Welcome, and thanks for reading Real-World AI for Teachers! If you’re new here, this free monthly newsletter is designed to help educators keep up with AI without having to keep up with everything about AI. Each issue includes a handful of important updates, practical ideas, and resources I think are worth your time. You can expect it in your inbox around the middle of each month, and I’ll never sell your email address or fill your inbox with extra messages. If something here sparks a question or an idea, feel free to reach out—I’m always happy to keep the conversation going.

Microsoft, the American Federation of Teachers, and the United Federation of Teachers announced a new AI safety and privacy standard for schools. Among the protections: student and educator data cannot be used to train AI models, schools keep control over how data is used and deleted, and AI systems must include human oversight.

What caught my attention is that these are meant to be enforceable protections, not just a list of nice ideas. As more AI tools find their way into classrooms, teachers shouldn’t need a law degree and a magnifying glass to figure out what happens to student data. This agreement may give districts a useful model for the questions they should be asking every AI vendor.

New York City announced a one-year moratorium on student-facing generative AI for students from 2-K through eighth grade, affecting nearly 600,000 students. High school students will still receive AI literacy instruction, while a smaller group will participate in tightly controlled classroom pilots.

I find this especially interesting because so much of the education conversation has moved quickly from “ban AI” to “teach students to use it.” New York City is asking a different question: when are students developmentally ready to use it? That feels like a conversation worth having. Just because a tool can help students over a difficult part of learning doesn’t always mean we should remove the difficulty.

New guidance from the U.S. Department of Education asks schools to judge classroom technology by its instructional value, evidence of effectiveness, transparency, and impact on student learning. The guidance even gives schools five questions to ask, beginning with a wonderfully simple one: What learning problem does it solve?

That may be one of the healthiest questions we can bring to AI right now. We have plenty of tools that can generate something impressive in ten seconds. The harder question is whether using that tool actually makes students better thinkers, writers, scientists, mathematicians, or learners. “Look what it can do!” is a pretty low bar for instructional technology.

OpenAI announced that ChatGPT for Teachers is expanding to 55 additional school systems across 20 states, reaching more than 100,000 additional educators and staff. OpenAI says it is now working with more than 100 K–12 organizations and providing access and training to more than 300,000 educators and staff.

The interesting part to me is the scale. AI in schools is moving beyond individual teachers quietly experimenting with a chatbot during their planning period. Districts are beginning to adopt, manage, and train around these tools systemwide. That means the conversation has to grow too—from “What prompt should I use?” to questions about privacy, instructional purpose, professional learning, and what we actually want AI to do in schools.

AI is pretty good at giving students answers.

That may be exactly why one of my favorite ways to use it is to ask for wrong ones.

For this activity, students use AI to generate believable mistakes about something they are learning. Then they have to figure out exactly what makes each answer wrong, explain the misconception behind it, and improve the answer.

The goal is to move students beyond knowing the right answer toward understanding why other answers fail.

Give students a concept, question, problem, or learning target they have already studied.

Instead of asking AI to explain it, have them ask for several plausible but incorrect responses.

For example, in my science classroom, I could give students:

Then students might use a prompt like:

The important word here is believable.

“Because the water gets angry” is technically wrong, but it doesn’t require much scientific thinking to reject it. I want errors that make students hesitate for a second.

Good distractors live close enough to the truth that students have to inspect them carefully.

Once students have their AI-generated wrong answers, the real work begins.

For each one, ask students to explain:

That last step matters.

Students aren’t simply hunting for errors like tiny academic exterminators. They are reconstructing the thinking.

You can also have pairs trade their incorrect answers and see whether another group can diagnose the misconception without being told what the AI was trying to imitate.

There’s another version of this activity that I especially like.

Give students a correct answer and challenge them to create a multiple-choice question with really good distractors.

AI can help brainstorm possibilities, but students decide which wrong answers are useful enough to keep.

Then ask:

What would someone have to misunderstand in order to choose this option?

That question changes the task completely.

Students begin thinking like assessment designers. They have to anticipate confusion, distinguish between closely related ideas, and decide what understanding the question is actually measuring.

If every wrong answer is obviously ridiculous, the question probably isn’t doing much diagnostic work.

A multiple-choice question where one answer is “photosynthesis” and the others are “pizza,” “France,” and “Tuesday” may produce excellent scores, but I’m not sure we should start printing certificates.

You can add this activity to almost any lesson without rebuilding the assignment.

After students have learned a concept, give them a question you might normally use for review and have them use a prompt like:

Then take the AI away.

Students have to analyze the responses using their notes, texts, data, calculations, or other evidence from the lesson.

You could even finish by revealing the AI conversation and asking students whether its supposed “misconceptions” were actually realistic.

Sometimes they won’t be.

That becomes another useful question:

Would a real student actually think this way, or is AI just making something that sounds wrong?

Correct answers can sometimes hide shallow understanding.

A student may remember that salt water is a mixture, that the mitochondria produce usable energy, or that the author’s evidence supports a particular claim without understanding the boundaries of that idea very well.

Wrong answers expose those boundaries.

To explain why an answer fails, students have to compare ideas, locate the flaw, retrieve what they know, and defend their reasoning. AI provides something worth arguing with, but students still have to supply the judgment.

That fits one of the principles I keep returning to in The Learning Forge: AI is most useful when it creates more thinking for students rather than removing the thinking they need to do.

And there is a nice bonus for teachers.

Listen carefully while students debate the wrong answers and you may discover misconceptions you didn’t know they had.

At that point, the AI-generated mistakes have done something much more valuable than simply giving students another practice question.

They’ve given you a window into their learning.

Want to see more? Check out the Learning Forge One-Pager for more ideas from the upcoming book.

With the school year starting, my work on The Learning Forge manuscript is on hold, while I work with my PLC at school to create the best learning experiences for our students.

I will be in Kennesaw, Georgia in a few weeks working with a small group of eager educators who want to know how AI tools can help them better differentiate for their students’ needs.

October’s newsletter will a list of topics that I will be presenting about in the next year.

Paul (and the Codium Educational Consulting team)

P.S.

Don’t forget to email me with examples of how you’ve used the tools and strategies that I’ve shared.