Digital & AI English · English Learning Transfer · Feedback & Performance · Secondary 2 English · English Atlas
The most important AI question in English tuition is not “Did AI help the student produce better work?” It is “What can the student now do when AI is gone?”
AI-free testing separates assisted performance from owned capability. It protects the tutor from mistaking polished output for learning and gives the student credible evidence of independence.
The AI-off transfer test
- Teach or practise with carefully bounded AI support.
- Close the AI tool.
- Change the surface of the task.
- Reduce tutor prompts.
- Ask the learner to perform independently.
- Delay another retest where possible.
- Check whether the same control appears in school work.
1. Assisted output is not the final evidence
The OECD Digital Education Outlook 2026 highlights an important risk: students can produce stronger work with general-purpose generative AI while the advantage disappears when the tool is removed. Tuition therefore needs a separate independence test.
2. Test the same mechanism, not the same question
If AI helped the student understand inference scope, do not reuse the exact passage. Give a different passage with the same underlying decision. Familiarity can hide dependence.
3. Reading should be AI-off before interpretation is judged secure
Give an unfamiliar text and ask the learner to build the passage model, select evidence and justify the inference without AI summaries or suggestions. The tutor can observe where the internal process now succeeds or breaks.
4. Writing should be AI-off before authorship is judged secure
Ask the learner to interpret the prompt, plan, draft and revise a fresh piece without AI. Compare the result with earlier assisted work. Has idea quality improved? Does paragraph development survive? Is vocabulary still controlled? Does the writing remain recognisably the student’s own?
5. Editing should be AI-off before detection is judged secure
Use fresh sentences or the learner’s own delayed draft. Do not highlight the error location. If AI previously explained the rule, the student should now detect and repair the pattern independently.
6. Vocabulary should be AI-off before retrieval is judged secure
Recognition inside an AI conversation is not enough. Ask the learner to retrieve precise words and collocations in new speaking or writing contexts without suggestions. Transfer requires independent access.
7. Oral performance should be AI-off before flexibility is judged secure
Ask fresh follow-up questions without model answers visible. The student should build the response live, adjust when challenged and recover after uncertainty. Spoken ownership becomes obvious when scripts disappear.
8. Timing should be tested without AI support
AI can speed planning and checking. A school assessment may not permit that support. Use short timed tasks to see whether the learner can still recognise the job, close decisions and recover when the tool is unavailable.
9. Remove one support at a time
If AI support has been heavy, do not jump directly to a full unaided paper. First remove generated examples, then hints, then feedback, then the tool entirely. This reveals which support the learner still depends on.
10. Test after a delay
An AI-supported insight may remain active in short-term memory. Retest days or weeks later. Delayed independent success is stronger evidence of learning than immediate repetition.
11. Change the surface
Change topic, genre, wording, examples or audience while preserving the same underlying mechanism. The farther the surface changes without destroying performance, the stronger the evidence of transfer.
12. Ask the learner to self-diagnose without AI
After an independent task, ask: what was the hardest decision? What error risk appeared? What would you practise next? If AI has supported metacognition effectively, the student should increasingly be able to run this diagnostic process personally.
13. Compare process, not just product
A slightly weaker AI-off paragraph can still show strong learning if the student now plans independently, recognises the target, self-corrects and explains choices. The independence test should inspect the route as well as the final mark.
14. Use school work as the final return path
The strongest evidence appears when the repaired behaviour shows up in ordinary school English where AI access may be restricted or absent. Marked work, timed tasks, oral performance and teacher comments can confirm whether the capability transferred.
15. Maren: AI-off writing transfer
Maren used AI-generated questions to strengthen evidence interpretation. One week later she receives a new prompt and writes without AI. Her paragraph explains what the example proves before any tutor question appears. The AI-supported cue has become an internal writing decision.
16. Iona: AI-off reading transfer
Iona used AI to compare alternative interpretations. Later she reads a new passage unaided, identifies two plausible readings and selects the narrower one using evidence. The comparative habit survives without the tool.
17. Leonie: AI-off execution transfer
Leonie used AI to simulate recovery scenarios. In a later timed task she becomes stuck, labels the question unresolved, moves on and returns during the review window without prompting. The recovery system now belongs to her.
18. Distinguish four transfer states
- AI-dependent: good performance disappears when AI is removed.
- Prompt-dependent: performance survives without AI but still needs tutor cues.
- Fragile independent: succeeds alone under favourable conditions but fails with delay or variation.
- Stable independent: succeeds across fresh tasks, delay and realistic school conditions.
19. Use AI-off baselines regularly
If AI is used frequently, schedule periodic independent baselines. Otherwise assisted performance can slowly become the new normal and the tutor loses visibility into what the learner owns.
20. Increase AI-off work before assessments
As a school assessment approaches, practise under the rules and conditions the learner is likely to face. If AI will not be available, independence should be tested before the assessment, not discovered during it.
21. Do not remove useful AI forever once independence is proven
The point is not to ban the tool. Once the learner can perform independently, AI can return for higher-value roles: tougher questioning, broader comparison, additional practice and verification. Independent capability becomes the foundation for better tool use.
22. A six-step AI-off audit
- Can the student recognise the task without AI?
- Can they retrieve the method?
- Can they perform it on new material?
- Can they self-check?
- Can they recover after a mistake?
- Can they do it again after a delay or at school?
23. Where this page fits
This guide owns the AI-off transfer test inside Secondary 2 English tuition. For when AI should enter a task, see When Should a Student Attempt English Work Before Using AI?. For broader school transfer, use How to Make English Tuition Transfer to School English. For progress evidence, use What Progress Should English Tutoring Produce?.
24. Further reading
25. Final test
AI-assisted learning has worked when the student can close the tool and still think. The clearest evidence is independent performance that survives new material, reduced prompting, delay and the ordinary conditions of school English.
Explore the Secondary 2 English hub for related guides.