A new school year always comes with its own set of questions. Which coursebook? Which exam class needs the most attention? This September, though, the question I keep hearing from colleagues is a bigger one: should we be using AI with our students at all?

It's a fair question, and the news hasn't made it any easier. Over the summer, some of the most respected universities in the world made headlines for taking AI out of certain classrooms. If you read those stories and felt a little relief, or maybe a little guilty about the AI tools you already use, you are in very good company.

So let's look at what is really happening, why it makes sense, and what it means for those of us teaching English, one essay at a time.

What the universities are actually doing

The most talked-about case is the University of Chicago. From this autumn, its Social Sciences Core, a required part of the undergraduate curriculum, is being taught as "analog" courses: texts on paper, discussions without laptops or phones, and assignments completed without AI assistance. The rule applies to instructors too, who are told not to rely on AI grading unless they have carefully checked it against human grading. When the policy became public, the chairs of the humanities Core were still discussing their own approach.

At Harvard, the debate is happening out in the open. The Dean of Harvard College suggested that writing-intensive courses might accept or even encourage AI, and many humanities professors pushed back. A Harvard Crimson analysis of around 600 autumn courses found that more than half of science and engineering courses allow some AI use, compared with only 27% in the arts and humanities. One English professor is keeping a complete ban in her courses, explaining that students discover their own writing style by wrestling with language themselves. At Princeton, the history department already bans AI for generating and editing text, and does not allow it even for brainstorming students' independent work.

One detail is worth keeping in mind. These are decisions about specific courses and departments, not whole universities turning their backs on AI. Chicago's policy still leaves room for professors to experiment thoughtfully with AI in consultation with their division, and Harvard has introduced a mandatory AI module in its first-year writing courses.

Why this is a good move

The reasoning behind these decisions is simple, and I believe it's right. Writing is how we find out what we think. When a machine does the drafting, students skip the very struggle that builds a mind of their own.

Research is starting to back this up.

Confidence comes first. Researchers from Microsoft and Carnegie Mellon surveyed 319 professionals about 936 real examples of using generative AI at work. People who had more confidence in the AI reported thinking less critically about its output, while people who had more confidence in their own abilities reported thinking more critically (Lee et al., 2025). Put simply, you need to trust yourself before you can judge a machine well.

Skipping the effort has a cost. In an MIT Media Lab study, participants who wrote essays with ChatGPT showed the weakest brain connectivity of three groups and struggled to quote from essays they had written only minutes earlier (Kosmyna et al., 2025). It was first released as a preprint with a small sample, so it's best read as a warning sign rather than a final verdict.

Shortcuts don't teach. In a field experiment with nearly a thousand high school maths students in Turkey, those given unrestricted access to a ChatGPT-style tutor did 48% better during practice. Once the tool was taken away for the exam, they scored 17% lower than students who never had AI at all (Bastani et al., 2025).

For us language teachers, this lands close to home. Our students' confidence in English is often fragile. If a chatbot always sounds more fluent than they do, why would a fifteen-year-old trust their own sentence? And a student who doesn't trust their own English is exactly the student who will accept whatever the AI says, mistakes included. Protecting that confidence isn't old-fashioned. It's what makes a critical AI user possible in the first place.

Why a ban isn't the whole story

Here is the part that tends to get lost in the headlines: the same research that warns us about AI also shows us how to use it well.

Look again at that maths experiment. A second group used a tutor built with teacher input, designed to give hints instead of answers. They did 127% better during practice, and on the exam they performed about the same as the students without AI. The harm was largely avoided. The difference was entirely in how the AI was set up and used.

The MIT study has a similar twist. In its final session, participants who had been writing on their own and only then switched to ChatGPT showed better memory recall and re-engaged a wider network of brain areas than those who went the other way round. Order matters: think first, then bring in the tool.

In our own field, the evidence for combining AI with human teaching keeps growing. As we shared in an earlier post, Escalante, Pack and Barrett (2023) found no significant difference in learners' writing improvement whether feedback came from GPT-4 or a human tutor, and learners' preferences were split almost evenly between the two. Newer studies go a step further:

• In a study with 68 EFL learners writing IELTS Task 2 essays, those who received ChatGPT feedback together with teacher feedback improved significantly more than those who received teacher feedback alone, across every scoring criterion (Asadi, Ebadi & Mohammadi, 2025).
• A randomised trial with 88 learners compared teacher feedback, AI feedback and a hybrid of the two. The hybrid group made the largest gains, particularly in task achievement, coherence and grammatical accuracy, and those learners reported feeling more confident and less anxious about their writing (Soori, Khojasteh & Javed, 2025).

That last finding is worth pausing on. Used alongside a teacher, AI feedback didn't weaken learners' confidence. It supported it.

To be fair, not every study points the same way. One recent study found that adding ChatGPT feedback to teacher feedback did not help students produce more complex sentence structures than teacher feedback alone (Alkhalifah & Almutlaq, 2026). Most of this research also involves small groups of learners, so there is plenty still to learn.

Blanket bans have a practical downside too. As a columnist in Princeton's student newspaper pointed out, students who ignore the rules may simply use AI in secret and gain an unfair advantage over the ones who follow them. Clear guidance tends to work better than silence.

Making it work in your classroom this year

So where does all this leave us? Not with a choice between "AI everywhere" and "AI nowhere", but with a sensible division of labour: students do the thinking, AI helps with the heavy lifting, and the teacher stays in charge of both.

Here are a few ideas to start the year with.

Keep the first draft human. Have students write in class, by hand if you like, before any tool comes near their work. This is the MIT finding put into practice: brain first, AI second. Many of you already do this, as we discussed in our post on plagiarism.

Put AI on your side of the desk. Let it take on the repetitive part of the job, such as marking errors across a stack of thirty essays, and spend the time you save on what it can't do: talking a student through their argument, or rewriting a weak paragraph together. That human conversation is where the hybrid studies saw their gains.

Teach students to argue with AI. Give them an AI-corrected paragraph and ask what they disagree with, or what the AI missed. The Bastani research team has since reported that training students with deliberately flawed AI examples improved their ability to spot and correct AI errors. Every time a student successfully questions a machine, their confidence grows a little.

Stay the final judge of grades. Whatever tool you use, ours included, compare its marks with your own judgement regularly, especially in the first weeks. It's the same standard Chicago set for its instructors, and it's a good one.

Make the rules visible. Tell students when AI is off limits, when it's welcome after they've drafted on their own, and why. Students can't build healthy habits around a tool they are only ever told to hide. If you'd like a starting point for that conversation, our post on ethics for a new culture of teaching is a good place to begin.

Final thought

The universities going "analog" are protecting something precious: students' ability to think for themselves and trust their own voice. That is exactly what young people will need in order to question AI, rather than simply follow it.

Our job this year is much the same. Protect the thinking, keep the human conversation at the centre of our lessons, and let AI take on the workload that keeps us away from our students.

Wishing you a calm start and a wonderful school year.

References:

1. UChicago Goes "Analog" for Some Courses, Inside Higher Ed, 26 August 2026. Link
2. The University of Chicago Expands Bans on AI and Other Technology, Michael T. Nietzel: Forbes, 24 August 2026. Link
3. Sosc Core to Institute AI Ban, Technology-Free Classrooms This Fall, The Chicago Maroon, August 2026. Link
4. Harvard Humanities Faculty Push Back on Deming's Call for AI Use in Writing Courses, Abigail S. Gerstein & Amann S. Mahajan: The Harvard Crimson, 3 September 2026. Link
5. Humanities lag behind STEM in AI policy. They must catch up., The Daily Princetonian, February 2026. Link
6. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers, Hao-Ping (Hank) Lee, Advait Sarkar, Lev Tankelevitch, Ian Drosos, Sean Rintel, Richard Banks & Nicholas Wilson: Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 2025. Link
7. Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, Nataliya Kosmyna, Eugene Hauptmann, Ye Tong Yuan, Jessica Situ, Xian-Hao Liao, Ashly Vivian Beresnitzky, Iris Braunstein & Pattie Maes: arXiv preprint, 2025. Link
8. Generative AI without guardrails can harm learning: Evidence from high school mathematics, Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı & Rei Mariman: Proceedings of the National Academy of Sciences, volume 122, issue 26, 2025. Link
9. Summary of the Bastani et al. study, Knowledge at Wharton. Link
10. Unpacking the Unintended Consequences of AI in Education, Hamsa Bastani: SRI Seminar Series, University of Toronto. Link
11. AI-generated feedback on writing: insights into efficacy and ENL student preference, Juan Escalante, Austin Pack & Alex Barrett: International Journal of Educational Technology in Higher Education, volume 20, Article 57, 2023. Link
12. The impact of integrating ChatGPT with teachers' feedback on EFL writing skills, Marjan Asadi, Saman Ebadi & Laleh Mohammadi: Thinking Skills and Creativity, volume 56, Article 101766, 2025. Link
13. Comparing Teacher E-Feedback, AI Feedback, and Hybrid Feedback in Enhancing EFL Writing Skills, Afshin Soori, Laleh Khojasteh & Fareeha Javed: Technology in Language Teaching & Learning, volume 7, issue 3, Article 102626, 2025. Link
14. The Combined Impact of ChatGPT and Teacher Feedback on the Syntactic Complexity of EFL Learners' Writing, Eman Alkhalifah & Sana Almutlaq: SAGE Open, 2026. Link