This is the second of two. The first was what I say before teaching anyone a prompt.
I facilitate workshops on using AI for planning, and the first thing I show people is more of a hack than anything else: use the AI to work out how to use the AI.
The session runs on a document of reusable profiles, tested workflow prompts and a toolkit of optional lenses. But the tab people actually go back to is the troubleshooting guide, which is just a list of things that go wrong and what to try instead. That part isn’t a secret and it doesn’t need CARL, so here it is.
“It’s too generic and doesn’t fit my class”
The most common one, by a distance. AI doesn’t know your students, your room, or how you teach unless you tell it.
Instead of: “Create a math lesson.”
Try: “Create a Grade 3 math lesson. Here’s my class profile. Make it hands-on and doable in 15 minutes using manipulatives.”
The profile is the part that keeps paying off. Write your class down once as a reusable block: grade and subject, rough size, the spread of reading levels, what supports are needed, what works with this group and what falls flat. Then paste it in each time instead of retyping it.
One rule while you’re writing it. Run the hallway test. If a colleague read that profile over your shoulder, could they identify a specific student? If so, make it more general. Use ranges instead of counts, and “several students need extended time” rather than a number. AI works just as well with “some” as with anything more precise, and there’s no version of this where naming a kid helps.
“It’s too long and overwhelming”
AI defaults to comprehensive. You’re allowed to just say no.
“Make this half as long.” “Keep it under 200 words.” “Just give me the hook and guided practice.”
Set the constraint up front rather than trimming afterward.
“It made something up and I don’t trust it”
AI fills gaps with plausible-sounding information. That’s the whole problem, because plausible-sounding and correct look identical on a screen.
Ask where things came from. “Where did this information come from?” Then check the specifics yourself, particularly dates, statistics and curriculum language. .
Or, ask the AI to admit when it doesn’t know or is unsure. “If you are unsure about any of the info you provided, let me know.” That way, you can work on getting that info yourself, or trigger the AI to research it for you.
“You said this is a BC Grade 5 Big Idea. Can you confirm that’s accurate? Or if you are unsure, let me know.”
“It’s close, but not quite right”
Here’s the expensive mistake. You get something 80 percent right, the independent practice is wrong, so you retype the whole prompt and receive a different 80 percent.
“This is good, but change the independent practice from written work to a hands-on task. Keep everything else.”
That last clause is the whole move. Without it the model helpfully rewrites the parts you liked. AI needs clearer direction, not a full restart.
“I’m stuck and none of this helped”
Then ask it. Directly.
“I’m trying to create [X], but this isn’t working because [reason]. How should I change my prompt?”
It feels circular the first time. It’s also the fastest way to get better at this, which is why I open the session with it instead of ending there.
The bottom line
- Write your class down once and paste it in. Run the hallway test first
- Say how long you want it, because AI defaults to comprehensive
- Ask where information came from, then check the specifics yourself
- Say “keep everything else” instead of starting over
- When nothing works, ask the AI what was wrong with your prompt
None of that needs CARL. It’s what I’d want a teacher to leave a session with, and it works in whatever tool is already open.
We build these into a workflow so you don’t have to remember them at 9 p.m. But they’re yours either way. You’re the one deciding whether the answer is any good.
Read Why Not Just Use ChatGPT, or try one of the four moves in your next planning prompt.
