Using an AI chatbot and a purpose-built IA guide platform aren’t the same decision, even when the question you’re typing looks identical. The tool type changes what the AI is actually contributing to your work, and by extension, what the integrity risk is. IB Maths students are currently working with at least three distinct categories: general-purpose chatbots, purpose-built IA guide platforms, and AI features embedded in graphing or CAS calculator environments. Treating them as interchangeable means making an academic honesty judgement without realising you’ve made one.
It’s worth knowing what a purpose-built IA platform actually looks like in practice. A February 2026 Revision Village Blog post on Mathematics IA Resources describes an IA-specific resource suite covering the full exploration journey, from brainstorming to finalising the written work, with an IA Support Hub that includes guidance on academic honesty, responsible AI use, and mathematical software. That’s structurally different from typing into a general chatbot or running a calculator solver.
The ethical stakes rise as a tool moves from supporting your thinking toward shaping or partially generating assessed work. Both cases come down to the same two criteria: whether AI is scaffolding your reasoning or replacing it, and whether anything it produces has been independently verified against your own understanding.
The Core Decision Rule: Scaffold vs Replacement
Academic integrity in IB Maths doesn’t usually collapse all at once. It erodes at a specific crossing point: when AI stops serving the student’s thinking and starts substituting for it. That boundary – scaffold versus replacement – is where the decision gets made, often quietly, often mid-session. An Oxford University case study on AI use in an IB school found that AI can support scaffolding, practice, and self-monitoring, but that feedback accuracy is uneven and effective use depends on teacher mediation and clear guidance. The finding cuts both ways: AI works as a thinking aid, less reliably as a thinking substitute. The assessed work must come from your own mathematical reasoning, with AI in a support role rather than as an invisible co-author. That line is easier to cross than most students expect.
Before submitting anything AI has touched, use this quick self-audit to make the scaffold/replacement call concrete.
- Assessed-source check: Is AI contributing to the assessed maths, explanation, or reflection itself – not just planning or understanding? If yes → replacement risk.
- First-attempt check: Did you attempt the problem, derivation, or IA plan before seeing any AI solution path? If not → replacement risk; the work needs to be done independently.
- Rebuild check: Could you reproduce the full reasoning tomorrow – the steps and why each one is valid – without the AI chat open? If not → treat this as replacement and redo the work yourself.
- Voice/wording check: Did any AI-generated wording become your final explanation or reflection with only light editing or paraphrasing? If yes → replacement risk; the authorship of that text is no longer clearly yours.
- Record-keeping check: Have you noted the date, tool, what you asked, what you kept – idea or check only versus solution text – and one verification method used? If not → do it before you close the session; a 30-second note made now is worth far more than a reconstruction later.
Evaluating Grey Areas
Borderline AI uses often look similar on the surface even when their integrity risk is very different. Checking symbolic algebra with AI before submitting worked solutions is one of the more common grey areas. If you’ve already solved the problem independently and use a tool only to catch transcription slips or sign errors, your reasoning remains the source of the solution and AI acts as a scaffold. Feed the question straight into a chatbot and copy over its procedure or answer, and the first-attempt and rebuild checks both fail. AI has replaced your mathematical work, not assisted it.
Using a chatbot to explain a mathematical concept rather than immediately asking a teacher is generally a scaffold use – but only when you can restate and, where appropriate, re-derive the idea yourself, and cross-check it against trusted resources. When you treat the explanation as something to interrogate and compare with your notes or textbook, your understanding stays in charge. Accept it without that check, and your apparent grasp of the concept comes from the model rather than your own reasoning. That gap surfaces in the IA, and in the examination.
The rebuild check is what separates genuine statistical thinking from statistical theatre in the IA. Asking AI to suggest appropriate statistical tests for an exploration dataset can support planning – but only if you already have a rough sense of suitable methods and can evaluate each suggestion against your understanding of the variables, assumptions, and sample size. When you can justify why the chosen test matches your research question, and what would count as evidence against it, AI has sharpened your thinking. If instead you let AI select the test and design decisions you cannot later justify – failing the rebuild check on why this test, with which assumptions – the AI is quietly directing the mathematical core of the exploration, not supporting it.
At examination, the gap between ‘written in my own words’ and ‘genuinely my thinking’ is very hard to disguise. Prompting AI to draft model IA reflections and then paraphrasing them is a replacement use even when every sentence has been reworded. Reflective commentary is an assessed component precisely because it is meant to reveal your reasoning process – not the model’s. When AI generates the underlying structure and ideas, the voice/wording check fails regardless of how much surface editing follows. The words change; the author doesn’t. The only acceptable role for AI in this scenario is helping you understand the assessment criteria, not drafting content you reshape into your submission.
The Hidden Risk in AI Mathematical Output
Even on the scaffold side of the boundary, AI-produced mathematics carries an accuracy problem that’s easy to underestimate. A 2025 arXiv preprint analysing state-of-the-art models on high-school-level word problems found persistent errors in arithmetic, spatial reasoning, and multi-step deduction – including cases where the final numeric answer is correct but the intermediate reasoning is flawed. In IB Maths terms, that failure mode might look like a misapplied integration technique, a sign error in a differential equation, or an algebraic step that doesn’t actually follow – all producing a result that looks plausible until it doesn’t. If you’re not yet confident in the topic, you’re poorly placed to spot these flaws. Uncritical reliance on AI output combines an integrity problem with a practical one: incorrect methods can travel undetected into revision notes and assessed work.
That’s why verification has to mean independent mathematical checking – not asking another model whether the first one was right. Strong verification treats AI output as a hypothesis. You re-derive key steps yourself, substitute results back into original equations to check they hold, confirm domains and constraints, and where possible reach the same conclusion by a different route – algebraic, graphical, or numerical. Weak verification leaves hidden reasoning errors untouched. Re-prompting the same chatbot, switching tools without doing your own checks, matching only the final answer to a markscheme while not understanding the path – none of these close the gap. A reliable rule: if you can’t name at least one independent check you performed, you haven’t verified the AI’s maths, and it shouldn’t appear in assessed work.
Using AI to Build Understanding
The practical difference between an AI interaction that helps and one that harms isn’t which tool is open – it’s whether the verification and rebuild behaviours described above are actually happening. When they are, the same tools that create problems when accepted passively become genuinely useful. You might take a worked example from a chatbot and actively search for errors or alternative approaches, turning the model’s known unreliability into practice material. You might read an AI explanation of a concept alongside your textbook and class notes, checking where they align and where they diverge, until you can express the idea cleanly in your own words. For IA planning, you might use AI to generate a wide list of possible investigation topics, then independently research, narrow, and refine those options against your interests and mathematical strengths.
All three patterns keep your judgement at the centre. They mirror strong verification behaviours: rebuilding arguments yourself, cross-checking against reliable sources, comparing different solution paths rather than trusting the first convincing output. In that mode, AI functions as a thinking scaffold – one that stretches and tests your understanding rather than quietly doing the work for you.
Carrying the Framework into New AI Situations
New AI tools will keep appearing throughout your IB Maths course, each arriving with its own implicit promise of convenience. The core question stays the same: is this scaffolding your reasoning, or replacing it?
Apply that test consistently. Keep a brief record of how you used the tool, and independently verify any mathematical output you rely on.
The IB assessor reading your exploration isn’t evaluating the tool you used. They’re evaluating the thinking they find on the page.
