Course

AI for Assessment & Feedback

Design valid, fair assessment tasks and give faster, fairer feedback — with your professional judgement in front, and never a single student's data in a general tool.

Modules

Each module: clear learning outcomes → short, accurate input grounded in the NSW DoE assessment principles and NESA integrity → a hands-on activity using the Lessio assessment generator → interactive knowledge checks. Mapped to the Australian Professional Standards for Teachers.

Click a module to read it.

1. 1

   What assessment AI can and can't do

   Validity, reliability and the on-balance judgement — and the clear line between where AI helps (drafting tasks, rubrics, feedback frames, question banks) and where it must never go (knowing the student, the grade decision).~45 min

   By the end of this module you'll be able to:

   • Define validity and reliability and explain why each matters for an assessment task.
   • Explain why an A–E grade is an on-balance judgement against the NESA Common Grade Scale — and why that decision can never be delegated to AI.
   • Sort assessment work into 'AI may draft' versus 'teacher only', and justify the split.

   Standards2.3 Curriculum, assessment and reporting5.1 Assess student learning5.3 Make consistent and comparable judgements

   Start from what assessment is for

   Before any tool, hold the purpose. Assessment in NSW does two jobs: it gathers evidence of learning to inform an on-balance judgement, and — done as assessment for/as learning — it makes explicit to students where they are, where they are going, and how to get there. AI can help you produce the materials for both. It cannot do the judging, and it does not know your students.

   Two words that govern everything: validity and reliability

   Concept	Plain meaning	The question you ask	Where AI helps / can't
   Validity	The task measures what it intends to measure — the actual syllabus outcome, not reading speed, not luck.	"Does this task give a student a fair chance to show this outcome?"	AI can draft a task and flag alignment; you confirm it measures the outcome.
   Reliability	Results are stable and consistent across markers, classes and occasions.	"Would another teacher mark this the same way?"	AI can draft clear marking guidelines; moderation between teachers delivers reliability — AI cannot.

   A task can be reliable but not valid (consistently measuring the wrong thing — e.g. testing comprehension when the outcome is reasoning). Validity comes first.

   The on-balance judgement — and why it stays human

   An A–E grade is not an average of marks. It is an on-balance judgement of a student's achievement against the NESA Common Grade Scale, drawn from multiple formative and summative tasks across the period. It weighs the whole picture of a learner. That is a professional, accountable act — and under the Australian Framework for Generative AI in Schools, accountability stays with the teacher. An AI has no standing to decide a child's grade, and "the AI suggested a B" is never an answer to a parent.

   Where AI genuinely helps assessment

   • A first draft of a task — beating the blank page, then shaped by you.
   • A draft set of marking guidelines or a rubric you align and own.
   • Feedback frames — a strength / a priority / a next step against your success criteria.
   • Question banks — graduated items, MCQs with distractors, retrieval-practice sets (always verified).

   Where it must stay out — non-negotiable

   • The grade decision and the on-balance judgement.
   • Knowing the student — what is fair for this child, what happened in class, the reasonable adjustment they need.
   • Any identifiable student work in a general tool (Module 4 makes this concrete).

   The principle that runs through this course: AI drafts; you align, verify, judge and own. The teacher stays the assessor and the author.

   Activity — draw your own line (10 min)

   Open the Lessio generator and produce an assessment task (task + marking guidelines) for a unit you teach. Now draw a line down a page: on the left, list what the tool drafted for you (the saved time); on the right, list every decision that is still yours — does it measure the outcome? is it fair? what grade does the evidence support? Notice that everything on the right is professional judgement, and none of it is going away.

   Knowledge check

   1A multiple-choice quiz gives identical results every time you run it, but the outcome it's meant to assess is 'analyses and evaluates'. Is it reliable? Valid? What's the lesson?

   Show answer

   2Why can an A–E grade never be handed to an AI to decide?

   Show answer

   3Give one assessment job AI may draft and one it must never do.

   Show answer

   Mark this module complete
2. 2

   Designing valid, fair tasks with AI

   Write tasks in plain English so the most students can access them independently; align marking guidelines to the success criteria AND the syllabus outcomes; build in reasonable adjustments — and verify every outcome code against the official syllabus.~60 min

   By the end of this module you'll be able to:

   • Draft an assessment task in direct, plain English so the greatest number of students can access it independently.
   • Produce marking guidelines that reflect the success criteria AND the syllabus outcomes, using active, observable verbs.
   • Build in reasonable adjustments for students with disability and verify every outcome code against the official NESA syllabus.

   Standards2.3 Curriculum, assessment and reporting5.1 Assess student learning1.6 Strategies to support students with disability

   A fair task starts with plain English

   The NSW DoE assessment principles are clear: write tasks in direct, plain English so the greatest number of students can access them independently. A task no one can read is invalid before a student starts — you end up measuring reading comprehension, not the outcome. So:

   • Short sentences, active voice, one instruction per line.
   • Command words students know (describe, explain, justify, calculate).
   • Define or remove unnecessary jargon; keep only the subject vocabulary the outcome requires.

   AI is excellent at a plain-English pass — but you confirm it didn't quietly drop the rigour the outcome needs.

   Alignment — the spine of a valid task

   A valid task and its marking guidelines line up with two things at once:

   1. the success criteria you'll share with students, and
   2. the syllabus outcomes the task assesses.

   The DoE principle: marking guidelines must directly reflect the success criteria and the syllabus outcomes, and align to the stage's level of achievement, using active observable verbs. "Shows good understanding" is not markable; "correctly applies the cosine rule to find an unknown side" is. Make AI draft to observable verbs, then check every criterion ties back to a real outcome.

   Quick alignment test: for each mark a student can earn, can you name the outcome it evidences and the success criterion it matches? If not, the guideline isn't ready.

   Reasonable adjustments — fair and legal

   Fair access for students with disability is not optional — it is required under the Disability Standards for Education 2005. Adjustments (extra time, a reader/scribe, a simplified instruction format, an alternative mode of response) must keep the task targeting the same outcome at the same standard — they change the access, not the learning. AI can quickly draft an adjusted format; you confirm it still assesses the outcome and suits the actual student (described as a de-identified need, never named).

   Verify the outcome codes — a confident code can be wrong (this really happened)

   A generic AI-generated Year 10 Mathematics program labelled the sine rule, cosine rule and area of a non-right-angled triangle as MA5-TRG-C-02, a Core outcome.

   It's wrong. In the NSW Mathematics K–10 syllabus that content is a Path outcome — MA5-TRG-P-01. MA5-TRG-C-02 is actually bearings, and angles of elevation and depression. The code looked real and sat inside a real-looking program — and was completely mislabelled. If that error rode into an assessment, you'd be marking the wrong outcome and potentially examining Path content as Core.

   The discipline, whatever the tool: verify every outcome code against the official NESA syllabus before the task goes anywhere. Lessio grounds its drafts in the verbatim syllabus to reduce exactly this slip — but the check is still yours.

   Use Lessio to build the whole task

   Lessio's assessment output is a task + marking guidelines drafted to the real outcomes and the DoE standards. That gives you an aligned starting point; your job is the plain-English polish, the adjustments, the code check, and making it fair for your cohort.

   Activity — draft, align, adjust, verify (15 min)

   Using the Lessio assessment generator, draft a task + marking guidelines for an outcome you teach. Then do four things by hand: (1) run a plain-English pass so a struggling reader could access it independently; (2) check each marking criterion names an outcome and matches a success criterion; (3) add one reasonable adjustment that keeps the same outcome; (4) verify the outcome code against the official NESA syllabus. The edits you make are the evidence of your professional judgement.

   Knowledge check

   1Why does writing a task in plain English protect its *validity*, not just its readability?

   Show answer

   2A marking guideline reads 'demonstrates good understanding of forces'. What's wrong, and how do you fix it?

   Show answer

   3An AI task labels the cosine rule as MA5-TRG-C-02. Why is that dangerous in an assessment, and what do you do?

   Show answer

   Mark this module complete
3. 3

   Marking guidelines, rubrics & consistency

   Turn marking guidelines into student-friendly 'I can…' criteria; align band descriptors to the NESA Common Grade Scale; and use moderation — not an algorithm — to get consistent teacher judgement. AI drafts the rubric; it never marks.~45 min

   By the end of this module you'll be able to:

   • Convert marking guidelines into student-friendly 'I can…' success criteria without losing the standard.
   • Draft band descriptors aligned to the NESA Common Grade Scale using active, observable language.
   • Explain how teacher moderation delivers consistency — and why AI may draft a rubric but must never be the marker.

   Standards2.3 Curriculum, assessment and reporting5.3 Make consistent and comparable judgements5.1 Assess student learning

   From marking guidelines to 'I can…' criteria

   Students do better when they can see the standard before they start. Translate your marking guidelines into student-friendly success criteria — short, concrete 'I can…' statements — while staying faithful to the original standard. This is assessment as learning: it shows students where they are going. AI is genuinely good at this rewrite; you check it didn't soften the rigour.

   Marking guideline (teacher)	Student-friendly criterion
   "Selects and justifies an appropriate method, with accurate working."	"I can choose a sensible method, show my working, and explain why I chose it."
   "Analyses the effect of technique on meaning, supported by textual evidence."	"I can explain how the writer's techniques change the meaning, using quotes as proof."

   Band descriptors and the Common Grade Scale

   For A–E reporting, descriptors should align to the NESA Common Grade Scale — the standard reference for what an A, B, C, D or E performance looks like. Draft band descriptors that:

   • use active, observable verbs (what the student does, not how 'good' it is),
   • describe a coherent picture at each grade (not a checklist of marks), and
   • align to the stage's expected level of achievement.

   AI can draft a five-band descriptor set quickly; you align it to the Common Grade Scale and your task, and confirm each band genuinely steps up from the one below.

   Consistency is a human process — moderation

   Reliability between markers doesn't come from a tool; it comes from moderation: teachers marking common samples together, agreeing the standard, and checking their judgements are comparable across classes and markers. This is professional, collaborative, on-balance work — and it's exactly the part an algorithm can't do for you.

   AI as a drafting aid, never the marker. Let AI draft the rubric and the student-facing criteria. Do not let it allocate the grade — that is the consistent, comparable, accountable judgement moderation exists to protect.

   A useful AI move for moderation: ask it to generate a discussion starter — borderline scenarios, or questions that surface where two teachers might disagree — to sharpen a moderation meeting. It primes the conversation; the teachers reach the agreement.

   Where AI can mislead you on rubrics

   • It may invent band boundaries that sound authoritative but don't match the Common Grade Scale — align them yourself.
   • It may produce descriptors that are vague ("good", "sound") rather than observable — push for verbs.
   • It cannot know whether this student's response sits in this band — only you, in moderation, can.

   Activity — rubric to criteria, then a moderation starter (12 min)

   Using Lessio's assessment output (or a rubric you already have), have AI draft student-friendly 'I can…' criteria for your task — then check each against the original standard. Next, ask AI (in NSWEduChat or your tool) for a moderation discussion starter: three borderline 'is this a C or a B?' scenarios for your task. Bring those to your next faculty marking conversation.

   Knowledge check

   1What is the danger when AI rewrites your rubric as student-friendly 'I can…' criteria?

   Show answer

   2Where does consistency between markers actually come from, and what does that mean for AI?

   Show answer

   3When you draft band descriptors with AI, what two things must you verify?

   Show answer

   Mark this module complete
4. 4

   Feedback that moves learning

   Generic feedback frames versus genuinely personalised feedback; assessment-for-learning (where the student is, where they're going, how to get there); de-identification as the hard line; and feedback that is both warm and specific.~45 min

   By the end of this module you'll be able to:

   • Structure feedback around the three assessment-for-learning questions: where the student is, where they're going, how to get there.
   • Use AI to draft a feedback frame, then personalise it with warmth and specificity for the actual learner.
   • Apply the de-identification rule so no identifiable student work ever enters a general AI tool.

   Standards5.2 Provide feedback to students on their learning4.5 Use ICT safely, responsibly and ethically1.5 Differentiate teaching to meet specific learning needs

   Good feedback answers three questions

   Feedback moves learning when it makes three things explicit — the heart of assessment for learning:

   1. Where is the student now? (against the success criteria, honestly)
   2. Where are they going? (the next standard / the outcome)
   3. How do they get there? (one or two specific, actionable steps)

   A grade alone does none of this. AI can help you produce feedback that does all three — fast — without you writing the same scaffold thirty times.

   Frames vs personalised feedback — use both, knowingly

   • A generic feedback frame is a reusable structure: one genuine strength · one priority for improvement · one specific next step, tied to your success criteria. AI is excellent at drafting frames, and they save real time.
   • Personalised feedback is that frame made true for this child — naming what they actually did, in language that lands for them. This is the part only you can finish, because only you know the learner.

   The workflow: AI drafts the frame → you personalise and verify. Never the reverse.

   The hard line: never paste identifiable student work into a general tool

   This is the rule that protects you and your students:

   Never enter a student's identifiable work or personal information into a general AI tool.

   No names, student/NESA numbers, nor a response that could identify the child. This is the Privacy and Personal Information Protection Act 1998 (NSW) and the Child Safe Standards — not etiquette. Your safe options:

   • De-identify — strip names and identifying detail; feed only the de-identified response or, better, just the success criteria and have AI build a frame you apply yourself.
   • Use your school's approved, secured environment where policy permits (NSWEduChat is department-built and secured and allows limited personal information; a public chatbot never does).
   • Lessio sidesteps it entirely — it is built student-data-free: you describe a cohort or a task, never a child.

   When in doubt, generate the frame from the criteria, not the comment from the child's work.

   Warmth + specificity — both, or it doesn't work

   Warm-but-vague ("Great effort!") doesn't move learning; specific-but-cold can deflate. Effective feedback is both: it names a real strength, is honest about the priority, gives a concrete next step, and is addressed to the student in a tone that keeps them in the game. Tell AI to be warm, specific, jargon-free — and to invent nothing that isn't in the response (no fabricated quotes, no praise for work that isn't there).

   A caution

   AI feedback can sound fluent and generic — the same three comments for every student. Personalisation is the antidote, and it's the professional move: the frame is the time-saver, you are the teacher the student hears.

   Activity — frame from criteria, then personalise (12 min)

   Take a set you're marking. Give AI only your success criteria (not an identifiable student's work) and ask for a feedback frame: a strength, a priority, a next step, addressed to the student, warm and specific, inventing nothing. Then take that frame and personalise it for two de-identified students — naming what each actually did. Notice how the frame saved the scaffolding and your judgement supplied the meaning.

   Knowledge check

   1What three questions should feedback make explicit, and what's the name for this purpose of assessment?

   Show answer

   2You want AI to help with feedback but the work identifies the student. What do you do?

   Show answer

   3Why isn't an AI-drafted feedback frame enough on its own?

   Show answer

   Mark this module complete
5. 5

   Integrity & authorship in the AI age

   Schools decide if GenAI is permitted per task; NESA malpractice (plagiarism, collusion, misrepresentation); assuring authorship by DESIGN — drafts, checkpoints, discussion, self-reflection — not unreliable 'AI detectors'; and teaching students to use AI with integrity, with HSC/RoSA in view.~45 min

   By the end of this module you'll be able to:

   • State the school's task-by-task position on GenAI and what NESA counts as malpractice.
   • Design an assessment that assures authorship through process, rather than relying on 'AI detectors'.
   • Teach students to use AI with integrity, with HSC/RoSA implications in view.

   Standards5.1 Assess student learning4.5 Use ICT safely, responsibly and ethically7.1 Meet professional ethics and responsibilities

   Schools decide — task by task

   There is no single rule that GenAI is always banned or always allowed. Schools decide whether generative AI is permitted for a given assessment task — and the task instructions must make that explicit to students. Some tasks will welcome AI as a drafting tool to be disclosed; others (especially those evidencing an individual's unaided skill) will prohibit it. Your first integrity move is to state the rule on the task.

   What NESA counts as malpractice

   For HSC and RoSA, integrity is paramount. Malpractice includes:

   • Plagiarism — presenting others' (or AI's) work as one's own.
   • Collusion — improper collaboration passed off as individual work.
   • Misrepresentation — submitting AI-generated work as the student's own thinking.

   Malpractice can jeopardise a student's award. Students need to know this plainly — and the line between permitted, disclosed AI assistance and misrepresentation must be unambiguous on every task.

   Assure authorship by design, not by detection

   The temptation is to reach for an "AI detector". Don't rely on them — they are unreliable, producing false positives (penalising innocent students, disproportionately some EAL/D writers) and false negatives. NESA's guidance is to assure authorship over time through how you design the task:

   • Drafts and checkpoints — collect planning, drafts and in-progress work, so you see the thinking develop.
   • Discussion / questioning — a short viva or conference where the student explains their choices.
   • Self-reflection — students account for their process and any AI use.
   • In-class components — supervised writing or problem-solving that anchors authorship.

   Design tasks where the process is visible. If you can see the work develop and the student can explain it, authorship is assured far more reliably than any detector could claim.

   Teach students to use AI with integrity

   Students will use AI regardless — so the ethical task is to teach them to use it honestly: as a tutor and drafting aid they critique, improve on, and disclose — not a shortcut that does their thinking. Be explicit about when AI is allowed for a task, how to acknowledge it, and why misrepresentation risks their results. Your own honest, disclosed practice (Modules across this course) sets the norm and builds the AI literacy the Australian Framework expects.

   Keep HSC/RoSA in view

   For senior and credentialled work, the stakes are formal. Build authorship assurance into the design of major tasks from the start, make the AI rule explicit, and document process. Protecting a student's award is part of the job.

   Activity — redesign one task for authorship (12 min)

   Take a real assessment task. First, write the AI rule onto it (permitted? for what? how disclosed?). Then redesign it to assure authorship by design — add at least one of: a drafted checkpoint, a short explain-your-thinking conference, or an in-class component. Use Lessio's assessment output to draft the revised task + marking guidelines, then add the integrity scaffold yourself. Note how the redesign makes the process visible without resorting to detection.

   Knowledge check

   1A student used AI on a task where your instructions didn't say whether that was allowed. Whose gap is that, and what's the fix?

   Show answer

   2Why shouldn't you rely on an 'AI detector' to catch misuse, and what does NESA suggest instead?

   Show answer

   3Name the three forms of malpractice that can jeopardise a student's HSC/RoSA award.

   Show answer

   Mark this module complete
6. 6

   Capstone — build, critique & log it

   Build a full assessment (task + marking guidelines) in Lessio, critique it for validity, fairness and integrity, self-assess against the Ethical-Use Checklist, and log it as PD.~50 min

   By the end of this module you'll be able to:

   • Build a complete assessment — task and aligned marking guidelines — with AI, end to end.
   • Critique it for validity, fairness and integrity, and improve it with your professional judgement.
   • Self-assess against the Ethical-Use Checklist, reflect, and record the hours as Standards-relevant PD.

   Standards2.3 Curriculum, assessment and reporting5.1 Assess student learning6.2 Engage in professional learning

   The task — a real, defensible assessment

   Choose an outcome you'll assess next term. Using the Lessio assessment generator, build:

   1. an assessment task written in direct, plain English, and
   2. its marking guidelines — aligned to the success criteria and the syllabus outcome, in active, observable verbs.

   Then improve it with your professional judgement so you could defend it in a faculty meeting or a registration audit.

   Critique it on three fronts

   • Validity — does the task actually measure the intended outcome? Do the marking guidelines tie each mark to that outcome? Is every outcome code verified against the official NESA syllabus (remember MA5-TRG)?
   • Fairness — can the greatest number of students access the wording independently? Have you built a reasonable adjustment that keeps the same outcome (Disability Standards 2005)? Are contexts and examples inclusive and culturally safe?
   • Integrity — is the AI rule explicit on the task? Does the design assure authorship (a draft, a checkpoint, a conference, or an in-class component) rather than lean on detection?

   What good looks like

   A connected, syllabus-accurate assessment you'd genuinely use — drafted by AI, unmistakably shaped and owned by you. Your edits — the code check, the plain-English pass, the adjustment, the authorship scaffold — are the visible evidence of your professional judgement and of teacher-in-the-loop practice.

   Self-assessment — the Ethical-Use Checklist

   Run your capstone against all five checklist items on this page. Every box should be honestly tickable. If one isn't, fix the artefact — that is the learning.

   Reflection — write a short response

   • What did AI genuinely save you time on, and what did you have to fix?
   • Where did your professional judgement change the task or the marking guidelines?
   • How did you assure authorship without relying on detection?
   • One rule you'll keep for using AI in assessment and feedback from now on.

   Log it as professional learning

   This module is your assessment: a complete, critiqued task + marking guidelines, plus your ethical-use reflection — keep it as evidence of practice. Since NESA removed the Accredited/Elective PD distinction in 2024, Standards-relevant learning like this counts toward your 100 maintenance hours — record it in your eTAMS PD log against the Standards it addresses (especially Standard 5, plus 2.3, 4.5 and 7). Your school can also run the course as part of its professional-learning plan or a staff development day.

   Activity — build, critique, self-assess, log (15 min)

   Generate your assessment (task + marking guidelines) in Lessio, run the three-front critique above, tick it against the Ethical-Use Checklist, write your four-point reflection, and log the hours in eTAMS. Keep the artefact and reflection together as your evidence.

   Knowledge check

   1What turns an AI-generated assessment into defensible professional work?

   Show answer

   2Your capstone fails one item on the Ethical-Use Checklist. What should you do, and why?

   Show answer

   3How does this course count toward your NESA professional-development hours?

   Show answer

   Mark this module complete

Take-away prompt library

Ready, RICE-shaped prompts for common NSW teaching jobs (Module 3). De-identified — copy one, swap in your details, and use it today.

You are a NSW [stage] [subject] teacher. Rewrite the marking guidelines below as student-friendly success criteria ('I can…' statements) for a de-identified [stage/year] class, staying faithful to the original standard and outcome — do not make any criterion easier or vaguer than the source. Use plain English and active verbs. Then, for the top criterion, add one short example of what 'achieved' looks like. Finally, list any place you were unsure you'd kept the original standard, so I can verify it.

[paste marking guidelines]

You are a NSW [subject] teacher. Using ONLY these success criteria [paste], write a reusable feedback frame I can personalise for each student: one genuine strength, one priority for improvement, and one specific, actionable next step — structured around where the student is, where they're going, and how to get there. Warm, specific, jargon-free, addressed to the student. Do NOT ask for or assume any identifiable student work, and invent no quotes or content. Flag anything you're unsure of so I can check it.

You are a NSW [stage] [subject] teacher and an accessibility reviewer. Review the de-identified assessment task below for plain-English accessibility so the greatest number of students can access it independently: flag long or complex sentences, unnecessary jargon, and ambiguous instructions, and suggest a clearer rewrite for each — WITHOUT lowering the rigour the outcome requires (tell me if any suggestion risks that). Use Australian spelling. List anything you couldn't be sure about so I can verify it.

[paste task]

You are a NSW [subject] teacher. Write eight questions on [topic] for a de-identified [stage/year] class, graduated from recall to application, each tagged with the skill it assesses, plus a clear answer key. Plain English, Australian context. Then FLAG any question whose answer could be ambiguous or whose wording could mislead, and flag anything you couldn't verify, so I can check it before use.

You are a NSW [subject] teacher. Write six multiple-choice questions on [topic] for a de-identified [stage/year] class. For each: one correct answer and three plausible distractors that each reflect a specific likely misconception (briefly name the misconception each distractor targets), and indicate the correct option. Plain English, Australian context. Note any item where more than one option could be defensible, and flag anything you're unsure of so I can verify it.

You are a NSW [subject] teacher facilitating a marking-moderation meeting. Using the de-identified task and marking guidelines below, write three short borderline scenarios (e.g. 'is this a C or a B?') and five questions that would surface where two markers might disagree, so we can agree a consistent standard against the NESA Common Grade Scale. Do NOT assign grades yourself — the teachers will decide. Flag any part of the guidelines that is vague or hard to apply consistently so we can sharpen it.

[paste task + marking guidelines]

Standards-relevant professional learning, mapped to the APST · content verified against national and NSW frameworks, June 2026 · self-log the hours in eTAMS.