IB Maths AI IA: Criteria, Topics and Structure Guide
By Michael Thompson · Former IB Diploma Programme coordinator; 10 years at Bromsgrove School · Published 4 September 2026
The IB Math AI IA - formally the Mathematical Exploration - accounts for 20 marks and a significant portion of your final grade, so understanding exactly how it is assessed matters before you write a single word. From May 2027 onward, the course is updated and SL and HL now share identical assessment criteria, each worth 20 marks in total across four criteria. The IA asks you to investigate a real-world question using the mathematical tools of Applications and Interpretation: statistics, modelling, probability, and technology-driven analysis. This guide covers the criteria, topic ideas, structure, and the specific things examiners reward - and penalise - so you can plan your exploration with confidence.
Key Takeaways
- Four criteria, 20 marks total.: From the updated curriculum, the IA is marked on Problem Specification (4), Abstraction (6), Computation (4), and Interpretation (6), with SL and HL now sharing identical criteria.
- Real data and technology are central to AI explorations.: Examiners expect you to engage meaningfully with genuine data sets and appropriate digital tools, but technology use alone does not earn marks - your mathematical reasoning must drive the analysis.
- HL students must demonstrate greater mathematical sophistication.: Although the criteria descriptors are identical across SL and HL, examiners assess the level of complexity and abstraction expected for the course level you are studying.
- Reflection is embedded in Interpretation.: Examiners want you to evaluate your model's limitations, consider what the results actually mean in context, and discuss what you would do differently - not just report answers.
- Length guidance is roughly 12-20 pages.: There is no strict word count in the updated criteria, but explorations that are too short risk missing depth while very long ones often contain padding that dilutes the mathematical focus.
- The most common mistake is a question that is too broad or too trivial.: A focused, specific research question - one that requires genuine mathematical work to answer - is the single biggest factor separating strong explorations from weak ones.
In This Article
- What the IB Math AI IA is and how it is weighted
- The four IB Math AI IA criteria explained
- IB Math AI IA topic ideas: SL and HL directions
- How technology use is assessed in an AI exploration
- What examiners want from the Interpretation criterion
- IB Math AI IA structure, length, and common mistakes
- What to do next
1. What the IB Math AI IA is and how it is weighted
The IB Math AI IA (Internal Assessment) is a piece of independent mathematical work formally called the Mathematical Exploration. It sits at the heart of the DP Mathematics: Applications and Interpretation course, and it counts towards your final grade whether you are taking SL or HL. Unlike the written papers, the Exploration is submitted during the course rather than sat in an exam hall, which means your grade here is locked in before May examinations begin.
From the updated curriculum with first assessment in May 2029, SL and HL share identical IA criteria totalling 20 marks, broken down as follows, per the IB Mathematics: Applications and Interpretation updates page:
| Criterion | Focus | Marks |
|---|---|---|
| A: Problem specification | Defining a clear, tractable question | 4 |
| B: Abstraction | Building a mathematical model | 6 |
| C: Computation | Accuracy and appropriate use of technology | 4 |
| D: Interpretation | Connecting results back to the real context | 6 |
One counter-intuitive point worth noting early: SL and HL students are assessed against the same criteria and the same mark totals. The distinction between levels shows up in the sophistication of mathematics expected, not in a separate marking grid. A stronger mathematical treatment is therefore the only lever available to HL students aiming to separate themselves from SL candidates on this component.
Because the Applications and Interpretation course emphasises modelling and real-world data over abstract proof, the Exploration is a natural fit. The criteria reward exactly the skills the course builds: framing a genuine problem, constructing a workable model, computing results with technology, and drawing conclusions that hold up to scrutiny.
2. The four IB Math AI IA criteria explained
!IB Math AI IA criteria scoring grid showing four criteria and their marks out of 20.webp6007
Per the International Baccalaureate, the ib math ai ia is marked out of 20 across four criteria. SL and HL students are assessed against identical descriptors - the distinction in practice is the mathematical sophistication expected at each level, not a separate mark scheme.
| Criterion | Focus | Marks |
|---|---|---|
| A: Problem specification | Defining the question and context | 4 |
| B: Abstraction | Mathematical modelling and representation | 6 |
| C: Computation | Accuracy and appropriate use of technology | 4 |
| D: Interpretation | Making sense of results in context | 6 |
Criterion A: Problem specification (4 marks)
Examiners want a focused, self-contained question - not a broad topic. "How does music streaming affect artist revenue?" is a topic. "Can a linear regression model predict monthly Spotify royalties from stream count for independent UK artists?" is a question. Marks are lost when the aim is vague, changes mid-exploration, or could be answered without any real mathematics.
Criterion B: Abstraction (6 marks)
This is the highest-weighted criterion alongside D, and the one where mathematical depth matters most. Examiners look for a clear, justified choice of mathematical model or structure. A common mistake is describing data collection at length without explaining why a particular model fits the situation. For HL students, a superficial model that an SL student could equally produce will not reach the top band, even if it is correctly executed.
Criterion C: Computation (4 marks)
Accuracy counts, but so does process. Showing only a GDC output with no explanation of the steps taken earns partial credit at best. Examiners expect you to demonstrate that you understand what the technology has produced, not just that you can copy it across.
Criterion D: Interpretation (6 marks)
Results must connect back to the original question in context. Saying "the correlation coefficient is 0.87" is computation. Saying what that means for your specific dataset, noting the limitations of that figure, and assessing whether your model answers the original question - that is interpretation. Section 5 of this article examines this criterion in more depth.
3. IB Math AI IA topic ideas: SL and HL directions
The IB Math AI IA is designed around real-world data and applied modelling, which makes certain topic areas a natural fit. Statistics, regression and correlation, probability modelling, financial mathematics, and structured data analysis all align closely with the course's emphasis on contextual mathematics.
The counter-intuitive trap many students fall into is choosing a topic that is interesting but data-poor. A research question needs enough data points to run meaningful analysis, and enough mathematical structure to satisfy the Abstraction criterion (worth 6 marks). Personal surveys of 20 classmates rarely do either.
Topic areas that suit the AI course
- Statistics and data analysis: comparing distributions across real datasets, for example sports performance metrics or public health records
- Regression and correlation: fitting linear, quadratic, or exponential models to collected data with residual analysis
- Probability modelling: using binomial or normal distributions to model real phenomena
- Financial mathematics: loan amortisation, investment growth, or currency exchange modelling using compound interest models
Example research questions (written for this guide)
SL-level examples:
- "How well does a linear regression model predict monthly electricity consumption from average temperature in the UK?"
- "To what extent does a normal distribution model the distribution of resting heart rates among 16-18 year olds?"
HL-level examples:
- "How accurately can a logistic growth model, compared to a linear regression, describe the cumulative adoption of electric vehicles in Norway between 2010 and 2023, and what do the residuals reveal about the model's assumptions?"
- "Using exponential smoothing and regression analysis, how reliably can consumer price index data from three economies be modelled, and what are the limitations of extrapolating beyond the dataset?"
The HL examples add a second mathematical layer (model comparison, residual critique, multi-variable consideration) and require explicit evaluation of model assumptions. That critical reflection is what separates a strong HL exploration from a competent SL one.
Curriculum changes affecting topic choice
From the updated curriculum with first assessment May 2029, several HL topics have been removed: Poisson distribution, hypothesis tests for the population mean (both normal and Poisson), and hypothesis testing for the correlation coefficient are no longer part of the HL syllabus. Building an ib math ai ia around any of these for post-2029 cohorts creates a mismatch with what examiners will be equipped to assess.
4. How technology use is assessed in an AI exploration
The Applications and Interpretation course is built around technology. GDCs, spreadsheets, and statistical software such as Desmos, GeoGebra, or R are expected tools, not optional extras. Examiners know this, which is why the "Use of Mathematics" criterion does not reward you for running a tool. It rewards you for showing that you understand what the tool did and why.
Generating output is the starting point, not the finishing line. The question the criterion is really asking is: can you reason about the mathematics behind the result, justify why you chose this method, and engage critically with what the output means?
The gotcha most students miss
A high R² from a regression does not automatically mean the model is good. A student who reports R² = 0.94 and concludes "the model fits well" has produced a calculator screenshot, not mathematical analysis. The R² value alone cannot tell you whether the relationship is actually linear, whether a single outlier is inflating the fit, or whether the model would hold outside the data range. Examiners flag this pattern repeatedly, and it costs marks across multiple criteria, not just one.
Weak vs stronger approach: regression example
(These examples are written for this guide to illustrate the contrast.)
| Dimension | Weak approach | Stronger approach |
|---|---|---|
| Method choice | Runs linear regression without explanation | Justifies why linear regression suits the data structure; considers alternatives |
| Output reported | States R² value and equation | States R², then plots and inspects residuals for patterns |
| Critique | Moves on to the next section | Discusses what the model cannot account for, such as seasonality or confounding variables |
| Technology role | Tool produces answer | Tool produces output; student interrogates it |
The practical takeaway: after any software output, ask yourself one question before writing the next sentence. "What would a reader need to know to trust, or distrust, this result?" Answering that question in your IA is what "Use of Mathematics" at the higher performance bands actually looks like.
5. What examiners want from the Interpretation criterion
Interpretation is worth 6 marks, making it the highest-weighted single criterion alongside Abstraction. The distinction that separates a 5 from a 6 here is not mathematical sophistication. It is whether you treat your results as the start of a conversation or the end of one.
Genuine critical engagement means asking what your maths cannot tell you, not just reporting what it can. That includes discussing the limitations of your model, explaining what the numbers mean in their real-world context, and considering what a different method or data set might reveal.
The contrast between weak and strong reflection is clearest with an example:
- Weak: "My model shows a strong correlation, therefore the relationship is confirmed."
- Strong: "The correlation is strong, but the sample covered only 30 data points from a single region. That restricts generalisation. A larger sample, or a longitudinal data set tracking the same variables over time, would test whether the relationship holds under different conditions."
The second version does three things the first does not: it names a specific limitation, explains why that limitation matters, and proposes a concrete remedy.
One non-obvious pitfall: examiners can tell when reflection has been written after the mathematics was finished and appended as a tidy closing paragraph. Reflection woven into each analytical stage scores higher than a single retrospective block, because it demonstrates that your thinking shaped the exploration rather than tidied it up. Write a limiting comment each time you obtain a result, not once at the end.
6. IB Math AI IA structure, length, and common mistakes
A well-structured ib math ai ia follows a clear sequence that mirrors how a mathematician actually works through a problem:
- Introduction and research question - state what you are investigating and why it is mathematically interesting.
- Background context - explain any real-world setting, define variables, and establish the mathematical tools you will use.
- Mathematical exploration body - this is the bulk of the work: calculations, models, graphs, and GDC or software outputs with your own reasoning woven around them.
- Results and interpretation - draw meaning from your outputs, address limitations, and link findings back to the original question.
- Conclusion - reflect on what the mathematics revealed and what you would do differently.
On length: most successful explorations run between 12 and 20 pages. The IB does not set a formal word count, but examiners note that very short submissions often lack sufficient mathematical development, while padded ones tend to bury the reasoning under redundant graphs and re-stated outputs.
A counter-intuitive gotcha: an exploration with 15 pages of GeoGebra screenshots and no accompanying algebraic reasoning will score poorly on every criterion, not just Technology Use. The screenshots are evidence, not argument.
Most common mistakes to avoid:
- Research question is too vague ("how does maths relate to sport?") or too broad to resolve within 20 pages
- Technology output is copied in without the student explaining the mathematical steps behind it
- Critical reflection is absent or confined to a single sentence in the conclusion
- The mathematics chosen does not connect logically to the stated research question
- Bibliography is missing or incomplete, which affects the Personal Engagement criterion in moderation
7. What to do next
Your most urgent task has nothing to do with choosing a topic. Ask your teacher this week for your school's internal IA submission deadline. Schools typically set this several months before the IBO's external moderation window, and missing it usually means no extension.
Once you have that date, check whether your chosen topic touches any content removed under the post-2027 Mathematics: Applications and Interpretation syllabus update. At SL, the trapezoidal rule is gone; at HL, Poisson distributions and the vector product are out. An exploration built around removed content will still be marked, but your teacher may flag a mismatch during the internal review stage.
Confirm both the deadline and your topic's syllabus alignment with your teacher before you write a single line of your exploration.
FAQ
How many marks is the IB Math AI IA out of?
The IA is marked out of 20 across four criteria - Problem Specification (4), Abstraction (6), Computation (4), and Interpretation (6) - with SL and HL now sharing identical criteria from the updated curriculum.
What is the difference between IB Math AI SL and HL IA expectations?
SL and HL share identical criteria descriptors, but HL students are expected to demonstrate a greater level of mathematical complexity and sophistication in their exploration to access the higher criterion levels.
How long should an IB Math AI IA be?
There is no strict word count, but most successful explorations run to approximately 12-20 pages; below that range it is difficult to demonstrate sufficient depth across all four criteria.
Can I use software like Excel or GeoGebra in my IB Math AI IA?
Yes - technology use is expected in an AI exploration, but examiners want you to justify your method choices and interpret the outputs critically, not simply reproduce calculator or software results.
What are the IB Math AI IA grade boundaries?
Grade boundaries are set by the IBO after each exam session and vary year to year; the official boundaries for each session are published on the IBO results pages and through your school's IB coordinator.
What topics work well for an IB Math AI IA?
Topics that involve real data sets and allow for statistical modelling, regression analysis, or mathematical modelling of real-world phenomena tend to suit the AI course well - the key is choosing a question specific enough to require genuine mathematical work.
References
- Mathematics: applications and interpretation updates - International Baccalaureate® - https://ibo.org/university-admission/latest-curriculum-updates/dp-mathematics-applications-and-interpretation-updates