What the examiners said about IB Maths AI SL — and exactly what to do about it
Built from four of the IB's own AI SL subject reports (May 2024 TZ2, November 2024 TZ2, November 2025 TZ1 and TZ3), read in full. They were supplied by the site owner; the IB makes subject reports available to IB World Schools through its Programme Resource Centre. We have not seen every timezone's report, and none from May 2025. Everything here is in our own words: we do not reproduce the reports or any exam question, and every worked example is our own. The question topics are our own matching from the reports' comments (a few parts on topics outside our topic list are marked "not mapped"); we show no marks per question because we did not use the markschemes.
Summaries are in our own words with links to the originals. Predictions are informed guesses, not a guarantee. You'll meet these notes as you revise too: under practice questions, inside mark schemes, in your AI marking and in “What to do next”.
Ranked for May 2027
What to focus on for May 2027
November 2026 candidates can use the same list. Ranked by how often each area appeared and how weakly it was answered.
Informed prediction, not a guarantee: any topic on the specification can be examined.
1
Statistics: hypothesis tests, probability and distributions
The largest share of the questions we mapped in every session; the reports repeatedly name χ²/t-test set-up and conclusions, conditional and without-replacement probability, and binomial/normal bounds.
A finance question with regular payments appears in every session and loses marks on TVM signs and periods; sequence questions confuse terms, sums and n; accuracy rules cost marks throughout.
Calculus: tangents and normals, rates, trapezoidal rule, optimisation
Named as difficult in every session: not differentiating to find a gradient, interpreting a derivative, the trapezoidal rule and unscaffolded optimisation.
Models: exponential, sinusoidal and inverse notation
Exponential and log models, sinusoidal parameters and function notation cost marks in every session, mostly through interpretation rather than calculation.
The 10 mistakes that cost the most marks (and 7 more)
Ordered by how often and how widely they were reported. The first 5 are free in full; Pro unlocks the fix and a worked example for every other one.
#1Accuracy: 2 s.f. answers, early rounding, and ignoring the accuracy asked for
What examiners saw: In every AI SL report we read. Answers were given to 2 s.f. (or truncated) with no unrounded value shown, which scores nothing if the method is not visible; intermediate values were rounded and reused; and questions that asked for 4 s.f., 2 decimal places, the nearest dollar or a whole number of years were answered to 3 s.f. out of habit. A value such as 0.340 written as 0.34 is only 2 s.f.
Fix: Write the full calculator value, then round once: 3 s.f. unless the question says otherwise (keep trailing zeros). Read each question for its own rounding instruction and for context (whole years, whole items).
Worked example (our own): A ramp rises at 23.5° over a horizontal distance of 12 m. Find the height, then the area of the triangular side.
Loses marks
h = 12 tan 23.5° = 5.2 m
Area = ½ × 12 × 5.2 = 31.2 m²
Earns the marks
h = 12 tan 23.5° = 5.2177… m (keep in the GDC)
Area = ½ × 12 × 5.2177… = 31.306… ≈ 31.3 m²
Why: Rounding h to 2 s.f. changed the final answer at 3 s.f. Only the final answer is rounded.
#2No working shown: bare GDC answers and calculator notation
What examiners saw: All four reports: a wrong answer with no working scores zero, and a correct answer on its own is not always full marks. Examiners saw calculator syntax and E-notation output instead of mathematical notation, missing diagrams for normal questions, no supporting calculations for Voronoi distances, and formulas substituted into without writing them first. When two answers are given, only the first is marked.
Fix: Write the formula or distribution first (X ~ B(n, p), P(X ≤ 3) = …), then the unrounded value, then the rounded answer. Convert calculator E-notation to a × 10ⁿ. Cross out the attempt you do not want marked.
Worked example (our own): X ~ B(12, 0.35). Find P(X ≤ 3).
Loses marks
binomcdf(12, 0.35, 3) = 0.35
Earns the marks
X ~ B(12, 0.35)
P(X ≤ 3) = 0.34665…
≈ 0.347 (3 s.f.)
Why: Calculator syntax is not accepted as working, and 0.35 is only 2 s.f. With the distribution and unrounded value written, a slip can still earn the method mark.
#3χ² and t-tests: hypotheses, the comparison and the conclusion in context
What examiners saw: Recurring in every session: hypotheses that leave out the variables or the population means, "dependent" instead of "not independent", a two-tailed alternative for a one-tailed test, the pooled/unpooled setting mixed up, a p-value misread from E-notation (so greater than 1) or given to 1 s.f., no comparison with the significance level (which loses the mark), a correct comparison followed by the wrong decision, and a generic conclusion where one in context was asked for. Expected frequencies were also left at the wrong accuracy.
Fix: Use a fixed layout: H₀ and H₁ in symbols with the population parameter (μ₁ = μ₂; μ₁ > μ₂) or in words naming both variables for χ²; the p-value; "p < 0.05, so reject H₀" (or not); then one sentence in context.
Worked example (our own): A t-test compares mean reaction times of two groups, A and B, to test whether A is slower. The GDC gives p = 0.031 at the 5% level.
Loses marks
H₀: the times are the same. H₁: they are different
p is small so A is slower
Earns the marks
H₀: μ_A = μ_B; H₁: μ_A > μ_B (μ = population mean reaction time)
p = 0.031 < 0.05, so reject H₀
There is evidence at the 5% level that the mean reaction time of group A is greater than that of group B
Why: The test is one-tailed and about population means. The decision needs the comparison, and the conclusion must be in context.
#4Finance: compound-interest formula with regular payments, TVM signs and periods
What examiners saw: Every session: the compound interest formula was used where regular payments made it invalid; PV and PMT had the same sign; N was entered in years instead of months; P/Y and C/Y were confused; answers were not given to the nearest dollar or 2 decimal places as asked; and the interest paid was confused with the interest rate.
Fix: With regular payments, use the TVM solver. Money in and money out have opposite signs. N counts payments (years × payments per year). Write down every value you enter (N, I%, PV, PMT, FV, P/Y, C/Y), so a slip can still earn method marks.
Worked example (our own): A loan of $15 000 is repaid monthly over 4 years at 6% per year, compounded monthly. Find the monthly payment and the total interest.
#5Probability: adding instead of multiplying, conditional sample spaces, without replacement
What examiners saw: Candidates added probabilities that should be multiplied (and doubled a probability that should be squared), gave probabilities greater than 1, treated conditional events as independent instead of restricting the sample space, and treated selections without replacement as binomial. Tree diagrams and tables were rarely drawn.
Fix: Draw the tree or table. Along a branch multiply; between outcomes add. For "given that", the denominator is the size of the given group. Without replacement, the second probability changes.
Worked example (our own): A bag has 5 red and 3 blue counters. Two are taken without replacement. Find P(both red).
Loses marks
(5/8)² = 25/64
or 5/8 + 4/7
Earns the marks
P(red then red) = 5/8 × 4/7
= 20/56 = 5/14
Why: Without replacement the second draw has 4 red out of 7. "And" along a branch means multiply.
#6Binomial and normal: recognising the model, the bound, the tail
What examiners saw: Candidates did not recognise a binomial situation (or its assumptions), used the wrong bound for "at least" or "fewer than", did not know E(X) = np although it is in the formula booklet, assumed an expected number must be a whole number, shaded or used the wrong tail of a normal distribution, and used the complementary area in the inverse normal.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#7Calculus: not differentiating, f(a) for f′(a), and the derivative as a rate
What examiners saw: Candidates used a normal's gradient as the tangent's, never differentiated, found f(1) where f′(1) was needed, stopped at f′(1) without writing the tangent's equation, ignored an instruction to use their derivative, stopped at the x-coordinate of a minimum, described a derivative as a height or an average rate, and left out the constant of integration.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#8Basic statistics: range as an interval, frequency tables, outlier fences, Spearman's ranks
What examiners saw: Candidates gave the range as an interval of values instead of one number, confused range with IQR, found the mean of the frequencies or of the values ignoring frequencies, did quartiles and means by hand instead of on the GDC, did not round an outlier boundary to suit discrete data, computed Spearman's coefficient between a variable and its own ranks, mishandled tied ranks, and used r² when r was asked.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#9Exponential and log models: e⁰ = 1, the equation vs the inequality, reading the model
What examiners saw: Candidates treated a term with exponent 0 as zero, rounded b to 1 in f(t) = abᵗ, misread log₁₀x in a model, set an equation where the context needed an inequality (and then rounded the wrong way), gave calculator mantissas as answers, and commented on a model without using its equation.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
What examiners saw: Candidates used the arithmetic formula in a geometric situation, found a sum when a term was needed (or the reverse), wrote u₄₂ for "the term equal to 42", found an extra term, found a nearby term instead of the one asked for, did not round n up to a whole number of days, and listed terms instead of using the formula.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#11Trapezoidal rule and integrals: heights, strip width and dx
What examiners saw: Candidates substituted x-values instead of heights, used the wrong strip width (for example the number of values instead of the number of strips), misread the booklet formula, substituted limits when an indefinite integral was asked, omitted dx, or used limits outside the domain. Exact answers were rarely given when asked.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#12Interpreting in context: parameters, derivatives and justifications
What examiners saw: Weaker candidates could not say what a gradient or intercept means in context, gave only one of the two when comparing costs, described a regression gradient loosely, explained sampling weaknesses too briefly to show understanding, and justified a choice with reasons that did not use the model or the numbers.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#13Voronoi diagrams and bisectors: midpoint, negative reciprocal, vertices
What examiners saw: Candidates used a site instead of the midpoint, did not take the negative reciprocal, rounded the gradient early, wrote a point or the wrong line instead of the equation of a vertical bisector, found an intersection of given edges instead of the required vertex, and could not find a bearing or a meeting point of two bisectors.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#14Triangles and 3D: wrong rule, wrong angle, missing units
What examiners saw: Candidates assumed a right angle that was not there (or used the sine rule in a right-angled triangle), found an angle other than the one named, used the wrong radius or only one of two areas, missed an arc length in a curved surface, forgot units where they were required, and did not give an answer in standard form when asked. Drawing a diagram helped.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#15Sinusoidal models: amplitude, period vs b, principal axis
What examiners saw: Candidates used the period as b, gave a negative amplitude, confused the minimum with the principal axis, left out "h(t) =" when writing the model, found only some of the times that satisfied a condition, and did not use diagrams for circular motion.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#16Function notation and answer form: f⁻¹(a), ranges and inequalities
What examiners saw: Candidates evaluated the function when its inverse was asked for, read f⁻¹(−2) as 1/f(−2) or f′(−2), wrote the range of a rational function as the region between two values when it was outside them, and gave two numbers where the answer was an interval.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
#17Exploration (IA): templated statistics, research questions and reflection only at the end
What examiners saw: The IA sections of all four reports: statistics dominate, often templated; a research question, rationale or personal hypothesis is not needed (a clear aim is); χ² tests are set out incorrectly (significance level stated late, expected values missing, expected frequencies below 5 not handled); models are chosen by R² or listed before the data are seen; outliers are removed without reason; GDC screenshots and calculator notation replace proper communication; and reflection is a list of generic limitations at the end.
Pro: the fix and a worked example (wrong vs right method) for this mistake are in Pro. See plans
How often each topic appeared, and for how many marks
Every question in the papers we read, matched to a topic. Columns marked * count questions (q) because the marks could not be verified from the mark scheme. Hover or tap a cell to see the questions.
6q (Paper 1 (TZ2) Q1, Paper 1 (TZ2) Q3, Paper 1 (TZ2) Q6, Paper 1 (TZ2) Q8, Paper 2 (TZ2) Q1, Paper 2 (TZ2) Q4)
6q (Paper 1 (TZ2) Q2, Paper 1 (TZ2) Q7, Paper 1 (TZ2) Q9, Paper 1 (TZ2) Q11, Paper 2 (TZ2) Q1, Paper 2 (TZ2) Q3)
12q (Paper 1 (TZ1) Q4, Paper 1 (TZ1) Q7, Paper 1 (TZ1) Q9, Paper 2 (TZ1) Q1, Paper 2 (TZ1) Q3, Paper 2 (TZ1) Q4, Paper 1 (TZ3) Q1, Paper 1 (TZ3) Q3, Paper 1 (TZ3) Q4, Paper 1 (TZ3) Q7, Paper 1 (TZ3) Q10, Paper 2 (TZ3) Q2)
4q (Paper 1 (TZ2) Q2, Paper 1 (TZ2) Q11, Paper 2 (TZ2) Q2, Paper 2 (TZ2) Q5)
4q (Paper 1 (TZ2) Q1, Paper 1 (TZ2) Q4, Paper 2 (TZ2) Q1, Paper 2 (TZ2) Q5)
9q (Paper 1 (TZ1) Q2, Paper 1 (TZ1) Q6, Paper 1 (TZ1) Q12, Paper 2 (TZ1) Q5, Paper 1 (TZ3) Q5, Paper 1 (TZ3) Q11, Paper 2 (TZ3) Q2, Paper 2 (TZ3) Q3, Paper 2 (TZ3) Q5)
4q (Paper 1 (TZ2) Q3, Paper 1 (TZ2) Q8, Paper 1 (TZ2) Q10, Paper 1 (TZ2) Q13)
5q (Paper 1 (TZ1) Q1, Paper 1 (TZ1) Q2, Paper 1 (TZ1) Q5, Paper 2 (TZ3) Q1, Paper 2 (TZ3) Q3)
3/3
Confidence: high for the series shown — every question was read and mapped by hand. "Series" = how many of the 3 series examined the topic.
Paper by paper
Every paper we analysed, question by question
Topics are ours (matched from each report's question-by-question comments). We don't reproduce the questions: open the official paper from our past-paper page.
Parallel & perpendicular lines; Constructing a Voronoi diagram
Q3
Trapezium rule (numerical integration); Definite integrals & area under a curve
Q4
Continuous distributions (normal on GDC); Chi-squared goodness-of-fit test
Q5
Volume & surface area of 3D shapes; Optimisation problems
Pro
Your mistakes vs the examiners' hot-spots
We set the topics in My mistakes and your practice scores against this page's mistakes library and focus list, and rank where they overlap, with a practice link for each.
Free account: how many of your notebook entries match, and the top one. Pro: the full ranked list, your focus areas and your risk list.
A one-page PDF of the 10 most common mark-losing mistakes and how to avoid them. Print it and stick it inside your revision folder.
Sources
Where this comes from
Built from four of the IB's own AI SL subject reports (May 2024 TZ2, November 2024 TZ2, November 2025 TZ1 and TZ3), read in full. They were supplied by the site owner; the IB makes subject reports available to IB World Schools through its Programme Resource Centre. We have not seen every timezone's report, and none from May 2025. Everything here is in our own words: we do not reproduce the reports or any exam question, and every worked example is our own. The question topics are our own matching from the reports' comments (a few parts on topics outside our topic list are marked "not mapped"); we show no marks per question because we did not use the markschemes.
What do examiners say students get wrong in IB Maths AI SL?
The most repeated points are: Accuracy; No working shown; χ² and t-tests. Each one on this page has the fix and, where free, a worked example showing the wrong and right method.
Which IB Maths AI SL topics come up most?
In the material we analysed, the biggest areas were Statistics and probability, Geometry and trigonometry, Functions.
Is this a prediction of the May 2027 paper?
No one outside the exam board knows what will be on the next paper. The "focus" list is an informed prediction from how often topics appeared and how weakly they were answered; it is not a guarantee, so revise the whole specification.
Where does this information come from?
We read 4 International Baccalaureate examiner reports (May 2024 TZ2, November 2024 TZ2, November 2025 TZ1, November 2025 TZ3) in full, summarised them in our own words and linked each point to the original.
IB Math Revision