Questioning AI output: seven checks before you trust an answer

Key facts

• In a 2025 study of nearly 1,000 secondary school maths students in Türkiye, practising with plain GPT-4 raised practice grades by 48%, but once the AI was taken away those students scored 17% lower than students who never had it.^[1]
• The OECD's Digital Education Outlook 2026 (January 2026) found that leaning on AI tools that hand out direct answers can improve task performance without the matching learning gains.^[2]
• The AI literacy framework published by the OECD and the European Commission in 2026 sets out 19 competences; one of them is deciding whether an AI output should be accepted, revised or rejected.^[3]
• Malaysia's Ministry of Education (KPM) put its AI literacy guide into force on 8 May 2026. It expects students to use AI without setting aside honesty and academic integrity.^[4]
• Malaysia's Personal Data Protection Act 2010 (Act 709) covers personal data used in commercial transactions, and the Personal Data Protection Department (JPDP) regulates it.^[5]

Why check at all

AI chatbots are good at drafting, explaining and translating. They can also be wrong in ways that sound right. The OECD and European Commission framework calls this a "hallucination": the AI invents information that sounds plausible but is not factual.^[3] It can invent a fee, a deadline, a quote or a whole reference.

How you use AI also changes what you learn. In the Türkiye study, a tutor version of the chatbot, built to give hints instead of answers, largely avoided the drop in exam scores.^[1] That is one study, in one subject and one school. The habit it points to works anywhere: think first, then use AI, then check.

Seven checks

Check 1: Does the source say that?

Underline every name, number, date, requirement and promise in the AI answer. Find each one in a source you trust: the brief, the fact sheet, the official page or the library item. Give each claim one of three verdicts.

"In no source" is where invented details hide: a guarantee nobody made, a service the client does not offer, a prize a person never won. If the AI gives a link or a reference, open it. A reference that does not exist, or does not say what the AI claims, counts as in no source.

Check 2: How old is it, and who published it?

Find the date on every source. Fees, entry requirements, salaries and application windows change every year, so a figure that was right two years ago can be wrong today. When two sources disagree, the newer one from the organisation in charge (the college, the ministry, the agency) usually beats an old flyer or a forum post.

If no source answers a question, the answer is unknown. Unknown is not true and not false. Write that it is unknown, and where someone could check.

Check 3: Do the numbers add up?

Check the currency (RM or US$), the period (per month or per year), percent against percentage points, and which year a figure is for. Then redo every sum, total and average yourself, by hand or on a calculator. A wrong total looks just as confident as a right one, so never use an AI number you have not checked.

Check 4: Did the translation keep the meaning?

Malaysians often mix Bahasa Malaysia, English and other languages in one message. A machine translation can flatten the tone, drop the politeness or change a course name or a legal term. Professional translators check technical terms for accuracy as a normal part of their job.^[6] Compare the translation with the original, and ask a fluent speaker when the meaning matters.

Check 5: Who is left out?

Ask whose situation the answer ignores: someone without a smartphone, a rural family, a person with a disability, someone who reads mainly in Tamil or Chinese, a family with less money. Fairness and inclusivity are two of the seven principles in Malaysia's national AI ethics guidelines (AIGE).^[7]

Check 6: What did you paste in?

Never type IC numbers, home addresses, phone numbers, health details, photos of other people or anyone's records into an AI tool, even to "tidy them up". What you paste may be stored or used in ways you cannot see. Privacy and data security is one of the AIGE principles,^[7] and the Personal Data Protection Act 2010 protects personal data in commercial transactions.^[5] In First Move every person and record in a practice task is made up, so you never need real details.

Check 7: Accept, fix or reject, and was AI the right tool?

After checks 1 to 6, decide: accept the output, fix it or reject it, and be ready to say why. The OECD and European Commission framework treats this as a core AI skill, together with deciding whether AI suits the task in the first place.^[3]

Skip AI when the point is to show or build your own thinking (unless your teacher or facilitator allows it), when a decision affects someone's health, money or legal position, and when you are upset and need a person. Talk to a trusted adult, a counsellor or a teacher instead. The same rule covers AI suggestions about you: in your Future Map, accept, edit or reject each one, and say why.

Where you use these checks in First Move

A printable checklist is attached. It follows the same seven checks.

What the official guidance says

KPM (Malaysia). The Ministry of Education issued its AI literacy guide through a professional circular, SPI Bil. 2 Tahun 2026, dated 8 May 2026 and in force from that date.^[4] It applies to all education institutions under KPM, and private institutions registered with KPM are encouraged to use it. It covers four areas: AI governance, teaching and learning, lifelong learning and AI ethics. Students are expected to be active users with integrity, using AI for creativity and problem solving without setting aside honesty and academic integrity. Parents and the community are asked to watch over safe, ethical use outside school hours and to protect students' personal data.^[4] Your college or university may have its own rules too, so ask what applies to you.

AIGE (Malaysia). MOSTI launched the National Guidelines on AI Governance and Ethics in September 2024. They are voluntary and set seven principles: fairness; reliability, safety and control; privacy and data security; inclusivity; transparency; accountability; and human benefit.^[7]

OECD and European Commission. Empowering Learners for the Age of AI, approved by the OECD's PISA Governing Board on 8 April 2026, is an AI literacy framework for primary and secondary education. Its 19 competences sit in four domains: Engage with AI, Create with AI, Manage AI and Shape AI. It feeds into the PISA 2029 assessment of media and AI literacy.^[3]

UNESCO. UNESCO's AI competency framework for students (August 2024) has 12 competencies across four dimensions (a human-centred mindset, ethics of AI, AI techniques and applications, and AI system design) and three levels: understand, apply and create.^[8] The human-centred mindset is about understanding and asserting your own agency in relation to AI.^[9]

This guide mostly practises the "Engage with AI" and "Manage AI" competences, and UNESCO's human-centred mindset and ethics dimensions, at a beginner level. That mapping is ours, not the OECD's or UNESCO's, and First Move uses these frameworks as references, not as accreditation.

What this means for you

• Treat every AI answer as a draft to question. The more specific and confident it sounds, the more check 1 matters.
• Spend your checking time on what would hurt if it were wrong: money, dates, entry requirements and anything about a real person.
• Practising with AI is not the same as learning. Try on your own first, then use AI, then check.
• "I could not confirm this; check it at the college office" is an honest, useful answer. A confident guess is not.
• Your own personal details, and other people's, stay out of AI tools.

Your next step

1. Take one AI answer you got this week about a course, a fee or a job, and run checks 1 to 3 on it. Note what you changed.
2. Open the attached checklist and write your own one-line rule at the top: what you will check first the next time an AI tool gives you a number, a date or a deadline.

For facilitators

Judge the reasoning, not the polish. Keep the first attempt, the revised answer and the source references, and accept a spoken or BM explanation as fully as a written one. Record each learner as emerging, demonstrated with support or demonstrated independently on "uses and verifies AI appropriately", with one short example. These are teaching notes, not scores.