AI for students and teachers
Learn with it instead of outsourcing your thinking - and mark work honestly.
Learning goal: You use it as a tutor and a marker, not a ghostwriter.
1 / 9
A plain guide to using AI to actually learn and teach better, not just outsource the work.
Before you start (prerequisites)
- Access to a chatbot (many schools and universities now provide or approve specific ones - check yours).
- Your institution's current academic integrity policy on AI use, read at least once.
- A clear idea of what you're trying to learn, not just what task you need finished.
- For teachers: awareness that detection tools are unreliable, so policy needs to rely on more than a single scan.
- No technical skill needed. Curiosity and a bit of discipline are the actual requirements.
Do this first:
- Read your school's or university's specific AI policy, not just general rumours about it.
- Try asking a chatbot to quiz you on something you're currently studying, rather than asking it to explain the topic.
- Ask it to explain the same concept two different ways and see which one actually helps you understand.
Where things stand (as of 2026)
AI is genuinely good at explaining concepts multiple ways, generating practice questions, and giving instant feedback on drafts. It's much less reliable for citations (it can invent plausible-looking references that don't exist), and it can't reliably tell you whether a piece of writing was AI-generated - detection tools produce both false positives and false negatives regularly enough that no serious institution should rely on them alone. Policies on acceptable use vary hugely between schools, subjects, and countries, and are being rewritten often, so what was "fine" last term might not be fine this term. Always check current local policy rather than assuming.
The mental model
animated · saying the unsaid
Four versions of the same request, and how the spread of answers collapses.
what you said
Make me a website.
what you left unsaid
- for what?
- who visits it?
- what should they do?
- does it need to sell?
The model will fill every one of these gaps. Silently. With the most average answer it knows.
three attempts, same prompt
Three tries, three unrelated things. You are rolling dice.
The useful question isn't "can AI do this for me" - it almost always can. The question is "does letting it do this help me learn, or just help me finish." Summarising a chapter you've already read, to check your understanding, is different from asking it to write the essay you haven't done the reading for. Same tool, completely different outcome.
Scenarios that work
Testing your own understanding
low stakeswhat you want
Check if you actually understood a topic you just studied
what most people type
explain photosynthesis to me
the briefed version
I just read about photosynthesis. Quiz me with 5 questions of increasing difficulty, don't give me the answers up front, and tell me after each answer whether I'm right and why.
Stake: Worst case is a confusing quiz, not a wrong grade - always cross-check with your course material.
Getting feedback on a draft essay
medium stakeswhat you want
Improve your own essay before submitting
what most people type
fix my essay
the briefed version
Here is my essay draft: [paste]. Don't rewrite it for me. Instead, give feedback on structure, argument clarity, and weak evidence, as bullet points, so I can revise it myself.
Stake: Submitting a fully AI-rewritten essay as your own can breach academic integrity policy and undermines your own learning.
Finding sources and citations
high stakeswhat you want
Get reading suggestions for a research topic
what most people type
give me sources on climate migration
the briefed version
Suggest 3 general themes to research on climate migration, but do not give me specific citations or reference lists - I know those can be inaccurate. Instead point me to the type of source I should search for (e.g. UN reports, peer-reviewed journals) so I can find real ones myself.
Stake: Submitting a fabricated citation in academic work is a serious integrity breach, even if unintentional.
Generating practice questions (for teachers)
medium stakeswhat you want
Create a differentiated worksheet quickly
what most people type
make a maths worksheet
the briefed version
Create a worksheet of 10 algebra questions for Year 8, in three tiers: 3 easy, 4 medium, 3 challenging, matching the UK national curriculum level for that year. Include an answer key.
Stake: An error in an answer key undermines lesson time and student trust if caught mid-class.
Handling a suspected AI-written submission (for teachers)
high stakeswhat you want
Decide what to do when a piece of work feels off
what most people type
run this through an AI detector and treat the score as fact
the briefed version
Note the specific things that feel inconsistent (vocabulary jump, no reflection of classroom discussion, unusual structure for this student). Have a conversation with the student about their process and drafts before drawing any conclusion, per your institution's academic integrity procedure.
Stake: Wrongly accusing a student based on an unreliable detector score can cause real harm and undermine trust; so can letting genuine misconduct slide.
Where it fails, and what it costs you
The single biggest failure mode for students is outsourcing the thinking, not just the typing. If AI writes the essay, you didn't practise the skill the essay was meant to build - and that gap shows up later, in an exam, an interview, or a job where you're expected to actually reason it through yourself. That's a cost you pay later, quietly, when it's harder to fix.
Citations are the sharpest specific risk: chatbots can generate references that look completely real - plausible author names, journal titles, page numbers - that simply don't exist. Submitting one of these in academic work is treated the same as fabricating a source yourself, because functionally, it is.
For teachers, the biggest risk is leaning on detection tools as if they were reliable evidence. As of 2026, these tools produce meaningful rates of both false positives (flagging genuine student work) and false negatives (missing AI-generated work), so treating a single score as proof, in either direction, is not defensible and can damage a student's trust and academic record unfairly.
How to check the answer
- For any citation or reference, search for it independently in a real database before it appears in your work.
- Ask yourself honestly: did I understand this well enough to explain it to someone else without notes? If not, the AI did the thinking, not you.
- Compare an AI explanation against your actual course material or textbook - they should broadly agree; investigate any conflict.
- For teachers, never treat a single detection tool score as conclusive; always combine it with drafts, process evidence, or a conversation.
- Check your specific institution's current AI policy before starting a piece of work, not after.
How to read the docs and find the truth
Start with your own institution's official academic integrity or AI-use policy document - this is the actual authority for your situation, not a general internet opinion. For subject-specific guidance, professional bodies (exam boards, subject associations) increasingly publish their own AI guidance, and these are more authoritative than a teacher's personal blog. For how a specific AI tool works or its limitations, check the provider's own documentation and any "known limitations" or safety pages on their site (openai.com, anthropic.com, google.com) rather than social media claims. For citation accuracy, always verify against a genuine academic database (Google Scholar, your institution's library catalogue, JSTOR) - never trust a citation because it looks well-formatted.
Your one-page playbook
TASK: [what you're working on] GOAL: learn the skill / check understanding / save admin time (pick one, be honest) WHAT I ALREADY KNOW: [so it doesn't just repeat that back] WHAT I WANT AI TO DO: explain / quiz / give feedback / generate practice (not: write it for me) POLICY CHECK: [confirmed against current institution policy - yes/no] VERIFY: [citations checked independently, understanding tested by explaining aloud]
Check yourself
- What's the difference between using AI to check your understanding versus using it to avoid doing the thinking?
- Why should you never copy an AI-suggested citation straight into a bibliography?
- Why is it unfair, as of 2026, to treat an AI-detection tool's score as definitive proof either way?
- Name one prompt strategy that keeps essay feedback useful without letting AI do your writing for you.