ChatGPT in Schools: A Teacher's Perspective 18 Months In

Eighteen months ago, when ChatGPT first became widely available, I had a conversation with a colleague in the staffroom that I suspect was happening in schools across the city. She teaches English. She'd just received an S5 essay that was suspiciously polished — coherent argument, varied sentence structures, not a single grammar error. Her question to me was simple: "What do I do now?"

I didn't have a good answer then. I think I have a better one now, though it's not the one most people expect.

What I got wrong initially

My first instinct was to treat ChatGPT as primarily a cheating problem. Write better assessment questions. Require in-class work. Use detection tools. This wasn't wrong, exactly, but it was defensive in a way that missed the bigger picture.

The thing about framing AI as a cheating threat is that it puts teachers in an adversarial relationship with a tool that students will use for their entire working lives. I was essentially teaching students that the correct response to a powerful capability is to hide it.

About six months in, I shifted my approach. Not because I stopped caring about academic integrity, but because I realised that integrity in 2024 means something different than it did in 2010. A student who learns to use AI thoughtfully, critically, and honestly is developing a legitimate skill. A student who learns to use it deceptively is not.

What's actually changed in my classroom

I now explicitly teach students to use AI as a thinking tool, with clear rules about when that's appropriate.

For research and exploration: fine. "Ask ChatGPT to explain three different perspectives on this environmental issue" is a valid starting point for an essay — not the essay itself.

For checking reasoning: useful. "I think the answer is X — does ChatGPT agree, and if not, what's different about its reasoning?" This is better than just submitting X and hoping for the best.

For generating finished products: not acceptable, and the reason matters. It's not about the tool being bad. It's about the student not learning. If ChatGPT writes your essay, you haven't practised writing. If you haven't practised, you can't write without it. The dependency compounds.

I've also noticed something that surprised me: students who use ChatGPT poorly — who just paste in questions and copy answers — tend to perform worse over time, not just in my assessments but in tasks where AI isn't available. The students who use it well — who argue with it, question its outputs, use it to find ideas and then develop those ideas themselves — are getting sharper.

The accuracy problem is real and underappreciated

This is the thing I spend the most time on with students now. ChatGPT is confidently wrong in ways that are genuinely dangerous for learning.

I had a student submit a report on computer networking that cited a "2022 study by Stanford researchers" on latency optimisation. I couldn't find the study. I asked him for the source. He went back to ChatGPT. The study didn't exist — the tool had invented it, complete with plausible-sounding author names and a journal title.

This isn't an occasional glitch. It's an inherent feature of how large language models work. They predict plausible text, not true text. For topics where plausible and true overlap most of the time — they can seem very reliable. For specific factual claims, dates, statistics, source citations — they are genuinely unreliable and your child needs to know this.

My rule: every factual claim from ChatGPT needs an independent source. Not another AI tool. An actual source — a textbook, a verified website, a published article. This takes longer. It's also how research actually works.

What I tell parents

Three things.

First: don't assume your child isn't using AI for homework because they haven't mentioned it. The survey I ran this year showed that most students assumed their parents didn't know what ChatGPT was. That assumption creates a gap in the conversation at home.

Second: the questions to ask aren't about whether they used it. They're about how. "Did ChatGPT give you any ideas?" is a fine starting conversation. "Can you explain to me what the essay is actually arguing?" is the follow-up that reveals whether the tool helped them think or thought for them.

Third: resist the urge to see this as a school problem that schools should solve. Schools are adapting — some faster, some slower, some more thoughtfully than others. But the habits around technology use are formed at home, in the evenings, during the hours when teachers aren't watching. Your role in this is real.

Where I am 18 months in

I'm cautiously optimistic. The panic of the first few months has given way to something more interesting: a genuine rethinking of what we're trying to achieve with school assessments. If a task can be completed by AI in 30 seconds, that's worth asking whether the task was testing what we thought it was testing.

Some of my colleagues find this uncomfortable. I find it clarifying. It's forcing us to be more specific about what we actually want students to be able to do — and that's a conversation education needed to have regardless of AI.

ChatGPT didn't break schools. It revealed what was already fragile, and gave us a chance to fix it.