Teaching · Practical guide
Use AI to Offer Support Without Lowering the Learning Goal
Design graduated support for a shared learning outcome, with a worked example, equivalent practice routes and checks that preserve learner independence.
Identify the barrier rather than labelling the learner
Start with evidence from the task: a learner can describe a ratio but confuses the denominator, or can explain a concept orally but struggles with a long written instruction. These observations suggest different responses. They do not justify asking AI to diagnose a disability, assign an ability category or infer motivation.
Keep personal information out of practice prompts. Describe the instructional barrier and the prerequisite knowledge instead. Consult the appropriate support service for individual adjustments. CAST’s framework is useful for considering options in access and expression, but it should not be turned into a claim that every learner belongs to a fixed visual, auditory or other learning type.
Keep one outcome and vary the support
Choose a common outcome, such as “calculate a percentage and explain which quantity is the whole”. You can offer a labelled worked example, a partially completed example and an independent problem. The level of prompting changes; the mathematical meaning does not. Tell learners which support they may use during practice and what they will later need to demonstrate independently.
Distinguish support from replacement. Providing a reminder to identify part and whole supports the calculation. Supplying the final answer to every task prevents you from seeing whether the learner can perform it. Plan how hints will be reduced and how understanding will be checked on a fresh problem.
Worked example: three routes through the same concept
Fictional problem: 9 of 60 seats are occupied. The target is 9/60 × 100 = 15%. Route A gives a complete model: “occupied seats are the part, all seats are the whole; divide 9 by 60, then multiply by 100.” The learner explains why the denominator is 60.
Route B provides “___ / ___ × 100” and asks the learner to fill the quantities and explain the result. Route C gives only the scenario and asks for both working and interpretation. All routes address the same relationship, but they provide different evidence of independent performance.
Follow with 14 of 80 seats. The answer is 17.5%. A learner who can solve and explain this new example without the scaffold has stronger evidence of transfer than one who simply repeats the original 15%. The two examples are illustrative; no effectiveness or classroom trial is claimed.
Choose meaningful options for response
If the outcome is mathematical reasoning rather than formal writing, a short spoken explanation accompanied by the calculation may reveal the relevant understanding. If the outcome explicitly includes professional written reporting, an oral response alone may not cover it. Match the format to what is being assessed and to approved adjustments.
When using AI to generate alternate explanations, keep numbers and definitions consistent across versions. Check whether an analogy introduces a new misconception. Ask the learner which step is unclear rather than forcing a particular support route based on a presumed type. Provide an accessible non-AI route when tool access is unequal.
Review progress without automated profiling
Record task-level evidence: “identified the whole correctly”, “needed a prompt to convert to percent”, or “explained the new example independently”. These statements help choose the next teaching step. Avoid broad labels such as “low ability” based on one response or a model’s interpretation of language.
A useful next task changes one feature at a time. If the learner understands the relationship but makes arithmetic errors, retain the relationship and provide calculation practice. If the denominator is still confused, return to identifying part and whole with simple whole-number examples. An AI suggestion is a possible teaching move, not an authoritative diagnosis.
Practice: plan and fade support in 20 minutes
Use four minutes to choose an outcome and a specific barrier. Spend six minutes drafting three support levels. Use five minutes to solve a fresh independent check, then five minutes to write what evidence would cause you to retain, reduce or change the support.
Example answer: for 12 failed checks out of 150, start with a labelled ratio, then remove labels on a new problem. The answer is 8%. A good interpretation states that eight in every hundred checks fail in the given data. Reduce support when the learner independently identifies the quantities and explains the conversion on a different example, not merely when they copy a model correctly.
Self-review: is the outcome unchanged across routes? Is the support based on observed work rather than a diagnosis or fixed learner label? Does the final check show independent understanding? Keep the support plan flexible and note when specialist advice or an approved adjustment is needed.
Reusable prompt
Select and copy this template, or customise your own using the free builder.
Create three levels of support for [learning outcome] and this observed task barrier: [non-sensitive description]. Keep the same objective and accurate content. Give a worked model, a partially completed example, an independent task and explained answers. State how to fade hints and what evidence to observe. Do not diagnose learners, infer ability or use fixed learning-style labels.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.