4.3: Reasons not to let AI brainstorm, write, revise, or edit (in college)

Most people struggle with writing, and most people wish there were an easier or faster way to write. It can be very tempting to turn some of the work over to AI. It can seem like AI is leveling the playing field by giving students who have a hard time a chance to boost their grades. But what seems to help might hurt in the long run. 

Often, writing is hard because doing it involves hard thinking, and that is where we learn. (See AI can generate decent-sounding text. Do we still need to learn to write?)

Some writing teachers I respect are experimenting with allowing AI assistance with brainstorms, drafts, and revisions, resulting in hybrid, part-AI writing. The Elon University Student Guide to AI suggests, “Think of the writing process as a human-AI loop. You remain in charge of the writing but use AI at certain stages of the process, working back and forth to create the final paper.” 

Certainly, some learning and thinking can happen if we let AI come up with ideas and words and then add or make changes. AI outputs can stimulate thinking as well as replacing it. But in my view, on balance, letting AI do some of the writing takes away chances for students to develop voice, confidence, and critical thinking. 

Brainstorming, organizing, writing, and revising in college are essential. Yes, “AI is here,” but that doesn’t mean we have to use it for everything. Below, I’ve explained why I ask students to use AI in the writing process only for feedback that leaves it to them to make changes to their drafts.

Note: I leave it to learning disability specialists to identify cases where specific uses of generative AI in the writing process should be considered accommodations. For students with writing disabilities, some kinds of friction in the writing process may be more destructive and limiting than useful and necessary. 

Thinking by brainstorming and outlining

In brainstorming and outlining, we wrestle with ideas. If AI supplies ideas and organizes them, we miss chances to develop mental muscle and intellectual confidence. 

Imagine you are starting on a paper about ways to prevent teen social media addiction. You’ll rack your brain for examples you’ve heard of and approaches you’re curious about. Maybe you do some research and find studies on the effectiveness of media literacy programs versus apps that limit access to TikTok and other platforms. To get some words down, you’ll have to reflect on what stood out to you in the research and how it connects to your prior assumptions and examples you’ve seen. You might get stuck, but there are plenty of strategies that help, from voice typing to timed writing with partners to giving yourself permission to “write the worst junk in the world,” as Anne Lamott puts it. (See the section on Brainstorming in my textbook, How Arguments Work: A Guide to Writing and Analyzing Texts in College for more strategies.)

Once you have a mess of a brainstorm, if you try to make an outline or organize it into paragraphs, you’ll need to reread what you wrote and notice when you switched from one point to another. You’ll reflect on which points seem important. Maybe you’ll consider which example from the beginning of the brainstorm relates to the study you mention later and what point they both illustrate that could become a topic sentence. As you sort these things out, you start feeling like you have a map in mind and you know much better what you want to say.

Chatbot rewrites can make a piece of writing worse and kill your voice

If you ask a chatbot to polish your essay, it might just butcher it. AI rewrites can sound academic and sophisticated, but AI often changes the meaning as it edits. It can be harder when you have a polished sounding rewrite in front of you to figure out whether it’s what you want to say or how you would say it. For this reason, I don’t ask AI for rewrites myself.

Here’s an example where a ChatGPT rewrite makes student writing worse. I gave ChatGPT the introductory paragraph from “We Don’t Care About Child Slaves” by Jennifer Vo-Nguyen of Portland State University, a student essay selected for inclusion in a college writing textbook, EmpoWORD: A Student-Centered Anthology and Handbook for College Writers. 

Here is the original student-written opening from EmpoWORD: A Student-Centered Anthology and Handbook for College Writers:

“When you walk into the mall or any department store, your main goal is to snatch a deal, right? You scout for the prettiest dress with the lowest price or the best fitting jeans with the biggest discount. And once you find it, you go to the checkout and purchase it right away. Congratulations—now it’s all yours! But here’s the thing: the item that you just purchased could have possibly been made from the sweat, blood, and tears of a six-year-old child in Vietnam. According to the International Labour Organization (ILO), one in ten Vietnamese children aged 5 to 17 are slave workers, and Vietnam is the second biggest source of imported goods to the US.”

Here is the ChatGPT rewrite (See also the full chat session transcript.):

“When you step into a mall or department store, your primary goal is often to find a great deal. You scour the racks for the prettiest dress at the lowest price or the best-fitting jeans with the biggest discount. Once you find the perfect item, you head to the checkout, purchase it, and it's all yours. But have you ever stopped to consider the true cost of that bargain? The item you just bought might have been made by the sweat, blood, and tears of a six-year-old child in Vietnam. According to the International Labour Organization (ILO), one in ten Vietnamese children aged 5 to 17 are involved in child labor, and Vietnam is a major source of imported goods to the US.”

The rewrite sounds fine, even a little smoother than the original. But it also sounds less alive. It wouldn’t get a better grade in my class. Most teachers hope students will write something that expresses their voice–their curiosity, interests, strong opinions, passions, or feelings of uncertainty. That’s where the energy is–that’s what we like to read!

ChatGPT’s rewrite also detracts from the student’s point. Jennifer Vo-Nguyen clearly intends to protest what she sees as slavery, not just labor. Using the word “slave” expresses moral outrage more than “child labor.” “Slavery” echoes the language of her sources, such as a BBC article titled “Vietnam’s Lost Children in Labyrinth of Slave Labour.” 

Below you see ChatGPT’s edits in the screenshot below with the EditGPT extension.

The red crossouts seem so authoritative. What if this student had taken the advice because she didn’t feel confident? Would the teacher still have chosen the more boring, AI–polished version as a model for inclusion in a textbook?

Grammar suggestions can get in the way, too

Grammarly’s basic identifications of grammar errors and suggested fixes can be helpful (I use them myself), but it’s key to stay skeptical. Sometimes the grammar suggestions are wrong, and sometimes they change the meaning completely. Remember that Grammarly is always trying to sell you its product by suggesting that you need it because it knows better.

Grammarly even claims it can do better than Dr. Martin Luther King, Junior. His “I Have a Dream” speech, is surely one of the most powerful and influential speeches in American history, but when I put it into Grammarly, the app claimed to find 16 errors and 5 ways to “improve your text” for clarity. It dangled the promise of 18 “Pro tips” for paying subscribers. Some of its suggestions, such as suggestions for comma insertion, were fine but not important.  Others were just wrong. Here in the first paragraph, it highlighted the phrase “been seared” in red: 

Below is a screenshot of the Grammarly explanation. It suggests that MLK should “Correct the verb” by changing “been seared” to “suffered.” 

It’s true that “been seared” is passive voice, and sometimes passive voice is overused or less effective than active voice. That might be why Grammarly flagged it. Here, however, passive voice is needed. “Seared” vividly evokes suffering through burning, connecting to the “flames” mentioned later in the sentence and adding to the drama and specificity of the sentence. It reminds us of “searing,” which describes intensely damaging and fiery effects, as in “a searing critique.” ChatGPT’s “suffered” is more generic. Given the powerful, epic, biblical style Dr. King was aiming for, “seared” fits beautifully and should not be sacrificed to the Grammarly “fix.”

Chatbot sentence-level style and grammar suggestions can also get in the way of more interesting, varied, and authentic expression. Chatbots are biased toward standardized English (See Don’t trust AI: it’s biased). I tested this by giving ChatGPT the opening paragraph of Dr. Vershawn Ashanti Young’s famous essay “Should Writers Use They Own English?” When I instructed it simply to “Revise this,” it rewrote the paragraph without any of the “Black English” (as Young calls it there) that Young had so consciously and defiantly embraced. When I gave it the same passage and asked “How could this be improved?” it again rewrote the passage without any Black English. This time it also injected an admonition to ensure “proper grammar.”

Removing linguistic variation goes against the idea of “Students' Right to Their Own Language,” a concept affirmed by College Composition and Communication (1974) of the National Council of Teachers of English in 1974 and reaffirmed in 2003 and 2014. They declared, “We affirm the students' right to their own patterns and varieties of language -- the dialects of their nurture or whatever dialects in which they find their own identity and style. . . . ”

As we’ll see in Getting the most out of AI feedback, some uses of AI feedback may support linguistic justice and help you build confidence in your voice and the way you might choose to reach readers through language variation and code meshing. But we’ve got to keep a tight rein on AI to make that work.

Further reading

• The introduction to How Arguments Work: A Guide to Writing and Analyzing Texts in College, including Why Study Argument? and A Closer Look at Fast and Slow Thinking
• The Seductions of AI for the Writer’s Mind by Meghan O’Rourke in The New York Times, July 18, 2025