Teaching · Practical guide
Design an AI-Enhanced Course Around Evidence of Learning
Map outcomes, AI-supported activities and independent evidence into a coherent course sequence, then test one unit before scaling.
Start with outcomes rather than a tool list
A course is not enhanced merely because every week mentions AI. Start with what learners should be able to explain, perform and judge at the end. Identify where assistance could improve practice and where independent performance must still be demonstrated. Keep the subject learning visible alongside any AI-literacy outcome.
For an introductory quantitative course, the subject goal might be to interpret percentages accurately. The AI-literacy goal might be to verify generated calculations and document assistance. These goals can support one another, but neither should disappear into a broad instruction to “use a chatbot”. UNESCO’s teacher framework offers a useful professional-development context for making these decisions.
Build an alignment map
For each outcome, specify preparation, practice, feedback, evidence and the allowed role of AI. If an outcome requires explanation, the evidence should include an explanation. If it requires performing a procedure, a generated description alone is insufficient. Check that students practise the same type of thinking they will be asked to demonstrate.
Distinguish teacher use from learner use. A teacher may use an approved assistant to draft practice items and verify them before class, while learners complete those items without an account. This can be a valid design. Do not impose a tool requirement unless it serves the outcome and its access, privacy and support arrangements are workable.
Worked example: a three-session unit
Session 1, 60 minutes: introduce part, whole and percentage using a teacher-checked example; learners identify quantities in pairs, then complete an individual exit task. AI’s role is preparation of varied fictional examples, all checked before release. Evidence is each learner’s ratio and explanation, not the assistant’s output.
Session 2, 60 minutes: provide a deliberately flawed generated explanation. Learners locate the error, calculate independently, request or draft a targeted revision, and compare the result with the source data. Example: 18 failures out of 240 is 7.5%, not 18%. Evidence is a correction log showing the original claim, calculation and revised interpretation.
Session 3, 60 minutes: give new data and ask learners to solve and justify the result under the stated assessment conditions. Use 16 of 200: 8%. Ask a follow-up about 16 of 400: 4%. The numerator is unchanged but the reference total doubles. A learner who explains that difference demonstrates more than recognition of a copied answer.
This is an illustrative unit design, not a claim of improved results. A course team would adapt timing, prerequisites and approved assessment conditions before use.
Design feedback and access together
Plan when learners receive feedback and what they do with it. A comment that arrives after the final opportunity to revise may have a different purpose from feedback used in practice. Include time for learners to correct an error and explain the correction, rather than filling the schedule entirely with new content.
Provide an equivalent route when an AI service is unavailable or a learner cannot use it. For the second session, a teacher-prepared flawed response supports the same critique objective without a live model. This fallback also makes the activity reproducible. Keep personal information and protected assessment material out of unapproved services.
Evaluate the unit before expanding the whole course
Choose evidence of success before the trial: correct denominator selection, accurate arithmetic, justified interpretation and appropriate verification. Record practical issues such as time required, access difficulties and unclear instructions. Learner satisfaction can inform usability, but it does not alone establish learning.
Do not compare unlike classes or tasks and attribute every difference to AI. A small pilot can reveal problems and generate questions without proving effectiveness. Review the assessment approach at programme level where required, consistent with the broader concerns discussed in TEQSA’s reform resources. Keep a non-AI baseline so the team can ask whether the added tool actually serves a useful purpose.
Practice: map and stress-test one unit in 25 minutes
Use five minutes to choose two outcomes. Spend eight minutes mapping preparation, practice, feedback and evidence for each. Use six minutes to write a worked task and check its answer. Spend six minutes testing the plan against no internet, an incorrect generated response and a learner who needs an approved alternative format.
Example answer: the percentage unit retains printable source cards and a verified answer key; an incorrect AI response becomes material for the critique activity; the final task uses the same reasoning criteria across approved formats. The fallback preserves the learning objective rather than cancelling the lesson because a tool is unavailable.
Self-review: does every assessed outcome have prior practice? Is the role of AI, including prohibited uses, explicit at each stage? Are verification, access and fallback arrangements ready? Keep the completed map, task, answer and review notes together. Expand the course only after resolving the unit’s observed problems, rather than multiplying an untested activity across every week.
Reusable prompt
Select and copy this template, or customise your own using the free builder.
Create an alignment map for this course unit: [outcomes, prerequisites and duration]. For each outcome specify preparation, practice, feedback, evidence and permitted AI assistance. Include a fictional worked task with answer, independent verification, an accessible non-AI alternative and a fallback for tool failure. Identify approval questions and evaluation evidence. Do not claim measured learning gains or institutional adoption.Open the prompt builder →
Pause · Think · Check
Test your understanding
Choose one answer per question, then check your reasoning. This is practice; scores are not saved.