On November 30, 2022, OpenAI released ChatGPT - a conversational AI system built on their GPT-3.5 large language model - to the public for free. Within five days, it had one million users. By the time I write this on December 28, the number is estimated to be well north of that, and the system has become the fastest-growing consumer technology application in history.

I have spent the past four weeks testing ChatGPT extensively. I have asked it to write lesson plans, explain complex scientific concepts at different reading levels, generate quiz questions, draft parent communications, analyze student writing samples, create rubrics, and write essays on topics I teach. I have tried to break it, trick it, and find its limits.

Here is my honest assessment, written from the perspective of someone with 28 years of K-12 teaching experience, Google Certified Educator credentials at Level 1 and 2, an Adobe Creative Educator certification, and an Apple Teacher designation: this technology changes everything. Not eventually. Now.

What ChatGPT Can Do - And What It Means

Let me start with the capabilities, because I think many educators have not yet fully grasped what this system can do.

I asked ChatGPT to write a five-paragraph essay on the causes of the American Civil War, targeted at an 8th-grade reading level. It produced a coherent, well-organized essay in approximately four seconds. The essay was not brilliant - it was generic and lacked the kind of specific, surprising analysis that distinguishes excellent student writing. But it was competent. It would have earned a B or B+ in most 8th-grade classrooms.

I then asked it to write the same essay at a 12th-grade AP level, incorporating historiographical debate. It did so. I asked it to revise the essay to argue a specific thesis about economic factors. It did so. I asked it to add citations in Chicago format. It fabricated plausible-looking citations (this is a significant problem I will address below), but the formatting was correct.

The speed and quality of this output has immediate implications for how we think about writing assignments, homework, assessment design, and academic integrity. But those are the surface implications. The deeper implications are about what we value in education and why.

My Genuine Emotional Response

I want to be honest about something that I think many educators are feeling but few are saying publicly: my first reaction to ChatGPT was fear.

Not fear of obsolescence - I do not believe AI will replace teachers, for reasons I will explain. But fear of inadequacy. Fear that the assessment structures I have spent decades building are suddenly brittle. Fear that the writing assignments I have carefully designed can now be completed in seconds by a machine. Fear that the gap between what I know how to teach and what students need to learn just widened enormously, and I am not sure I know how to close it.

I think it is important to name that fear, because I believe it is widespread and it is driving some of the worst impulses in the early institutional responses to ChatGPT. Schools that are rushing to ban the tool, install AI detection software, and return to handwritten in-class essays are responding to fear, not to pedagogy. And fear-based decisions in education are almost always wrong.

What ChatGPT Cannot Do

After four weeks of intensive testing, the limitations are as important as the capabilities.

It fabricates information with complete confidence. ChatGPT does not know what is true. It generates text that is statistically probable given its training data. This means it will state falsehoods with the same fluent confidence it states facts. I asked it to name three peer-reviewed studies on the effectiveness of flipped classroom pedagogy. It generated three citations that looked real - complete with author names, journal titles, volume numbers, and page ranges. Two of the three citations were entirely fabricated. The journals existed, but the articles did not.

This is not a minor limitation. It is a fundamental characteristic of the technology, and it has direct implications for how we teach students to use it. Any educational use of ChatGPT must include explicit instruction in verification - the skill of checking the AI's outputs against authoritative sources.

It cannot reason about novel problems. ChatGPT excels at tasks that resemble its training data - summarizing known information, writing in familiar formats, answering well-documented questions. It struggles with genuine novelty: problems that require original analysis, creative synthesis of disparate sources, or reasoning from first principles about unfamiliar situations. This limitation points toward where human cognition remains irreplaceable in education.

It has no understanding of context, motivation, or learning progression. ChatGPT does not know who the student is, what they have already learned, what misconceptions they hold, or what they need to learn next. It cannot adjust its explanation based on a student's facial expression, body language, or the particular way they formulated a question. These are things teachers do constantly, often unconsciously, and they are essential to effective instruction.

Three Things Schools Should Do Right Now

Based on my first month with this technology, I have three immediate recommendations for schools.

First, do not ban ChatGPT. Banning it is both practically futile and pedagogically counterproductive. It is a free web application accessible from any personal device. Students will use it whether schools ban it or not. More importantly, learning to work effectively with AI systems is going to be a critical professional skill. Schools that ban AI tools are not protecting students from cheating; they are depriving them of essential preparation for the world they will enter.

Second, redesign assessments immediately. The five-paragraph essay assigned as homework and submitted as a finished product is now an obsolete assessment format. Not because it was ever particularly effective (the educational critique of the five-paragraph essay predates AI by decades), but because it is now trivially easy to outsource to a machine. Assessments need to move toward formats that AI cannot easily replicate: process-based writing (where the journey from brainstorm to draft to revision is documented and evaluated), oral defense of written work, in-class composition with specific prompts that connect to classroom discussions, and portfolio-based assessment that tracks growth over time.

Third, start using ChatGPT yourself - extensively - before forming opinions about policy. I am alarmed by the number of administrators and policymakers making decisions about AI in education without having spent meaningful time interacting with the technology. You cannot make informed policy about something you do not understand. Every educator and administrator should spend at least 10 hours experimenting with ChatGPT before participating in any policy discussion about it.

The Bigger Question

Behind the immediate practical concerns - cheating, assessment design, detection software - lies a much larger question that I believe will define the next decade of education: what is the purpose of learning to write?

If the purpose of writing instruction is to produce a polished text artifact, AI has rendered much of that instruction unnecessary. A machine can produce a competent essay faster and more reliably than most students.

But if the purpose of writing instruction is to develop thinking - to force students to organize ideas, construct arguments, weigh evidence, grapple with complexity, and communicate with precision - then the value of writing instruction is entirely undiminished. The thinking that happens during the writing process is the learning. The artifact is just evidence that the thinking occurred.

This distinction - between product and process, between artifact and cognition - is going to be the central philosophical question in education for the foreseeable future. And how we answer it will determine whether AI makes education better or worse.

I do not yet know the full answer. But after 28 years in classrooms, I know this: the teachers and schools that engage with this question honestly, that experiment courageously, and that prioritize student thinking over student outputs will be the ones that thrive.

The rest will be playing defense against a technology that is already smarter than their assessment designs. And defense, in education as in everything else, is a losing strategy.

Dr. Janette Camacho is a K-12 educator with 28+ years of classroom experience, a Google Certified Educator (Level 1 & 2), Adobe Creative Educator, and Apple Teacher. She is beginning what she expects will be an extended exploration of artificial intelligence in education.