AI In PracticeBlog Post

What I Say Before I Teach Anyone to Use AI

Jun 24, 20265 min read
What I Say Before I Teach Anyone to Use AI

“This feels like cheating.”

I hear that more than anything else when I run these sessions, usually as soon as the room warms up enough for someone to say it out loud. It’s worth answering first, because nothing else I teach lands until it’s settled.

Using AI to save time isn’t unethical. What matters is how and why. It’s already being used in a lot of creative ways as an accommodation, and however you decide to use it, if it helps you do your work better, it’s valid.

That’s the easy half. The harder half is that there are things I say before I teach anyone a single prompt, and they aren’t optional.

This is the first of two posts. This one is what comes before. The second is what to do when it isn’t working.

It’s a drill, not a magic wand

A screwdriver is great. You can get a lot done with one. But I’m not taking on a kitchen reno without a drill handy.

That’s roughly where I’ve landed on AI and planning. It’s a power tool, which means it’s faster and it makes a bigger mess when you point it at the wrong thing. Treat what comes out as a draft, not a deliverable.

The training data isn’t neutral

AI predicts patterns from what it was trained on, and what it was trained on is not the world.

A 2024 audit by the Data Provenance Initiative looked at nearly 4,000 public datasets across 67 countries. North American and European sources accounted for 93% of the text. African and South American organisations accounted for under 0.2%.

That’s not an argument against using it. It’s the reason the defaults skew the way they do, and it’s why “make this more inclusive” is a weaker instruction than naming what you actually want to see. If you don’t push, you get the middle of that distribution.

Two guardrails, and neither is negotiable

Protect privacy. Never put student names or identifying details into an AI tool. My rule of thumb: if you wouldn’t write it on a whiteboard, don’t paste it into AI.

Verify before you trust. AI can invent sources and oversimplify. Anything factual, curricular, sensitive, or headed for students gets checked. You can ask directly, “verify all factual information and give me links to your sources”, and then you have to actually open the links. The asking is not the checking.

I don’t teach detection, and I won’t

This is the part where I lose some rooms, and I’d rather lose them here than pretend.

Trying to catch AI use with detection tools is unreliable and unethical. A 2023 study in Patterns found detectors flagged more than half of essays by non-native English writers as AI-generated, while classifying native-speaker essays almost perfectly. A 2024 Common Sense Media survey found 20% of Black teens had been falsely accused of using AI on an assignment, against 7% of white teens. OpenAI quietly retired its own detector in 2023, citing its low rate of accuracy.

Disability advocates raise the same concern about neurodivergent students. I haven’t found a study on that one, so I’m not going to lend it the authority of the two above, but it’s the pattern you’d expect.

The point isn’t that students never misuse this. It’s that the tool we’d use to catch them is wrong often enough, and wrong in a patterned enough way, that it does more damage than the thing it’s policing. So the session is about designing learning thoughtfully. Not surveillance.

The one setup step worth doing

Everything above is a caution. Here’s the thing that actually saves you time, and you only do it once per tool.

Most AI tools have a personalization or memory setting. Paste in a short block telling it who you are: your grade and subject, that you want clear practical language rather than formal prose, your teaching priorities, and what to always include when you ask for a lesson. Then end it with something like “I’m the expert on my students. You’re my brainstorming partner, not a replacement for my professional judgment.”

It sounds small. It removes an enormous amount of repetitive explaining, and it stops later prompts feeling so generic.

If filling in a template isn’t your thing, there’s a shortcut. Upload two or three lesson plans or newsletters you’ve already written, with names removed, and ask it to analyse your writing and teaching style. Then tell it to remember that.

The bottom line

  • Using AI to save time isn’t cheating. How and why is what matters
  • Treat outputs as drafts, not deliverables
  • If you wouldn’t write it on a whiteboard, don’t paste it into AI
  • Ask for sources, then open them. Asking isn’t checking
  • Detection tools are unreliable and biased in ways that fall on specific students. Design instead of policing
  • Set up your tool once, so you stop re-explaining yourself

None of this requires CARL, and none of it is complicated. It’s just the part that has to come before the interesting bit.

Read part two, on what to do when the output isn’t working.

Illustrated portrait of Courtnay Boateng with CARL peeking over her shoulderWritten byCourtnay BoatengCo-Founder, CARLCourtnay Boateng is CARL’s co-founder. She holds an MEd from UBC Okanagan, where her research asked how AI could make equitable education more accessible. She focuses on AI use in education and workflow optimization, and always has a new idea in progress.More about who we are