Teachers look both ways before they tell me what they use AI for.
It happens at conferences, in the corridor, after the session has finished and the lanyards are coming off. Someone waits until the group has moved on, glances down the hall, and then admits they used it to draft next week's lesson plans. Or to build the end of unit test. Or to make three versions of a reading task for a class where three versions were always needed and never possible.
Nothing they describe is against their school's policy. Most of them could not tell you which rule they think they are breaking. The feeling is closer to shame than to risk, and it sounds like the thing you say when you have found a shortcut you have not earned.
Set that aside for now, because there is a second thing those same teachers say, and it is the one that actually stops them.
Almost nobody tells me they tried AI and decided against it. What they say is that they meant to keep going. They used it twice in March, thought it was useful, and then did not think about it again until someone mentioned it in a staff meeting in June.
Two versions of this come up again and again. The first is forgetting at the exact moment it would have helped, then remembering half an hour later, once the task is done. The second is remembering perfectly well, having eleven minutes before a meeting, and reaching for the old method because it is the one you can do without thinking.
Both are memory problems dressed as motivation problems. In the last article I argued that teachers hold on to old methods because the brain automates what works, and automation is efficient. A method you have used for nine years does not require you to remember it. It runs on its own. Anything new has to be actively recalled, and it has to be recalled precisely when you are busiest.
Advice about being more disciplined does nothing here. Neither does a calendar reminder, which you will dismiss with your thumb while walking to your next class.
Practical AI for Busy Teachers averages 4.84 out of 5 from the teachers taking it, so the problem is not that people dislike it. The stall happens at a predictable point, and it is not where you would expect.
Teachers get through how AI works and the prompting framework without much trouble. The drop comes at the module where the course stops explaining and starts asking you to apply it to your own class, your own topic, next week.
That is the moment the routine has to exist, and for most people it does not yet. The teachers who finish tend to do it at about an hour a week over a term. They are not faster than everyone else. They have a slot.
Pick one recurring task. Then remove the old way of doing it.
The task is your weekly retrieval questions. The short set of recall questions you use to start a lesson, close one, or find out what survived the weekend. It repeats on a fixed cycle. It is low stakes, so a mediocre first attempt costs nothing and nobody sees it. It eats twenty minutes a week without ever feeling like teaching.
That is the whole mechanism. You are no longer relying on remembering to use AI, because the task cannot begin without it. The prompt becomes the first step of a job you were going to do anyway, which means the old habit now does the remembering for you.
Do not start with a unit plan. Planning is the task teachers most want help with, and it is the worst place to build a first habit. The stakes make you slow down, argue with the output, and conclude that doing it yourself would have been quicker. It probably would have been, the first time.
Write the prompt before you need it.
If you open a blank chat on Tuesday morning with a class arriving in fifteen minutes, you will type something vague and get something vague back. Then you will blame the tool rather than the prompt. Do this part when you are not under pressure.
Your prompt needs four things, which the course calls RTFC. A role, a task, a format, and context. Build yours below. Three fields is enough to start, and the extra detail can wait until Wednesday.
Leave these until you have run the prompt once and seen what needed fixing. Every correction you write in here stops being a correction you make weekly.
Your prompt
Fill in the fields above and your prompt will appear here…
Open each one after you have seen the output. Then paste the fix into the class detail fields above and copy the prompt again.
This happens because the model has no idea what your class already holds. It guesses at the middle of the year level and lands under it.
Add to "what they are secure with": the two or three ideas they have genuinely locked in. The model will stop testing those and start building on them.Curriculum sequencing varies by country, school and teacher. The model uses whatever ordering was most common in its training, which will not be yours.
Add to "terms they have not met": the words you know are coming later. This is the single highest value line in the prompt, and it takes thirty seconds.This one you do not fix in the prompt. You fix it by reading the key, which takes about ninety seconds and is the part of the job that stays yours.
If it keeps happening on the same kind of question, that is worth noticing. It usually means the topic has genuine disagreement in it, which is often more interesting to teach than the question you asked for.
Teach. Do nothing about this.
A habit that begins on the busiest day of the week has already lost. Let the prompt sit unused for a day.
Run it. Change the topic, paste, read what comes back.
Some of it will be wrong. One question will sit too low. One will use a term your class has not met. The answer key will state something with more confidence than the evidence supports. Fix those, use the rest, and notice what you had to fix.
Fold yesterday's corrections into the prompt.
If the pitch was low, write in what your class is actually working at. If the vocabulary was off, name the three terms they know and the three they do not. Every correction you make once here stops being a correction you make weekly. Ten minutes, and you will not touch it again for months.
Use the same prompt for a different class or topic.
A prompt that works once feels like luck. Working twice is when people start treating it as equipment, and reach for it without deciding to.
Ask whether that was faster than doing it by hand, and answer with a number rather than an impression. Twelve minutes, twenty, whatever it was.
If the honest answer is no, the prompt is probably still too vague. Vagueness in, vagueness out.
Then decide what the time is for, specifically and a little selfishly. Leaving at four on Thursday counts. If you do not name it, the time gets absorbed within a fortnight and you finish the term convinced nothing was saved.
Week one is carried by novelty. Week two is not, and the old method is still there, comfortable and automatic.
Keep the day fixed. Keep the task identical, and resist the pull towards something more ambitious just because the first attempt worked. Hold the rule about not writing them by hand, especially in the week when you are behind, because that is the week the rule exists for.
Get through week two and you are not trialling anything. You have a workflow.
Tell a colleague what you are doing.
Not for accountability, which rarely works. A practice you conceal cannot be built into a routine, because you cannot mention it, compare it, or ask anyone whether their version is better. Hiding it keeps it fragile, and fragile things do not survive week two.
When retrieval questions feel automatic, add differentiating a text. Then rubrics. Then the planning tasks you were tempted to start with, which will be far easier by then, because you will know where your judgement has to stay.
That sequence is what Practical AI for Busy Teachers is built on. Six modules, thirty-three lessons, and a prompting framework you reuse rather than relearn. It is designed for an hour a week, because that is the pace at which teachers actually finish it and keep what they learned.
There is a free preview if you want to see the shape of it first. Otherwise, start with Sunday and ten minutes.
See the course and the free previewThe fix is not more time. Book the hour, take the module you stalled on, and apply it to one real task in the week that follows. That single application is what the next module assumes you have done.
Pick up where you stopped