Two piles, and the line between them
Teaching splits into two piles that look similar on a timetable and are completely different underneath.
The first pile is production. The parent email you have rewritten four times to sound firm but warm. The worksheet that needs a lower reading age version by Thursday. The rubric you never quite got round to writing properly. The first pass at report comments for thirty-one students. This work is real, it is exhausting, and almost none of it requires the professional judgement you trained for. AI is good at it, and it can hand you back genuine hours.
The second pile is judgement about a particular child. The grade that goes on a record. A safeguarding concern. A decision about additional support or an access arrangement. A reference that follows a student into their next school or their first job. These are consequential, they are personal, and they belong to you.
The whole course rests on one principle: AI drafts, the teacher decides, and the teacher must be able to explain any decision in their own words.
That last clause is the test. If you cannot explain a grade, a comment or a referral to a parent without saying that a tool produced it, you have crossed the line. Not because a policy forbids it, but because you have made a decision about a child that you cannot account for.
If your school already has an AI policy, it wins over anything here. Read it before you use any tool with student work in it. If your school has no policy yet, treat that as a reason to be more careful, not less.
In plain English
- Prompt:
- The instructions you type. Vague instructions produce vague output, which is most of why AI feels useless the first few times.
- Hallucination:
- Confidently invented content. A quotation, a date, a source or a curriculum reference that simply does not exist.
- Differentiation:
- Adapting the same material so different students can access it. One of the tasks AI genuinely speeds up.
- Formative feedback:
- Comments meant to improve the next attempt, as opposed to a grade that records a judgement.
What it genuinely takes off you
Rewriting a message you have already decided the content of. Producing three versions of a text at different reading ages. Turning your marking criteria into a rubric with described levels. Drafting starter questions, exit tickets or retrieval quizzes on a topic you know well. Summarising a long policy or a research paper into something usable on a Tuesday evening.
What these share is that you already hold the professional decision. The tool is doing the typing, not the thinking. If it gets something wrong you spot it immediately, because you know your subject and you know your class.
Draft a short email to a parent. I will edit it before sending.
Situation: [two or three sentences, no names]
What I need the parent to know: [one sentence]
What I am asking them to do: [one sentence]
Tone: calm, factual, warm but not apologetic
Rules: under 180 words. No jargon. No blame. Do not promise anything
I have not listed. Leave square brackets where a name or a date is needed.
What must stay with you
Any grade that goes on a record. Anything to do with safeguarding, ever, including talking a worry through with a chatbot. Decisions about special educational needs, access arrangements or support plans. References. Anything you would be asked to justify in a meeting with a parent.
Three practical reasons hold this line up.
You cannot explain it. These tools produce fluent reasoning that sounds convincing and may bear little relation to how the output was actually produced. Repeating it as your own is a costume, not an account.
It goes wrong quietly. One misread paragraph is a bad afternoon. The same flawed rule applied to a whole year group is a pattern, and patterns are what get discovered later.
Students and parents can tell. Being processed rather than taught is its own injury, separate from whether the outcome was right.
Never put a safeguarding concern, a named student's personal circumstances, or anything from a support plan into a general AI tool. Most of these tools are not school systems and you do not control where the text goes. Use your school's proper channel, which is a person.
โ Weak prompt
Prompt
Here are 30 pieces of student writing. Give each one a grade out of 9 and write the report comment.
Output
Thirty grades and thirty comments, produced in under a minute, all fluent and all plausible.
Fast and indefensible. You did not read the work, so you cannot defend a single grade, and the comments describe a generic student rather than the one sitting in front of you.
โ Good prompt
Prompt
Here is my marking criteria for a Year 9 persuasive writing task. Turn it into a rubric with four described levels per criterion, written plainly enough that two teachers would apply it the same way. Do not grade anything.
Output
A rubric with clear level descriptors and observable evidence for each band.
The tool improved the instrument. You still read the work and set every grade, and you can point at the criterion and the evidence when anyone asks why.
Checkpoint
AI drafts, the teacher decides. If you cannot explain a decision about a student in your own words, without mentioning a tool, you should not be making it that way.
The test to apply on Monday
Three questions before you use AI for anything.
Is a specific student the subject of this output? If yes, slow right down.
Would I be comfortable showing this prompt and this output to that student's parent? Discomfort is information.
If this is questioned in a year, can I explain the basis of it without mentioning software? If not, redesign the step.
I am thinking of using AI for this teaching task:
[describe it in two or three sentences]
Tell me:
- Whether any output would be about a specific named student
- What could go wrong for that student if the output is wrong
- Which parts I must do myself rather than merely review
- What I should keep so I can explain the decision later
Be blunt. Do not reassure me. If I should not automate this at all,
say that first and explain why.
Turn my marking criteria into a rubric.
Year group: [ ]
Task: [ ]
Criteria: [paste yours]
Four levels per criterion. Describe what the work looks like at each level
using observable features, not adjectives like good or excellent.
Add one sentence per criterion telling a student what to do to move up a level.
Do not invent criteria I have not given you.
Keep the prompt next to anything you reuse. Six months later, the prompt is usually the only thing that explains why the worksheet or rubric looks the way it does.
A note on being sceptical
If you have decided AI has no place in your classroom, that is a professional judgement, not a failure to keep up. Plenty of the best teaching happens with a whiteboard and a hard question. This course is not an argument that you should use these tools more. It is here so that whatever you decide, you decide it knowing what the tools actually do and where they actually fail.
๐ Quiz
Question 1 of 4What principle does this course rest on?