The September 2026 SAT: the test students couldn't agree on

Harder than usual, but not everywhere. Most students said it was hard, and the ones who struggled sound more shaken than August’s did. The strange part is how much their reports disagree.

Reading and Writing is where they agree. Module 2 drew steady complaints, vocabulary led them for the second test in a row, and plenty of students who expected an easy section didn’t get one. If you walked in treating Reading and Writing as the easy half, you weren’t alone. Math is where reports split. The back half of Module 2 drew the loudest complaints and the claim that it was the easiest part of the day. Two things barely came up at all: timing, and inference questions. Both were big themes in August.

The most common question after this test wasn’t about a question at all. It was which Module 2 did I get. Students answered it by listing the topics they remembered and checking them against someone else’s list.

That doesn’t work. A topic list tells you nothing about routing, because topics differ between forms for reasons that have nothing to do with difficulty. Two students who both got the harder Module 2 can share almost no questions, and a student on the easier one can hit a topic that sounds advanced. The worry underneath is fair. Routing is real, it runs off your Module 1 performance, and the harder Module 2 opens up the top of the scale. So if a hard Module 2 worried you, relax. It means you did well. Hoping a question you missed was unscored is the same move, and it fails for the same reason. Both try to work out a score before scores come out.

Geometry and measurement dominated what students reported from Math. The questions that hurt all shared one structure. Each one gave you a number that you had to convert before the formula would take it.

Take the pyramid question. It gives you the vertical height, the one running straight up the middle. But the lateral faces are triangles, and the height of a triangle is measured up its own face, not up the middle of the solid. That slant height is the square root of the vertical height squared plus half the base edge squared. Nothing about the surface area works until you have it.

The right-triangle question does the same thing. Drop the altitude to the hypotenuse and it cuts the hypotenuse into two pieces. The altitude isn’t their sum and it isn’t their average. You multiply the two pieces and take the square root. If the pieces are 4 and 9, the altitude is 6. Average them instead and you get 6.5, which is close enough to look like a real answer choice.

The mixture question does it with units. It gives concentrations as percent by mass, but the equation needs grams. Twenty percent of a 50 gram sample is 10 grams, and 10 is the number that goes in the equation. Add the percentages straight and you get an answer that is wrong in a way that never looks wrong.

In every one of these, the arithmetic is easy and the conversion is the question. That’s also why students said the calculator helped less than usual. Desmos will finish any of these in seconds once you have written the relationship down. But it can’t tell you that the height you were handed is the wrong height.

So do this before you substitute anything. Say out loud what the number you were given measures, then say what the letter in the formula measures. If those two sentences don’t match, the conversion is the question and you have found the real work.

Solids and triangles carried the section. Circles, proportional reasoning and a two-way table question showed up alongside them. The same setup kept coming back. A figure gets described in words, and you get one measurement that looks like it belongs in the formula. You have to convert it first.

The two-way table question worked the same way. It named a group, and that group becomes your denominator. If it asks how likely something is among the students who chose one option, you divide by that column’s total, not by everyone in the table. The overall total is sitting right there to catch you. Scale questions did it too. Double the sides of a solid and the area goes up four times, the volume eight times.

A smaller group ran backwards. Instead of giving you numbers and asking for an answer, they give you a property and ask which value makes it true. One asked for the constant that leaves a quadratic with exactly one solution, so you set the discriminant to zero and solve. Another asked for the point that can’t satisfy an inequality whatever the constant is. That shape has shown up on every recent form. Treat it as permanent.

Vocabulary carried this section, just like August. The traps were built differently though. August’s wrong answers were near-synonyms, a little too strong or too weak. September’s were often the word pointing the opposite way. That changes how you should work the question.

The clearest case put paradigmatic against idiosyncratic. Paradigmatic means it serves as the typical example of a pattern. Idiosyncratic means it is peculiar to one case. They are close to opposites, so knowing both definitions doesn’t help you. Only the sentence tells you which way the passage is going. The test is quick: find the clause before the blank and ask whether it is setting up a rule or an exception. A rule takes paradigmatic. An exception takes idiosyncratic.

Elusive worked the same way. The competing option described the researchers rather than the thing that kept escaping them, so the trap was about who the word attaches to. Lament split the choices between expressing regret and simply listing, which is a difference in tone, not meaning. In all three, reading the passage for topic leaves you with two answers that both look fine.

The section ran other things too. One question punctuated a series whose items already had commas inside them, which is the case where the separators have to become semicolons or the list stops being readable. Another put a blank in a sentence that already had its main verb. That means the answer had to be a participle like having decided, not a second finite verb like decided. And a literary excerpt asked what one sentence was doing rather than what it said. In those, the choices are jobs: conceding a point, setting up a contrast, giving an example.

Original questions written around what this exam tested, one for each argument above. Nothing here is a real test question.

One habit covers most of what hurt on this form. In both sections, the hard questions turned on reading a given correctly before you used it. Which height. Which denominator. Which direction the sentence points. None of them needed a harder method than the ones you already have.

So build the check in. Before you substitute a number, say what it measures. Before you pick a word, find the clause that sets the direction. You can learn both in an afternoon, and they work on every test after this one. Guessing at your routing never will.