Start with the hub: ChatGPT for South African Teachers (CAPS).
What FET assessment actually is (before AI)
The National Protocol for Assessment Grades R-12 treats assessment as internal work (school-based assessment and, where relevant, practical assessment tasks) plus end-of-year examinations.
In the FET phase the Protocol’s table sets school-based assessment at 25% and the end-of-year examination at 75%. Recorded formal tasks feed the SBA mark used for promotion. Classroom assessment is supposed to collect evidence of learner performance.
DBE’s NSC page is blunt about the certificate: candidates must meet the subject minimums and “provide full evidence in the SBA component in the subjects offered”. Umalusi quality-assures Grade 12 assessment and issues the NSC.
Assessment in the NSC, on DBE’s PAT page, comprises school-based assessment, practical assessment tasks, language oral assessment for official languages, and external examinations.
A PAT is compulsory for listed subjects (among them Agricultural Management Practices and Technology; several arts; technology subjects; Life Orientation; CAT and IT; Consumer Studies, Hospitality and Tourism). Language orals are listed there too. DBE states that the PAT mark must count 25% of the end-of-year examination mark for those subjects. Use the current year’s PAT documents, not last year’s memory.
None of those documents is an AI policy. They still tell you whose work the mark is supposed to represent: the learner’s.
Traffic lights for FET tasks
| Task | Light | Why |
|---|---|---|
| Teacher drafts a test, memo, extra worksheet or lesson outline | Green | Prep aide. You still check weighting, cognitive demand and topic against CAPS, the ATP and your school assessment plan. |
| Informal class practice, clearly not for the SBA file | Amber | Allowed if learners know it is practice, and if you can run it without devices. Prefer paper. |
| Homework “research” with a chatbot, no process evidence | Red | You cannot tell whose sentences you are marking. |
| Formal SBA task (test, project, assignment that goes in the file) | Red unless the task sheet says how AI may be used | The Protocol is collecting evidence of learner performance. |
| PAT, oral, practical, performance, design process | Red for generating the submitted artefact | These exist because a written exam cannot capture the skill. Outsourcing the artefact defeats the task. |
| NSC / Grade 10-11 November exams and controlled tests | Red | External or controlled conditions. No phone, no chatbot. |
| Revision pack the teacher prints | Green for the teacher | Learners work on paper. |
If you want an amber SBA design, write it on the paper: “You may use ChatGPT for five minutes to list possible sources. You may not paste prose from it. Attach the prompt you used.” If you cannot moderate that, keep the task red.
What teachers may do (green)
Use the hub’s assessment prompt, then do the professional work the model cannot:
Draft a [20/30/50]-mark formal or informal assessment for Grade [10/11/12] [Subject], Term [Y], topic “[topic as named in my CAPS/ATP]”.
Mix cognitive demand appropriate to FET. Language: [HL / FAL].
Include a full memorandum with marks and acceptable alternatives.
Do not invent CAPS page numbers or official codes.
State that I must check weighting and task type against my CAPS/ATP and school assessment plan.
Use South African examples. No US curriculum names.Then:
- Open Chapter 4 of your subject CAPS and the current ATP.
- Check that the topic belongs in this term and grade.
- Check mark allocation and task type against the school programme of assessment.
- Strip overseas examples, dollars, yards, and fake “NSC-style” labels the model invented.
- For a formal task, keep a clean copy for moderation in the teacher file.
You may also ask ChatGPT for extra practice items after a topic is taught, or for a plain-language recap. That is still teacher-side. Do not paste a real learner’s script into the tool to “mark it”. Describe the error in general terms, or anonymise until nothing identifying remains.
Marking that needs human judgement stays human. Auto-marking higher-order FET work with a chatbot is not a shortcut. It is a different, weaker assessment.
What learners may not do (red, unless the task says otherwise)
- Submit chatbot prose as an essay, history investigation, literature essay, or Life Orientation project.
- Generate a PAT report, tourism plan, CAT/IT artefact write-up, or practical design and hand it in as their own.
- Use ChatGPT in a controlled test or exam venue.
- Invent sources. Models fabricate references. That is an integrity problem even when the prose looks fluent.
AI-detector scores are not a substitute for task design. They misfire. Build tasks that need class process, oral defence, drafts dated in class, or practical evidence you saw.
Grade 12, Umalusi and irregularities
DBE notes that results may be withheld for irregularities, with an appeal route. Umalusi’s job is the credibility of the NSC, not your classroom chatbot. Do not wait for an exam-irregularity scandal to decide the Grade 11 homework rule.
If a learner’s take-home SBA is unusually fluent, unmatched to classwork, and stuffed with US examples, that is a teaching and integrity problem now. Talk to the HOD. Do not paste the script into a detector and call it proof.
Language orals and FAL
Orals are part of official-language assessment. A learner reading a chatbot speech is not an oral. You may use ChatGPT (teacher-side) to draft prompt cards or a list of follow-up questions, then run the oral yourself.
For FAL writing, chatbot “model essays” teach the wrong lesson if learners copy them. Prefer: generate a weak paragraph, then mark it together on the board.
POPIA, scripts and memos
Do not upload a class’s answer books, barcode sheets, or a spreadsheet of names and marks to ChatGPT. The Information Regulator administers POPIA. This page is not legal advice. Ask the school information officer.
Teacher files already hold assessment evidence. Keep that evidence in the school’s agreed store, not in a public chatbot history.
This week in an FET subject
- Pick one upcoming informal worksheet. Generate a draft, check CAPS/ATP, print.
- For the next SBA, write one line on the task sheet: red or amber, in plain language.
- Agree the same line with the other teacher of that grade.
- Tell Grade 12, once, that the NSC exam and the PAT are their own work. Do not outsource that speech to a generated letter.
School-wide colours: use the ChatGPT traffic-light page in this cluster once it is live. Assessment recording: CAPS assessment. Ethics: ChatGPT and the CAPS curriculum.