When AI helps create the work, how do we see the learning?
Research on AI-assisted learning, with practical ways to document reasoning, check references and give students a fair opportunity to explain their own understanding.
Read & put into practice
Clear slides, fluent abstracts and engaging visuals can make a project easier to follow. When AI can help produce them, an educational question becomes especially important: what did the student decide, what did they verify, and which parts can they explain independently?
This article offers ways to use AI while keeping the learner’s reasoning visible. It is for secondary students, teachers and families discussing project work. Every suggestion is conditional on the task’s rules: practice, examinations and competition submissions are different settings. Check the applicable instructions first. [5]
Key ideas to take away
- Examine both the finished work and what the learner can explain without AI.
- Keep proportionate records of drafts, revision decisions and source checks.
- Allow a fair explanation; do not treat a detector score as a verdict by itself.
Better output and deeper learning need separate evidence
OECD Digital Education Outlook 2026 distinguishes improved AI-assisted output from learning, emphasizing purposeful educational design. That distinction invites us to examine both a submitted product and the understanding a student can subsequently use independently. [1]
In Bastani and colleagues’ randomized study at one Turkish high school, AI improved mathematics practice performance. On a later test without AI, students who had used a general-purpose chatbot performed worse than the control group. Students who had used a tutor with learning safeguards did not show the same disadvantage. This does not establish the effects of every AI tool, subject or long-term project. [2]
Begin with a shared, explicit agreement
Before using a tool, the task owner should explain what is permitted, what must be disclosed and what must be done independently. For example, language support might be permitted while generating substitute experimental data is not. Stating boundaries in advance is more useful than leaving students to guess after completing the work.
UNESCO’s guidance emphasizes human agency, age-appropriate use and privacy. [4] Our practical application is to use non-identifying examples, avoid uploading student or unauthorized information, and have an appropriate adult check a tool’s conditions before its use by minors.
| Possible assistance | Student responsibility | A record to keep |
|---|---|---|
| Questioning our draft | Select relevant questions and explain why. | Original draft and questions actually used |
| Language support | Check that meaning and strength of claims remain accurate. | Before/after examples and reasons for edits |
| Suggesting search terms | Read original sources and check that they support the claim. | Link, document title and location of the relevant evidence |
Make reasoning visible without excessive paperwork
It is not necessary to save every chat message. Select important decisions: the starting question, evidence that changed the approach, and the reason for the final conclusion. Where AI is used, record the tool, date, purpose and material actually adopted according to the task’s requirements. Keep only necessary, permitted information.
A useful record goes beyond “AI checked it”. For example: “The tool suggested replacing ‘associated with’ with ‘caused’. I rejected that edit because the data are observational.” This illustrates a decision about both language and evidence. It is a hypothetical example, not a program participant’s record.
Question and first attempt
Show the starting idea, available information and unresolved difficulty before seeking help.
Selection and revision
Explain which suggestions were adopted or rejected and what was checked. A revision count is not an effort score.
What the student can do independently
Explain the core idea, sketch the reasoning or attempt a related task without AI, with agreed and appropriate support.
Check the evidence behind each reference
A plausible author name, journal and DOI are not enough. Open the document, verify its title, year and authors, then locate support for the sentence being written. Check the population, setting, method and stated limitations. If only the abstract is available, make clear that you reviewed only the abstract rather than implying that the full methods were examined.
Keep observed data, calculations and conceptual illustrations distinct. Label AI-generated imagery as conceptual when required, and do not use it as a substitute for experimental evidence. Simulations and hypothetical data can be useful for practice when readers can clearly tell that they are not observations from actual participants.
Give students a fair way to demonstrate understanding
Liang and colleagues found that the studied detectors could misclassify non-native English writing as AI-generated. This 2023 study does not evaluate every current tool, but it supports caution about treating a detection score as conclusive evidence. [3]
Where there is a concern, invite the student to explain drafts, sources and decisions with appropriate time and response options. Hesitant speech or developing English fluency should not automatically be read as lack of understanding. Writing, drawing or another suitable language may help, while preserving the task’s criteria and rules.
| What to explore | A possible question | Listen or look for |
|---|---|---|
| Reasoning | Why this approach rather than another? | A connection between the question, method and evidence |
| Limits | What evidence would change your conclusion? | Recognition of conditions and unknowns |
| Transfer | If this variable changed, what would you check next? | A reasoned next step, not simply a memorized script |
Applying the ideas in TechEd contexts
For SCOPE KOREA Oral or Poster Presentations, students can use these prompts to practice explaining their work and study communication sequences in the published archive. [6] This does not imply that archived projects used AI or that the organizer permits it at every stage.
For K-XCEL and IMEDDAT, support for reading or practice must be distinguished from answering an actual examination. If instructions are unclear, ask the organizer first. Across these contexts, useful habits include checking sources, recognizing limits and taking responsibility for explanations submitted under one’s own name. [5]
01Does disclosure make every use acceptable?
No. Disclosure provides transparency; it does not replace permission. If a task requires independent work or prohibits a tool, those rules still apply even when a student is willing to disclose its use.
02What if the tool sounds certain but the source cannot be checked?
Confidence is not a substitute for evidence. Look for the author’s, publisher’s or originating organization’s document. If it remains unverifiable, narrow the claim, mark it as unconfirmed or use a source you can actually inspect.
From reading to your next small step
Before submission: six transparency checks
Use this to review your own work. It is not an integrity certificate or an AI-detection system.
If a part remains difficult to explain, return to the reading and try it independently before submitting. That is a worthwhile learning opportunity.
Selections stay on this page, are not submitted and reset when the page is reopened.When students can explain what they believe, why the evidence supports it and what still needs checking, technology can sit within a learning process where their own thinking remains visible.
How this article was developed
This selected-source synthesis distinguishes an OECD research overview, Bastani and colleagues’ mathematics experiment, Liang and colleagues’ detector study, and UNESCO policy guidance. We checked the PNAS author manuscript and the notice correcting an author affiliation, a matter separate from the experimental results.
Examples, agreements and prompts are learning proposals, not official TechEd assessment criteria, and their effectiveness has not been tested in these programs. We have no data about archived participants’ AI use and infer none from photographs, videos or writing style. Sources were checked on 1 October 2026; tools can change afterward.
TECH EDUCATION operates the programs discussed and publishes this article, creating an interest in the subject. This is a source-based editorial analysis, not an independent evaluation, and has not undergone external peer review.
Explore the original sources
- 01OECD · Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education (2026)
Introduction and key messages · Task performance with AI and student learning are distinct
- 02Bastani, H. et al. · Generative AI without guardrails can harm learning: Evidence from high school mathematics (2025)
PNAS, 122(26), e2422633122 · Randomized trial in one Turkish high school · Author manuscript and affiliation-correction notice checked: DOI 10.1073/pnas.2518204122
- 03Liang, W. et al. · GPT detectors are biased against non-native English writers (2023)
Author version of Patterns, 4(7), 100779 · Studied the detectors and English writing samples available at the time
- 04Miao, F. & Holmes, W. · Guidance for generative AI in education and research (2023)
UNESCO · Human-centered policy guidance, age-appropriate use and privacy
- 05TECH EDUCATION · คู่มือการสอบ / Testing guide (2026)
Check actual exam instructions and the relevant round notice in TechEd Testing
- 06TECH EDUCATION · SCOPE KOREA GEN 03 (2026)
Examples of project explanation; not evidence that any presenter did or did not use AI
Sources checked October 1, 2026 · Please contact the team with a source if you identify a correction. Contact the team
Explore the related programs
Review the aims, activities and edition-specific conditions before choosing your next experience.
SCOPE KOREASCOPE KOREA | Student Conference on Presentation, Innovation & Evaluation K-XCELK-XCEL | Korea eXcellence through Competency, Experience & Learning IMEDDATMedical Science Aptitude Development and Testing