Pedagogy
Most arguments about AI in a calculus class start with the tool. The useful question is narrower: which parts of this course have to stay in the student’s head, and what is left over once that is settled? Here is where the public rules actually draw the line, five uses that keep the reasoning with the student, and three traps that quietly take it away.
This cycle we read the public rules rather than arguing about them: College Board’s AI guidance for 2026–27, its FAQ on acceptable use, the AP Exams calculator policy and the AP Calculus AB exam page — plus, for the classroom side, Cornell University’s institutional teaching guidance. Every quotation below was read on 18 October 2026 and is linked at the end with what it was used for. The five uses and three traps are our editorial opinion, and they are labelled as such; the rules they sit on are not.
The AP Calculus AB exam is 42 multiple-choice questions in 1 hour 40 minutes and 6 free-response questions in 1 hour 30 minutes, each section worth half the score. Read the parts and the split is not subtle:
Section I, Part A: 29 questions, 62 minutes, calculator not permitted.
Section II, Part B: 4 questions, 60 minutes, calculator not permitted.
Add it up. 122 of the exam’s 190 minutes — 64% — are sat without a calculator. If questions are weighted equally inside each section, those parts carry about 68% of the exam’s points: (29/42 × 50) + (4/6 × 50) = 34.5 + 33.3 = 67.9. The arithmetic is ours, from the published question counts and section weightings; the counts themselves are College Board’s.
Nor are these the easy parts. The no-calculator questions are where setup, algebra and justification live, and the free-response half without a calculator is the longer one: four questions in sixty minutes against two in thirty. That is our reading of the format, and it has a practical consequence for any classroom policy — a tool that removes the algebra has not removed the easy part of this exam, it has removed the part being tested.
The calculator policy is specific in the same way. Only approved handheld models are permitted, students may bring up to two, and “calculators may not be shared”. Alongside them sits the built-in graphing calculator in the testing app, which the policy says will be available only in the calculator-required parts, because “these exams continue to have parts where no calculator is allowed”. And the scope is closed explicitly: calculators may be used on the listed exams “and on no others”. On exam day there is no category of tool beyond what that page names.
One date matters for planning. College Board states that the number of multiple-choice questions and the timing were updated, “effective starting with the May 2027 exams”, and that course content has not changed. The 42-question, 100-minute shape above is the 2027 shape, and the exam sits on Monday 10 May 2027.
College Board’s AI guidance is written course by course, not as one blanket rule: “The following guidance on AI use is provided for specific courses.” The 2026–27 page carries policies for AP African American Studies, Art and Design, Capstone (Seminar and Research), Computer Science Principles and World Languages and Cultures. AP Calculus is not among them. That is worth saying plainly, because a great deal of staffroom argument assumes a calculus-specific AI rule exists, and as of 18 October 2026 it does not: exam-day conduct is governed by the general exam rules and the calculator policy above.
Where the Board does write a rule, the shape is consistent, and it is about purpose rather than about tools. The framing: AI “must be used ethically, responsibly, and intentionally to support student learning, not to bypass it”. Students “must read primary and secondary sources directly, perform their own analysis and synthesis of evidence, and make their own choices on how to communicate effectively”. The FAQ puts the floor where everyone can see it: “Students may not use AI tools to write or create their assignments for them.” Checking is allowed — “checking their writing — but not rewriting — for grammar and tone”.
Two course policies are worth borrowing even though neither is about calculus. AP Capstone requires interim checkpoints with the teacher, whose purpose is to “make their thinking and decision-making visible”, and “failure to complete the checkpoints will result in a score of 0 on the associated task”. AP Computer Science Principles lists what AI output can be — “incomplete code, code that creates or introduces biases, code with errors … or code complexities that make it difficult to understand and therefore explain the code” — and then puts the obligation on the student: review and understand anything co-written, and “be prepared to explain their code in detail, as required on the end-of-course exam”.
Our reading, not a College Board statement: translate that into calculus and the test is not “did you use a tool”. It is “can you explain and defend the step”. Everything below is built on that one sentence.
Each of these keeps the reasoning with the student and puts the tool where it does least harm — asking and checking rather than producing. Each ends with the rule we would write on the board.
Have students write their own explanation before anything is pasted anywhere: why the derivative of the area-so-far function is the curve itself, why a definite integral can come out negative, when substitution is the right first move. Then, and only then, they may ask for gaps in their paragraph. Our explainer of the Fundamental Theorem is written to be the thing they argue with first.
The tool may critique your explanation. It may not write it.
Give a worked solution with exactly one mistake in it and nothing to mark it out. For example:
∫ x ex² dx = ex² + C
The answer is missing its factor of ½, and nothing in it looks wrong. Differentiate back and it shows: d/dx of ex² is 2x ex², twice the integrand, so the correct antiderivative is ½ex² + C. A limit version of the same exercise: “limx→0 sin(3x)/x = 1”. Test it with a small number — at x = 0.001 the quotient is sin(0.003)/0.001 ≈ 2.999996, not 1 — and the true limit is 3. The error is found by checking, which is the skill that transfers.
Find it, name the rule that broke, then solve a fresh one with no help.
Related rates are a translation exercise before they are a calculus exercise: a 10 ft ladder leans on a wall, the foot slides away at 1 ft/s, how fast is the top falling when the foot is 6 ft out? The relation is x² + y² = 100, so at x = 6, y = 8; differentiating gives 2x x′ + 2y y′ = 0, so y′ = −(x/y)x′ = −(6/8)(1) = −0.75 ft/s: the top descends at three-quarters of a foot per second. Let the tool offer three wordings of the same relation; the class rebuilds the equation from scratch and defends each symbol.
The tool may translate. The equation is yours.
The student supplies the unit list — limits and continuity, derivative rules, applications of the derivative, integration, differential equations — and the tool drafts quick questions against it. The answering happens closed-book, and the marking happens against the textbook. Check the answer key before you use it: a wrong key is worse than no key, and a generated key is exactly the kind of output the Board’s own list warns about.
Generated questions, unaided answers, checked key.
Free-response credit usually hangs on a clause: “… because f is continuous on [a, b] and differentiable on (a, b)”, or “… because the integrand changes sign at x = 2”. Students write the sentence; the tool is allowed to flag that a reason is missing or that a hypothesis has gone unstated; the student rewrites it in their own words.
The reasoning words have to be the student’s own.
Wrong algebra does not announce itself. A dropped factor, a sign that flipped inside a substitution, a rule used where its hypothesis fails — all of it comes back in the same fluent voice as the correct work. College Board’s own description of AI failure modes in code is a fair description of what goes wrong in algebra too, though that is our analogy, not their claim. The cheap defence is a two-second check: differentiate back, substitute a number and see if the sign and size make sense, test the endpoints. Require the check in the homework itself, not as a suggestion.
Reading a correct solution feels like understanding it. Cornell’s guidance summarises the research bluntly: while some studies show learning gains from student use of generative AI, “especially in math or computer science courses, … often those gains have proved ephemeral”. That sentence is Cornell’s summary of the studies it names, and it matches what the exam format implies: the 62-minute and 60-minute no-calculator stretches do not care how familiar a method looked on a screen. Closed-book retrieval, short no-tool quizzes, and a rule that a tool-assisted homework problem earns credit only with a written unaided re-do are the obvious counter-measures.
If the take-home permits tools and the exam does not, the grade stops predicting the score, and the student finds out in May. AP Capstone’s checkpoints are the cleanest public precedent for the fix: make the thinking visible in the room, in short conversations, rather than inferring it from a submitted file. One in-class no-tool item a week, a whiteboard checkpoint, and grading the justification as well as the answer all move the same direction.
Our opinion, and it follows from the format rather than from taste:
It earns the twenty minutes twice over. Checking is the skill that survives contact with a tool, and explaining is what the exam already asks for.
Five limits, so the page can be trusted rather than just useful. We are not stating a policy for any school or district — your department’s rules and your school’s academic-integrity code govern, and this page is not advice about either. We are not claiming College Board’s guidance is settled: it is labelled for the 2026–27 year and can change, which is why every quotation carries the date we read it. We are making no claim about detecting AI use, and we would not build a class around detection — that is opinion, and it is why the mitigation here is task design. We name no tools, vendors or prices, deliberately: this is about what to ask students to do, not about what to buy. And the five uses, three traps, hand-work list and task template are editorial opinion, not research findings — the one research sentence quoted is Cornell’s summary of the studies it cites. Nothing above changes exam-day rules: on exam day, the calculator policy is the whole of the law, and it is narrower than any classroom arrangement described here.
Keep reading
Practice · 7 min read
The free-response answers are written by hand in paper booklets while the questions sit on a screen — what College Board has published about how that works, and the habits that follow from it.
Free, no spam
Roughly twice a month. Unsubscribe any time.
One email address is all we ask for. New articles, roughly twice a month, and nothing else.