Custom GPTs can turn ChatGPT from a general-purpose chatbot into a focused teaching tool for a specific subject. They allow teaching staff to specify the assistant’s role, provide approved subject material, and set clear rules for what students should (and should not) ask it to do. A GPT is a customised version of ChatGPT configured for a particular task or context. It combines three elements:
- instructions that determine its role, tone, workflow, and boundaries.
- knowledge, which consists of files uploaded as reference material; and selected
- capabilities, such as web search.
This makes a GPT materially different from an ordinary ChatGPT conversation. It can be designed to consistently work from your subject guide, assessment rubric, weekly resources, AI-use guidance, and other approved materials.
The knowledge feature is broadly a form of retrieval-augmented generation (RAG): when a student asks a question, the GPT can retrieve relevant content from the files you have uploaded and use it to augment its answer. Uploading files, however, does not tell the GPT how to teach. Instructions do that work. A useful distinction is that knowledge files provide facts and sources; instructions define behaviour.
This design control is the principal educational advantage. Rather than accepting generic answers and generic tutoring behaviour, lecturers can create an assistant with a bounded role in the subject. Ethan Mollick and Lilach Mollick describe several productive AI roles in education, including tutor, mentor, simulator and AI “student”. The key question is not whether AI is used, but what role it plays, and which intellectual work remains with the student.
The four downloadable templates below provide adaptable starting points. They are not ready-made solutions: each must be populated with current subject-matter content, assessment-specific AI-use settings, and local escalation pathways. In the GPT editor, create a GPT, configure its instructions and conversation starters, upload approved knowledge files, test it in Preview, then share it only through approved institutional channels
1: Build a GPT as a Feedback Mentor
A Feedback Mentor is an assessment-specific assistant that gives formative feedback and feedforward without becoming a marker, ghostwriter or substitute lecturer. Its role is to help students notice what matters in their own draft, interpret the published criteria and decide what they will revise. This is important because useful feedback is not simply information delivered to students; it should prompt them to make judgements, act on advice and monitor improvement.
This is particularly valuable in subjects where students need repeated opportunities to develop disciplinary writing, argument, analysis, presentation design or reflective practice. A mentor can provide timely formative interaction between classes, but its instructions should deliberately limit it to two or three high-impact priorities. It should identify an issue, link it to a criterion, ask a purposeful question, explain why it matters and invite the student to revise. The student must remain the author of the work.
Upload the assessment brief, rubric, relevant referencing guidance and assessment-specific AI-use instructions. Explicitly instruct the GPT not to allocate marks, estimate grades, rewrite passages, generate missing analysis or create references. A well-designed mentor, therefore, supports the learning process while preserving assessment validity and lecturers’ judgement.
Download the instructions:
2: Build a GPT as a Simulated Practice Partner
A Simulated Practice Partner gives students a structured rehearsal environment for an authentic interaction: for example, a professional conversation, interview, negotiation, client briefing, explanation of a complex issue or response to a difficult scenario. The GPT takes a defined role and responds to what the student says or does. It does not coach the student through every move during the simulation.
This role matters because professional capability is not acquired only by reading about practice. Students often need repeated, low-stakes opportunities to notice cues, make decisions, communicate under constraints and adapt when the situation changes. A simulation can make that rehearsal more accessible and repeatable, particularly where placements, live role-play or individual feedback are limited. It should complement—not replace—supervision, placements, professional judgement or high-stakes practical assessment.
The most effective structure is prepare, perform, debrief, retry. First, the GPT gives a short pre-brief: the goal, roles, boundaries and stopping point. During the scenario, it stays in role, reveals information progressively and allows plausible consequences. Afterwards, it clearly pauses the simulation and asks students to explain their reasoning, identify cues they noticed or missed, and set a goal for a second attempt.
Keep scenarios bounded: one observable skill, one main role and a manageable level of complexity. Use fictional or de-identified cases, specify safety limits, and prevent the GPT from diagnosing, counselling or impersonating identifiable people. Upload scenario facts, professional standards and criteria, while putting role behaviour and debrief rules in the instructions.
Download the instructions:
A: Build a GPT as a Simulated Practice Partner
B: Knowledge base: Teacher Education Parent Conversations Scenarios
3: Build a GPT as a Socratic Tutor
A Socratic Tutor supports understanding and application without simply supplying answers. Its role is to ask what the student already knows, explain one small idea or step, prompt retrieval or application, and invite reflection. It is particularly suitable for threshold concepts, common misconceptions, foundational disciplinary knowledge and low-stakes practice.
This matters because learning requires productive cognitive effort. Students benefit when they retrieve knowledge, explain ideas in their own words, compare cases and apply concepts to unfamiliar contexts. A tutor that provides long, comprehensive explanations on demand can create an illusion of understanding: students may recognise the explanation without being able to use the knowledge independently. The template is therefore designed to reduce avoidable confusion and cognitive overload without removing the intellectual work through which learning occurs.
A useful teaching cycle is activate, explain, retrieve, reflect. The tutor begins by asking what the student has tried or already understands. It then provides a concise explanation, example or analogy; asks the student to restate, compare or apply the idea; and periodically uses a no-notes retrieval or transfer question. If students struggle, it should provide a hint before a solution, then fade support as competence improves.
Upload the relevant weekly resources, approved explanations and terminology. Instruct the GPT to prioritise those sources, identify when the materials do not support an answer and avoid inventing references. It should also refuse requests to complete graded assessments while redirecting students towards planning, practice, or feedback on their own attempts.
Download the instructions: Build a GPT as a Socratic Tutor.
4: Build a GPT as a Student Subject Helper
The Student Subject Helper has a different, practical role: it helps students find, understand and act on authoritative subject information. It can explain assessment requirements, due dates, submission processes, learning guide content, subject resources and assessment-specific AI-use settings. It is a navigator and explainer, not the source of record or a decision-maker.
This role is important because students often lose time navigating dispersed information or asking repeated administrative questions. A well-designed helper can reduce this friction, improve access to routine information and direct students to relevant resources at the point of need. But accuracy is critical. For dates, assessment requirements, submission rules and university processes, the LMS, subject learning guide, assessment documentation and current institutional guidance remain authoritative.
The GPT should use a simple response pattern: give a concise answer, name the source, and state the next action. If the question is ambiguous, it should ask which assessment or process the student is referring to. If sources are absent or in conflict, or if a decision is required, it should refer the student to the Subject Coordinator or the appropriate university service.
Upload only current, student-facing documents: the subject learning guide, assessment briefs and rubrics, schedule, AI-use guidance, submission instructions and approved contact pathways. Explicitly prohibit the GPT from approving extensions, predicting grades, deciding integrity cases, collecting sensitive personal details or completing assessable work.
Download the instructions:
A: Build a GPT as a Student Subject Helper.
B: Student subject helper checklist
AI Author Statement. All the resources were made with help of Chat GPT (Project function with knowledge and system prompts). I have built and tested all the GPTs, feel free to build and test your own, they may behave differently in various teaching contexts.
Leave a Reply