Set clear expectations for teaching with AI
What should students do themselves, and where can AI help?
Choose the evidence of learning first. Then define permitted assistance, practise it with students, and publish the conditions for the specific activity and assessment.
In brief
- The learning outcome should explain why a use is allowed, limited, or required.
- Study activities and assessed work can have different conditions.
- Students need clear rules, accessible routes, and an opportunity to practise before assessment.
Start with the learning outcome
TUM ProLehre's faculty guidance recommends starting with intended learning outcomes, analysing assessment tasks, and considering students' prior knowledge. It distinguishes learning during the semester from summative assessment and asks instructors to consider fairness and professional practice.
Write down what a student should be able to explain or do at the end. Then identify which part of an AI-assisted result would still provide evidence of that ability.
For example, "produce a clear technical report" could involve several goals: selecting a method, interpreting measurements, supporting a claim, and writing for a reader. AI assistance with wording has a different effect from AI selecting the method or inventing the interpretation. Specify the skill the task will assess.
The following questions adapt ProLehre's four-page decision guide into a planning sheet:
- Which competence must students demonstrate independently?
- Could an AI system produce the assessed work while the student remains unable to explain it?
- Which assistance supports practice, and which replaces the skill being assessed?
- Can students judge the system's answer with their current subject knowledge?
- Is competent AI use itself an intended outcome? If so, what practice and assessment evidence will show it?
The German source recommends retaining core judgement skills, checking prior knowledge, and changing teaching and assessment together when outcomes change. Its recommendations do not establish one required policy for all modules.
Match assistance to the task
Scroll horizontally to view all columns.
| Outcome | Possible assistance | Evidence to preserve |
|---|---|---|
| Explain a concept | Ask for a question or hint after an initial attempt. | The student gives an explanation and handles a new example. |
| Evaluate an argument | Generate a counterargument for students to examine. | Students identify assumptions and support their judgement with sources. |
| Diagnose a code defect | Use AI feedback after inspecting the behaviour. | Students explain the cause, inspect the proposed change, and test relevant cases. |
| Choose an analysis method | Compare suggested methods and their limits. | Students justify the choice against the actual data and assumptions. |
| Use AI responsibly | Plan a division of work and record decisions. | Students show permission checks, verification, rejected output, and a reasoned final decision. |
These are adaptable examples, not permitted-aid rules. Check the applicable module description and examination format before adopting a task. An oral discussion, staged draft, or work sample can make reasoning visible when it fits the approved format. Adding a new assessment element requires the relevant examination rules to support it.
Try the task yourself with the kind of tool students can access. Inspect how much of the intended work it can produce. If the output hides the reasoning you need to assess, revise the task or evidence requirements. A different prompt alone may not solve that problem.
Publish activity-specific conditions
ProLehre recommends communicating permitted AI use early, in writing, and repeatedly. Explain the reason and demonstrate suitable use. A student should be able to find the current conditions without reconstructing them from announcements.
An activity sheet can state:
- The intended learning outcome and the task it applies to.
- Which assistance is allowed, restricted, or required.
- What the student must produce and explain independently.
- Which tools and input material may be used.
- What verification and assistance record the task requires.
- How students can take part with the available access and support.
- Where to ask before using an aid when the conditions are unclear.
Check the applicable TUM statutes and regulations, programme requirements, and local assessment instructions. The rule needs to describe the specific assessment; a general statement that "AI is allowed in this course" can leave students unsure about individual tasks.
Adapt a concrete announcement
Before adopting this example, decide the tool route, feedback alternative, required evidence, and permitted inputs for your actual task. Check that the wording matches its assessment rules and accessibility arrangements. Change the allowed assistance if writing the rationale is not one of your intended outcomes.
A declaration helps the reader understand the system's contribution. It does not authorize an otherwise prohibited use. Refer students to verify and declare for practical checks, while keeping the required format in your own task instructions.
Teach the verification you expect
Give students a short practice task before they need to submit work under the conditions. Use an example whose errors they can inspect.
For a methods assignment, supply a synthetic dataset and an AI-generated recommendation that overlooks one stated assumption. Ask students to identify the problem, compare a suitable alternative, and explain which evidence changed their decision. Provide a route that uses a supplied answer rather than requiring a paid account.
For code, supply a small program and a proposed change. Have students inspect the difference, explain the expected behaviour, and test a relevant edge case. Passing a test only shows what that test exercises. The student's explanation needs to connect the change to the defect.
The exercise and feedback should make the expected skill visible. Do not infer learning from the polish of a finished output or the number of prompts submitted. Use the study guide for examples of practice that leaves the student doing the explanation and checking.
Plan for equal access and assistive technology
Tool access, prior knowledge, and hardware can differ. As an editorial design recommendation, provide an institutional route or an equivalent activity when AI use is required. Avoid making a premium subscription or a powerful private laptop necessary to demonstrate the learning outcome. Explain the tool before assessing its use, and provide readable task instructions and accessible feedback.
Accessibility needs a separate route. TUM's guidance on reasonable accommodations describes individual measures, potentially including technical aids, with applications handled by the responsible examination board. The performance goals remain. A course's general AI restriction does not settle an individual accommodation request.
Discuss relevant technical features early with the student and the responsible support route. TUM's special-needs FAQ provides practical contacts and procedures. Students should not have to share medical information with a chatbot to obtain support.
ProLehre also recommends discussing accessibility-related processing of lecture material with the instructor or academic advisor. Consent to process materials and an accommodation are distinct questions; resolve both through the responsible people.
Keep teaching data and system decisions in scope
A demonstration can use synthetic student work. Before processing real submissions, feedback, grades, or participant material, confirm the purpose, permissions, and service route. TUM ProLehre's current public guidance says not to enter personal, sensitive, or confidential data into AI systems, including approved ones.
AI used for grading, admissions, or monitoring students raises different questions from a practice chatbot. The rules and rights guide explains the AI Act's roles, educational use cases, and current dates. Refer a proposed consequential deployment to the responsible institutional owner before introducing it.
When work is questioned, use the applicable rule and concrete evidence. ProLehre's decision guide warns that reliable monitoring of AI use is difficult. A detector score alone does not explain who did the work, which assistance was permitted, or what the student can demonstrate.
Sources and scope
The ProLehre faculty page and its German decision guide, TUM support routes, and the cited rules database were checked on 5 October 2026. This page reports public teaching guidance and offers labelled editorial examples. It does not supply a mandatory TUM policy, universal declaration form, or individual legal decision.
Recheck the task conditions when tools, outcomes, assessments, or institutional guidance change. For help adapting the activity, contact ProLehre; for a particular assessment decision, use the responsible examination route.