@MrMalbon@OECDEduSkills I'd add: "Which sentence in your source supports the claim, and why?" Naming a source is a useful start; explaining the connection could help distinguish checking from simply adding a plausible reference.
If you're using AI to adapt a GCSE science question, ask it to simplify the wording without explaining the science students need to work out. Then check the result: has it made the question easier to read, or quietly supplied part of the answer?
@TeacherToolkit For secondary teachers, I'd pair the attention cue with a quick check of the instruction: ask a student to explain the first step before everyone starts. A quiet room can still contain several different ideas about what the task actually requires.
An interesting finding from our recent AI marking study:
When TeachEdge disagreed with published examiner marks, the differences often came down to evaluation rather than subject knowledge.
How well had students developed their arguments?
How much credit should a final judgement receive?
It reminded me how much professional judgement goes into assessment.
AI marking isn't just about recognising correct answers. It's about assessing the quality of thinking.
@pj_schreiner I'd keep the reading reflection but build in a short revision after class discussion: which claim would you now change, and what in the text or discussion changed your mind? That gives students a reason to revisit their thinking, rather than just submit a finished answer.
@TeacherToolkit I'd protect a follow-up meeting before adding another CPD priority. Bring a few student answers showing how students used the feedback, then agree what to keep or change. That makes implementation part of the timetable, rather than an extra.
@PhdDfs I'd want to ask them....were they struggling to understand the notes, or has asking a chatbot become the first step even when the answer is in front of them?
When I was teaching A Level Economics, AI helped me plan a lesson on bond prices. The first version showed what happened, but not why. Pushing back took longer than planning alone, yet clarified the explanation. For teachers, speed isn't always the most useful test.
@mbateman I think it's getting harder to justify paying for individual teaching resources when AI can help teachers create and adapt them to their own classes. There's still checking and refining to do, but that applies to a worksheet you've bought too.
For teachers reviewing AI-drafted feedback, three checks before sharing:
1. Is each judgement supported by the student's answer?
2. Does it fit the task and mark scheme?
3. Is the next step clear enough to act on?
If any answer is no, revise it.
This was a small pilot conducted by us, not an independent study.
We've published all 67 results, our methodology and the shortcomings. I think it's important to be open about what AI marking gets right AND wrong.
Here's the blog, with a link to the full research:
https://t.co/ILWJbcIn1L
Can AI really mark an Economics essay as well as an experienced examiner?
Having spent years teaching A-Level Economics (and marking hundreds of essays!), I was keen to find out.
So we put TeachEdge to the test on 67 real examination answers.
The results surprised me. 🧵
Of course, TeachEdge didn't always agree with the examiner.
It sometimes marked weaker answers too generously and strong essays too harshly.
Evaluation was a particular challenge.
As any Economics teacher knows, spotting a weakness in an argument isn't the same as properly evaluating it!
These findings are useful for improving the marking.
For secondary teachers using AI to draft quizzes, check the wrong answers as carefully as the right one. In GCSE science, an option might reveal confusion between mass and weight. If it is just obviously wrong, choosing correctly tells you less.
Teachers using AI for lesson planning: decide what "good enough" looks like. If a usable worksheet becomes five extra versions, the time saved has gone straight back into work.
A useful department discussion is what to stop doing, as well as what AI can help with.
In an A Level Economics answer, “develop your analysis” is hard to act on.
“Explain how higher interest rates could change household spending, then consider who might benefit” gives the student a route forward.
Useful feedback names the missing thinking.
Teachers using AI for feedback: try this quick check.
Could this comment have been written without reading the student’s answer?
If yes, it needs work.
Useful feedback refers to what the student actually wrote and gives them a clear next step.
#AIinEducation