Teacher PD · Current NAS Jiaxing edition
Build something useful. Check it properly. Keep student data inside the right boundary.
Archive refresh, not archive republishing
Removed: “magic prompt” language, speed promises and claims that AI can automatically halve grading time.
Added: explicit data boundaries, a review gate, school-managed/local-AI context and clearer separation between drafting and high-stakes judgement.
Kept: the practical two-hour format, hands-on production, differentiation, assessment drafting and a take-away workflow teachers can reuse.
Workshop map
Facilitate this as a 100-minute core session, expand discussion and practice to fill two hours, or use each section independently for shorter department PD.
Decide what belongs in a public AI tool, a school-managed system, or nowhere in AI.
Turn a vague request into a usable teaching brief with context, constraints and a checking step.
Create and improve one real lesson resource without sharing identifiable student information.
Check accuracy, level, answer keys, inclusivity and whether the output actually fits the lesson.
Use AI to draft low-risk assessment materials while keeping high-stakes decisions human-led.
Choose one workflow to use this week and record what needs changing after classroom use.
Module 1 · Set the boundary
These zones are a practical workshop aid, not a substitute for school policy. Where a school rule is stricter, the school rule wins.
Usually suitable for an approved general-purpose AI service, subject to school rules.
AI can assist, but the teacher remains responsible for the final material or judgement.
Use established school procedures and approved systems instead of convenience.
Take one task you genuinely need to do. Remove names, IDs, private comments and anything else the AI does not need. Replace real student details with neutral descriptors such as “Grade 8, approximately B1 reading level” or use fictional examples.
Module 2 · Brief AI well
Give enough context to define the teaching problem, state what the output must do, constrain it, then require an explicit check or revision step.
Class, approximate level, topic, time and what students already know — without identifiable data.
State the precise output: worksheet, explanation, questions, model answer, examples or lesson sequence.
British English, length, CEFR range, task types, vocabulary limits, formatting and what to avoid.
Ask the model to flag uncertainty, verify its answer key, explain assumptions or revise against a rubric.
Reusable teacher brief
I teach [grade / approximate language level]. We are working on [topic or skill]. Students already know [relevant prior learning]. Create [specific resource] for a [duration] activity. It must: - use British English; - match approximately [CEFR / difficulty]; - include [required features]; - avoid [unwanted features]. Before presenting the final version, check that: 1. the instructions are unambiguous; 2. every answer in the key is actually supported; 3. the difficulty matches the level I gave you; 4. any factual claim that may be uncertain is clearly flagged. Then give me the classroom-ready version only.
Start with something weak such as “make me a lesson on advertising”. Add the real learning goal, age/level, lesson time, expected student output and a checking instruction. Compare the two drafts.
Module 3 · Build
The most useful classroom workflows usually involve a teacher asking for a draft, changing the constraints after seeing it and then manually selecting or editing the parts worth keeping.
Create the same core task at two language-support levels. Compare whether the learning objective stayed the same.
Generate a short mixed review from teacher-supplied concepts, then remove ambiguous items and rebalance difficulty.
Ask for contrasting examples that help students discriminate a grammar form, writing feature or concept.
Create a deliberately imperfect fictional response, then use it for annotation, correction or peer discussion.
Draft several versions of one question type so the teacher can choose the clearest rather than accepting the first output.
Generate two or three lesson sequences under the same constraints and compare trade-offs before selecting one.
Produce a small, testable resource rather than an entire unit: for example a starter, reading task, vocabulary review, model paragraph, discussion sequence or exit ticket. Save the first draft so you can see what changed after review.
Module 4 · Review & test
AI can confidently produce plausible mistakes. A polished layout or fluent explanation is not evidence that the content is correct or appropriate.
Are facts, grammar points, calculations, names and references correct?
Does this match the actual lesson objective, age group, language level and available time?
Do answer keys, distractors and examples genuinely work? Test them rather than assuming.
Can students understand the instructions without the AI conversation that produced them?
Is the language, layout and task load usable for the students who will receive it?
Have unnecessary copyrighted passages, images or unattributed material been avoided?
Did the workflow avoid exposing student or school data beyond the approved boundary?
Current NASJX case study
NASJX resource production has used separate generation, deterministic checks, model review and human adjudication rather than treating one AI answer as “finished”. Teachers do not need a multi-model pipeline for every worksheet, but the principle transfers: generate, inspect, test, correct and preserve the final human decision.
Module 5 · Assessment
Drafting question banks, rubrics and low-stakes practice can be productive. Final high-stakes judgements, sensitive feedback and decisions affecting students require the appropriate human and school process.
Public cloud, school-managed and local AI
A public AI service can be convenient for low-risk general drafting when permitted. School-managed or local systems can provide stronger organisational control for approved workflows as those systems are connected. “Local” does not automatically make a workflow safe; permissions, logging, access and data handling still matter.
Use only within school rules and without data the service does not need.
Use when the school has approved the service and its handling of the relevant data/task.
Useful for future protected workflows, but still requires authentication, permissions and governance.
Module 6 · Make it repeatable
Choose one recurring task. Define what AI does, what you check, what data is allowed and what “good enough to use” means. After one classroom cycle, record what saved effort and what created extra correction work.
Something you already do repeatedly.
Decide what information may enter the tool.
Context, task, constraints and check.
Verify the resource before students see it.
Keep the workflow only if it genuinely improves the work.
Take it with you
The printable handout condenses the risk zones, prompt brief, review checklist and action plan into a cleaner format for printing or saving as PDF from the browser.