, I got back pretty bad text.

Soon enough, I found that I needed to draft rough language pointing to ideas that had to be on each email. I used this original mini draft along with information from venue websites to craft simulated emails tailored to each venue.

When we offload the task of drafting, we become editors rather than creators of our own expression—the cognitive equivalent of watching someone else practicing a musical instrument and expecting to build technique.

The Problem of Premature Coherence

AI produces unnaturally coherent first drafts, but human thinking is naturally messy, recursive, full of false starts and tangled ideas. The struggle to create coherence from this chaos is where learning happens (Sommers, 1980). When AI provides instant coherence, it masks the actual state of our understanding.

Students revising AI drafts might produce polished final products while never developing the ability to create coherence from their own confused thoughts. They learn to recognize good writing without learning to produce it.

Coherence is essentially a function of thought, not of language. In a very real sense, finding a through line in one’s disorganized thinking is the heart of drafting. Writers need to let coherence mature, which is done more efficiently and effectively through drafting than through offloading.

The Syntax-Thought Connection

Linguistic research demonstrates that our syntactic structures shape our thinking capacity (Vygotsky, 1962). As we learn to write complex sentences, we develop the ability to think complex thoughts. The struggle to subordinate clauses, balance parallel structures, and manage logical flow isn't just about grammar—it's about developing a cognitive architecture.

When AI provides sophisticated syntactic structures, students might operate temporarily beyond their developmental level. Remove the AI assistance, and their expression reverts to their actual capability level. What are the learning gains from creating tinsel coherence artificially?

The Developmental Threshold

Despite these concerns, AI use during drafting can serve learning when used appropriately. The key lies in recognizing it as a tool that amplifies existing abilities rather than replacing foundational development. Writers should demonstrate certain capabilities before AI drafting becomes beneficial:

1. Established Voice: Can produce original text with consistent personal style.

2. Structural Understanding: Knows how arguments and narratives are constructed.

3. Revision Skills: Can identify weaknesses in their own writing. (Can read a rough draft.)

4. Metacognitive Awareness: Understands their own writing process.

5. Genre Knowledge: Familiar with conventions they're working within.

Beyond these matters, writers are ready to experiment with AI during drafting when they are aware of the role of the reader and can adjust expression to accommodate the reader’s needs. Without these foundations, AI drafting may mask deficits rather than enhance abilities.

When AI Drafting Enhances Learning

For experienced writers who've developed foundational skills, AI can provide valuable assistance:

Breaking Through Blocks

AI can provide momentum when stuck, generating alternative openings, suggesting structural approaches, or offering phrase variations. The key is using AI to overcome specific obstacles, not avoid the writing process entirely.

Exploring Possibilities

AI can expand a writer's range by demonstrating unfamiliar genre conventions, modeling different stylistic approaches, or suggesting organizational patterns outside the writer's usual habits. This works when writers can evaluate options critically and adapt them to their purposes.

Supporting Access Needs

For writers with specific challenges—learning characters, language barriers, or executive function difficulties—AI might remove barriers while preserving cognitive work. The goal is accessibility without sacrificing development.

Appropriate Educational Contexts

Advanced workshops might use AI in drafting to do case studies in different voices and styles ot for rapid prototyping of experimental forms.. In technical writing, where clarity and convention matter more than personal voice, AI can efficiently create documentation templates or user manuals.

Research contexts benefit from AI assistance in organizing and connecting established information, creating literature review structures, or developing initial frameworks for analysis.

*****

AI drafting becomes appropriate when it serves as a lever for existing ability rather than a substitute for developing ability. The timing matters: too early, and it prevents essential skill development; too late, and it misses opportunities for enhancement.

Writing development requires what Elbow (1973) calls finding your "sound on the page"—a distinctive voice developed through thousands of micro-decisions. AI drafting interrupts this when used prematurely but can enrich it when writers have already established their voice.

Certain human capacities can only be developed through direct, effortful practice. The question isn't whether AI can produce good drafts—it clearly can (well, adequate). The question is whether the struggle to produce our own imperfect drafts is an irreplaceable part of becoming fully developed writers. I think it is for what it’s worth.

AI drafting can either support learning or prevent it, depending on timing and context. The key lies in recognizing when we're ready. That recognition requires wisdom from educators and honesty from writers—qualities no AI can provide.

REFERENCES

Bereiter, C., & Scardamalia, M. (1987). The psychology of written composition. Lawrence Erlbaum Associates. https://eclass.uoa.gr/modules/document/file.php/PHS122/%CE%91%CF%81%CE%B8%CF%81%CE%B1/Bereiter_Scardamalia_Chapter1.pdf

Elbow, P. (1973). Writing without teachers. Oxford University Press. https://talkcurriculum.files.wordpress.com/2014/09/elbow-p-1973-writing-without-teachers-pp-12e28093751.pdf

James, K. H., & Engelhardt, L. (2012). The effects of handwriting experience on functional brain development in pre-literate children. Trends in Neuroscience and Education, 1(1), 32-42. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4274624/

Sommers, N. (1980). Revision strategies of student writers and experienced adult writers. College Composition and Communication, 31(4), 378-388. https://psychology.yale.edu/sites/default/files/sommers_onrevisionstrategies.pdf

Vygotsky, L. S. (1962). Thought and language. MIT Press.​​​​​​​​​​​​​​​​ https://img3.reoveme.com/m/bac6393f496a1d08.pdf