Practice question banks do not look like the SAT, and the gap is measurable. We mapped all 3,444 questions in our bank to College Board's 29 official skill tags, rolled them into the four content domains per section, and compared the result against the published blueprint. Our bank over-weights Problem-Solving and Data Analysis by more than seven percentage points and under-weights Advanced Math by six and a half. Those are our numbers, and we are publishing them because the drift is the point.
What the real test is made of
College Board's Digital SAT Suite Specifications Overview fixes the proportions. Every form is assembled to them, which is why a scaled score means the same thing across administrations.
| Section | Domain | Blueprint share |
|---|---|---|
| Reading and Writing | Craft and Structure | ~28% |
| Information and Ideas | ~26% | |
| Standard English Conventions | ~26% | |
| Expression of Ideas | ~20% | |
| Math | Algebra | ~35% |
| Advanced Math | ~35% | |
| Problem-Solving and Data Analysis | ~15% | |
| Geometry and Trigonometry | ~15% |
What our bank is made of
Counted on 2 September 2026. The 1,688 Reading and Writing questions and 1,756 Math questions were each tagged to one of the 29 official skills, then rolled up by the same skill-to-domain map the College Board taxonomy uses, so this is an exact lookup rather than keyword guessing.
| Reading and Writing domain | Questions | Our share | Blueprint | Drift |
|---|---|---|---|---|
| Information and Ideas | 507 | 30.0% | 26% | +4.0 |
| Craft and Structure | 437 | 25.9% | 28% | −2.1 |
| Standard English Conventions | 379 | 22.5% | 26% | −3.5 |
| Expression of Ideas | 365 | 21.6% | 20% | +1.6 |
| Math domain | Questions | Our share | Blueprint | Drift |
|---|---|---|---|---|
| Algebra | 582 | 33.1% | 35% | −1.9 |
| Advanced Math | 501 | 28.5% | 35% | −6.5 |
| Problem-Solving and Data Analysis | 389 | 22.2% | 15% | +7.2 |
| Geometry and Trigonometry | 284 | 16.2% | 15% | +1.2 |
Why banks drift the same way
Look at which domains are over-represented. Problem-Solving and Data Analysis is up seven points; Standard English Conventions is down three and a half. That pattern is not random, and we would expect to find it in most third-party banks.
Data-analysis questions are straightforward to write. You need a table or a chart and a question about it, and the item is self-contained. Advanced Math questions, the nonlinear functions and equivalent expressions that make up a third of the real Math section, take much longer to author and are far easier to get subtly wrong. Standard English Conventions has the same problem in reverse: writing a punctuation item that tests exactly one rule, with three genuinely wrong alternatives, is slow work.
So banks fill up with what is cheap to produce. If you practise on an unweighted bank, you spend disproportionate time on chart questions worth six to eight marks on test day, and too little on the two domains that together carry about 70% of Math.
The drift above is in the raw bank. Our mock forms are assembled to the blueprint targets rather than sampled at random, so a full-length test carries 14 to 17 Algebra questions and 6 to 8 Geometry ones regardless of what the underlying pool looks like. An unweighted bank is fine for drilling one skill and misleading as a simulation.
What we can and cannot tell you about difficulty
Here is where we have to be careful, because this is the claim most prep companies overstate.
| Difficulty label | Questions | Share |
|---|---|---|
| Easy | 1,240 | 36.0% |
| Medium | 1,136 | 33.0% |
| Hard | 1,068 | 31.0% |
Those are editorial labels applied when each question was written. They are not calibrated difficulty parameters derived from how students actually performed. Our adaptive engine converts them to numeric values for routing, but that conversion is a documented stand-in for real calibration, not a measurement, and we would rather say so than present a three-way label dressed up as psychometrics.
Real item calibration requires large volumes of student responses per question. Anyone publishing precise difficulty values for SAT questions without that response data is showing you a relabelled guess.
The one number that matches perfectly
Of the 1,756 Math questions, 421 are student-produced responses, the grid-ins with no answer choices. That is 24.0% of Math, against a published share of roughly 25%. On the real test it works out at about 11 of 44.
Grid-ins are worth attention out of proportion to their count, because they remove the one strategy that rescues you elsewhere. There is nothing to eliminate and nothing to work backwards from, so a shaky method that survives on multiple choice fails completely here. We covered the mechanics in SAT grid-in questions.
What to take from this
- Judge a question bank by its weighting, not its size. Ten thousand questions weighted wrong is worse practice than two thousand weighted right.
- Practise Math in the real ratio. Algebra and Advanced Math are about 70% of the section. If your revision does not look like that, it is not preparing you for the section you will sit.
- Treat difficulty labels as editorial. Including ours.
Frequently asked questions
How many questions are on the digital SAT by content domain?
Reading and Writing has 54 questions: roughly 28% Craft and Structure, 26% Information and Ideas, 26% Standard English Conventions and 20% Expression of Ideas. Math has 44: about 35% Algebra, 35% Advanced Math, 15% Problem-Solving and Data Analysis and 15% Geometry and Trigonometry. The proportions come from College Board's published specifications overview.
Do practice question banks match the real SAT?
Usually not. In our own bank of 3,444 tagged questions, Problem-Solving and Data Analysis is over-represented by 7.2 percentage points and Advanced Math under-represented by 6.5. The pattern reflects how expensive each question type is to write: chart questions are cheap to produce, nonlinear-function and punctuation items are not.
What percentage of SAT Math questions are grid-ins?
About 25% on the real test, roughly 11 of the 44 Math questions. In our tagged bank, 421 of 1,756 Math questions are student-produced responses, which is 24.0%, closely matching the published share.
Can anyone publish real SAT question difficulty values?
Not without large volumes of student response data per item. Calibrated difficulty is estimated from how students actually perform, so a bank that publishes precise difficulty numbers without that response volume is presenting editorial labels as measurements. Our own three-way Easy, Medium and Hard labels are editorial, and the numeric values our engine uses are a documented stand-in for real calibration.
Which SAT Math domain should I prioritise?
Algebra and Advanced Math, which together account for about 70% of the Math section, or roughly 28 to 34 of the 44 questions. Geometry and Trigonometry, where many students concentrate their revision, is only 6 to 8 questions.
How many questions does a good SAT question bank need?
Weighting matters more than volume. A bank needs enough items per domain to build full forms without repetition, but a large bank weighted wrongly gives worse practice than a smaller one assembled to the blueprint. Ask how forms are assembled, not how many questions exist.