How to read a GMAT question bank description without being…

A GMAT question bank is the single largest line item in most self-study budgets after tuition for a course. Candidates buy a book, a platform subscription, or an AI-generated deck expecting that volume will translate into score movement, then discover after a hundred hours that they have practised a question type the new exam no longer asks, or a difficulty curve that does not resemble the adaptive modules. This piece is a working audit. It walks through what to check on a product page, what to verify in the first thirty minutes of use, and which signals tell you the bank will quietly waste weeks of your preparation cycle. The aim is a checklist a candidate can apply before any purchase, plus a defence plan if the bank you already own turns out to be the wrong one.

1. Confirm the bank is built for the GMAT Focus edition, not the legacy GMAT

The first filter is non-negotiable. The current exam, commonly called the GMAT Focus, runs three sections — Quant, Verbal, and Data Insights — for a total of 64 scored items in about 2 hours 15 minutes. Banks published before this edition was introduced still surface on second-hand marketplaces and in older study threads, and their item shapes do not match what the adaptive engine will throw at you in October or November. A bank whose marketing copy says "GMAT" without naming Focus is, by default, suspect. The verbal section in particular dropped Sentence Correction, and the data interpretation layer moved almost entirely into a new section. A few minutes on the product page or in the publisher's FAQ will tell you whether Quant has 21 questions, whether Verbal has 23, and whether Data Insights is treated as a section in its own right.

Beyond item counts, look at the format inside each item. A Focus Quant bank should give you multiple-choice items with five options and a numeric entry. Verbal in the Focus edition is reading comprehension and critical reasoning only, with no sentence correction. Data Insights has multi-source reasoning, table analysis, graphics interpretation, two-part analysis, and data sufficiency — a set of five item families that share a section timer. If a bank's mock tests contain sentence correction, it was authored for the prior exam. The same logic applies if a "data insights" question type is missing from the bank's index. You are not buying extra practice; you are buying misaligned practice that can drill in a method the test will not reward.

Three quick checks on the publisher's own site

Search the bank's product page for the strings "GMAT Focus", "Focus Edition", and "Data Insights". If only "GMAT" appears, the bank is probably legacy content.
Open the bank's index or table of contents. Look for Sentence Correction. If it has its own chapter, the bank was written for the old Verbal section.
Check the publication date on the copyright page. Anything authored before the Focus launch, even if updated, is usually a re-skin of older material rather than a rebuild.

For most candidates, the right call is to retire any bank authored for the legacy GMAT the moment they switch to Focus preparation. Re-using legacy material for a different exam format is a quiet score-killer, because the time pressure in the Focus Data Insights section is materially different from anything the previous exam had. A bank that cannot describe how its 20 Data Insights items map to a 45-minute section is the wrong bank.

2. Audit the item-format coverage: 64 questions, five Data Insights families, real source data

Item-format coverage is the second filter. A bank might be Focus-branded and still underserve a specific question type that the engine uses as a tie-breaker between adjacent score levels. The Quant section in the Focus edition has 21 questions across problem solving and data sufficiency, both with a multi-choice framing. Verbal has 23 reading comprehension and critical reasoning items, with no sentence correction, no logic games, and no analogies. Data Insights has 20 items spread across five families: multi-source reasoning, table analysis, graphics interpretation, two-part analysis, and data sufficiency. A bank that gives you 1,000 quant items but only 60 data sufficiency items in total is not balanced for the section that carries the most weight per item.

The way to test this is to count. A serious bank should expose you to at least 300 items per major section, with the Data Insights families each getting a meaningful share. Two-part analysis in particular is a structurally unusual question type — two correct answers that must both be selected for credit, set inside a shared stem — and the practice effect on this item family is large. If a bank's "DI practice" is mostly graphics interpretation with a thin tail of the other four families, you will underperform on the section the test weights most heavily. In my experience, the most underprepared candidates I see in diagnostic sessions have done hundreds of graphics interpretation items and perhaps thirty two-part analysis items. The exposure is lopsided, and so is the score.

Format-fit signals worth a second look

Look for item stems with two correct answer choices selected simultaneously. If the bank has none, the two-part analysis family is absent.
Look for items that present two tabulated sources side by side. That is multi-source reasoning, and a well-built bank should have at least a few dozen of them.
Look for graphics interpretation items whose data is presented in real units with axis labels and time-series ticks. Banks that draw decorative charts are padding the count.

A second check is whether the bank provides the same on-screen calculator you will see on test day. The Focus exam gives candidates a built-in calculator on every Data Insights item. If the bank's DI questions do not include a calculator widget, you are practising in a different environment than the one you will be tested in. This matters more than it sounds: a candidate who has done 400 DI items in their head will often type into the on-screen calculator clumsily and lose 20 to 30 seconds per item, which is precisely the kind of pacing error that the exam's adaptive scoring punishes.

3. Difficulty calibration: a bank that is too easy is a worse trap than one that is too hard

Many candidates who contact me with a stuck score have, paradoxically, practised too many easy items. The GMAT Focus scoring is adaptive. As you answer correctly, the engine raises difficulty; as you answer incorrectly, it lowers it. A bank that is calibrated too low — items in the 400 to 550 difficulty range when the candidate is already scoring in the 640 to 700 band — trains the eye to find the easy answer fast. The candidate builds reflexes that are tuned to a difficulty band the engine will visit briefly at the start of the section and never again.

The right way to read a bank's difficulty claim is to take ten items from each section at random and time yourself. If you are finishing Quant items in well under two minutes each, the bank is too easy. A calibrated Focus Quant bank should make you think. Items in the upper half of a well-built bank take 2:30 to 4:00 minutes for most candidates scoring in the 650+ band, and not every item will be solvable in the time given. The test design intentionally includes some items that the average candidate will not finish, and a bank that pretends otherwise is misrepresenting the exam. For Verbal, a similar logic holds. Reading comprehension passages in a well-calibrated bank are dense, often science-adjacent, and the questions probe inference rather than retrieval.

How to test calibration in the first 30 minutes

Pull ten items at the bank's "hard" tier. Solve them timed.
Compute the per-item average. If it sits below 1:30 for Quant or below 1:10 for Verbal, the bank is miscalibrated down.
Count how many you got correct. If you are clearing 9 or 10 of 10, the bank is also too easy on content. Real Focus-level items should give a 650-band candidate 6 to 8 correct out of 10 on a hard set.

For most candidates, the practical filter is the publisher's own mock tests. A bank that ships with three or more adaptive mock tests is signalling that it understands how the engine scores. A bank that ships with none is asking you to trust the difficulty labelling on each item, which is a much weaker guarantee. Where the publisher does ship mocks, sit the first one within the first week of using the bank. If the score reads wildly above or below your target, the bank's difficulty curve is offset from the real exam, and the bank will train the wrong habits.

4. Source authenticity: practice items should be retired-test items, not textbook rewordings

The single largest differentiator between high-quality and low-quality GMAT banks is source. The official materials from GMAC — the test maker — consist of a small set of official practice exams, an official question pack, and the materials bundled with prep courses the council authorises. Retired test items are gold because they were written to the exact rubric the live engine uses. When a candidate practices a retired item, they are reading the test author's voice, which trains recognition of stem phrasing, distractor patterns, and difficulty gradients that AI-generated or textbook reworded items cannot reproduce.

Many third-party banks mix retired items with new items written to look like retired items. The new items are usually easier to spot than publishers admit: the distractors are too clean, the answer choices are too symmetric, and the difficulty curve is monotonic in a way retired items rarely are. A useful diagnostic is to solve a passage or item set, then look at the official explanations for that family of questions. If the bank's explanations mention test-maker conventions — such as how a critical reasoning stimulus is usually structured with a conclusion and two supporting premises — the bank is probably working from retired material. If the explanations read like a generic math tutor's notes, the items are probably reworded from textbooks.

AI-generated question banks, increasingly common, introduce a separate concern. Generative models can produce unlimited items in the right shape, but they have a tendency to invent conventions the real test does not use. A common failure mode is the AI item whose correct answer hinges on a reading of the passage that no careful reader would share. When a candidate practices such items, they begin to distrust the test, second-guess strong inferences, and slow down on items the engine considers easy. The fix is to keep AI-generated items in a clearly labelled subset, and never let them form more than a small fraction of total practice volume. For most candidates I'd personally pick one well-curated bank of retired items over an unlimited AI generator, because the feedback signal on the former is much tighter.

5. Exposure discipline: caps, sequencing, and the "do not repeat" rule

A practical question for any bank is whether the publisher supports exposure discipline. The GMAT has roughly 2,000 retired items in circulation across the major prep publishers. A serious candidate will eventually see most of them. After that, repetition is a risk. Re-solving an item you have already solved gives you a false confidence boost, because your eye remembers the answer rather than your method reconstructing it. The score on a re-solve is almost always higher, and the candidate incorrectly concludes that the bank is working. The truth is that the practice effect on re-seen items is small for the first re-solve and near-zero by the third.

Good banks support sequencing — they let you mark items as solved, filter unsolved items, and run timed sets that exclude the bank you have already cleared. Bad banks force you to re-encounter solved items unless you keep a parallel spreadsheet. The parallel-spreadsheet workaround is real and is what most tutors I know actually do, but it adds 30 to 60 minutes of overhead per week, which is time better spent on weak-topic drilling. When you read a product page, look for terms like "filter", "unsolved only", "spaced repetition", and "adaptive sequencing". If those terms are absent, plan to maintain your own progress tracker from day one.

Exposure rules that hold up across candidates

Cap total solved items at roughly 1,200 to 1,500 across your full prep cycle. Beyond that, the marginal item is a re-skin of something you have already seen.
Hold 200 to 300 items untouched as a final-week calibration set, and never look at them until ten days before the exam.
Mark every item you solve with a date, a section, a sub-topic, and a result. This is the spine of an error log, and the bank should not prevent you from maintaining it.

A second exposure question is how the bank handles item explanations. The strongest banks publish explanations for every option, not just the correct one, because the wrong options in GMAT items are written as designed distractors. A candidate who reads why each distractor is wrong learns the engine's logic. A candidate who reads only the correct answer learns nothing about the four traps they have just dodged. If a bank publishes the correct answer only, write the distractors down yourself and articulate why each is wrong. It is a 20-minute habit that compounds across hundreds of items.

6. Mock-test integrity: the bank should come with at least three full adaptive mocks

Mock tests are the calibration tool, and a bank that does not provide adaptive mocks is asking you to take a self-reported progress number at face value. The GMAT Focus adaptive engine changes the difficulty of item N+1 based on your response to item N, and a non-adaptive practice set cannot reproduce the pressure of the live exam. Look for a bank that provides at least three full adaptive mocks with the same time budgets you will see on test day: roughly 45 minutes for Quant, 45 minutes for Verbal, and 45 minutes for Data Insights, with a 10-minute break in the middle. The mocks should also enforce the on-screen calculator, the flag-for-review behaviour, and the section-end screen lock, because each of these is a small piece of the test-day experience that a candidate practising in a different environment has to relearn.

Schedule the mocks across your prep cycle. A common mistake is to sit four mocks in the last ten days. That sequence tells you almost nothing — fatigue, recency, and the absence of a feedback loop between attempts distort the score trajectory. A better sequence is mock one at the end of week three, mock two at the end of week six, mock three at the end of week nine, and a final mock in the last three days before the exam. This gives you four data points spaced at intervals long enough for the score to move on training, not on noise.

Mock-to-bank coherence: a quick rubric

Bank feature	High-integrity signal	Warning sign
Adaptive engine	Item difficulty changes by section and prior responses	Items are pre-sorted into fixed sets
Section timers	Exact 45-minute per-section clock with break option	Untimed or globally timed practice
Calculator	On-screen calculator present on every DI item	Calculator absent or external only
Review behaviour	Flag-for-review and section-end summary mirror the live UI	Free navigation between items at all times
Score reporting	Section scores, percentile, and time-per-item breakdown	Single composite score only

For most candidates the right call is to use the same bank's mocks for the entire cycle rather than mixing mocks from different publishers. Cross-bank mock scores can differ by 30 to 60 points because of calibration drift, and the resulting score swings will distort your read on whether your training is working. If you must mix, treat each bank's mocks as a separate score family and compare within families, not across them.

7. Price, licensing, and the "true cost per solved item" calculation

The headline price of a question bank is almost never the figure that matters. The figure that matters is the per-item cost across the items you will actually solve, and the licensing window in which you can solve them. A bank priced at 100 euros for 1,500 items, with a 6-month licence, comes out to roughly 7 cents per item. A bank priced at 80 euros for 600 items, with a 3-month licence, comes out to 13 cents per item and pressures you to finish in a tighter window. A candidate paying the lower absolute price for the more expensive per-item bank will often end up buying a second licence to finish what they planned, doubling the actual cost.

Beyond price, look for licensing terms on device count, account sharing, and offline access. Candidates who travel for work often need an offline mode; candidates who study at a shared desk need multi-device support; candidates who work with a tutor need account-sharing terms that the publisher explicitly permits. If a publisher's terms of service forbid account sharing and you study with a tutor, you have effectively doubled the price of the tutoring engagement because the tutor cannot see the bank interface. Read the terms before you pay, not after.

Cost-per-item worked example

Suppose a candidate has a 14-week prep window and plans to solve roughly 1,200 items at a steady cadence of 12 to 14 items per weekday. Bank A offers 1,800 items for 120 euros, valid for 9 months, with two-device support and three mocks. Bank B offers 700 items for 60 euros, valid for 4 months, with one-device support and one mock. The candidate will solve 1,200 items from Bank A, paying 10 cents per item. From Bank B, the candidate will exhaust the licence (700 items) and need to renew, paying roughly 17 cents per item across the full 1,200 solves. Bank A wins on the per-item calculation, on the licence length, and on the mock count, despite the higher sticker price. This is the kind of arithmetic the product page rarely encourages you to do, and it is the arithmetic that determines whether a bank is a good buy for your specific window.

8. Common pitfalls and how to avoid them

Five patterns repeat across the candidates I see in diagnostic sessions. Each is a question-bank error rather than a study error, and each is fixable with a one-line tactical change.

Pitfall 1: solving a bank's items in the default order, top to bottom

Default ordering is calibrated to the publisher's syllabus, not to your weak-topic map. Always run items in topic clusters driven by your error log, not by the bank's chapter order. If a candidate's quant error log shows that data sufficiency accounts for 40 percent of missed items, they should solve all 80 DI-style data sufficiency items in a single two-week block, not sprinkle them across the bank's default order.

Pitfall 2: ignoring the time stamp on a wrong answer

A bank that records the time you spent on each item is doing you a favour. A candidate who gets an item wrong in 45 seconds and a candidate who gets the same item wrong in 4:30 minutes have different problems — the first is rushing, the second is misreading. The first needs a pacing intervention; the second needs a comprehension intervention. Banks that do not record per-item time force you to keep a parallel log, which is doable but rarely sustained.

Pitfall 3: re-solving flagged items too soon

Re-solving a flagged item within 48 hours tells you almost nothing because short-term memory still recognises the answer. A practical rule is to re-solve flagged items no earlier than 10 days after the first attempt, ideally in a mixed-topic set rather than a topic-cluster set. Spaced re-solve is the single highest-leverage habit a question bank can support, and most candidates under-use it.

Pitfall 4: trusting bank-reported difficulty labels

Publisher difficulty labels are usually band labels, not item-level calibrations. A bank that labels every item as 600-700 difficulty is using a different convention from a bank that labels items on a 200-800 scale, and the two are not directly comparable. The only difficulty score you can trust is the one your mock tests return. If your mock scores sit in the 640 band but the bank's "hard" items feel like a 550 mock, the bank's labels are wrong for your level, and you should treat the bank's hard tier as a warm-up rather than a challenge.

Pitfall 5: using the bank for review, not for first attempts

Many candidates buy a bank, do a few dozen items, then read through the explanations of items they did not attempt. This is study-porn; it does not move the score. The bank is for first attempts, timed, with the answer recorded before the explanation is read. Review belongs in the error log, where you re-derive the correct path from your wrong path. If a candidate has spent more time reading explanations than solving items, the bank is functioning as a textbook, and a textbook is the wrong tool for the final 60 days of a prep cycle.

9. A short decision rubric for the actual purchase

When the price page is open and the buy button is highlighted, the candidate has a ten-minute window to apply the audit. The right call, in most cases, is to use a small checklist rather than a long essay. A workable rubric is the following nine-point check, with one point per satisfied criterion: bank is Focus-branded, item counts are balanced across the five DI families, on-screen calculator is present, three or more adaptive mocks are bundled, item explanations cover all five options, sequencing and unsolved-only filtering are available, licence covers the full prep window, per-item cost lands under 12 cents, and the publisher is named with a verifiable track record. Banks scoring 7 of 9 or higher are good buys; 5 or 6 means a longer look; 4 or below is a refund candidate if the trial is open, or a no-buy if it is not.

For most candidates, the practical move is to start a single bank at the start of the prep cycle, hold it for 12 weeks, and resist the temptation to add a second bank in week six. The error pattern of the typical underperformer is bank-hopping — solving 200 items in bank A, switching to bank B for a different voice, switching again to bank C, and ending the cycle with no bank's items mastered. A single bank used end-to-end almost always beats two banks used in alternation. The exceptions are candidates whose first bank turns out to be miscalibrated, and the audit above is designed to catch those exceptions before the cycle starts rather than at week eight.

The next concrete step for any candidate weighing a bank purchase is to run a 30-item diagnostic on the bank's free trial or sample set. Time the items, count the correct answers, and record the per-item time. The result will tell you, in roughly 90 minutes of work, whether the bank is miscalibrated for your level. If the diagnostic shows a 50 percent accuracy rate at a comfortable 2:30 per item, the bank is well-matched. If the diagnostic shows an 80 percent accuracy rate at 1:00 per item, the bank is too easy. If the diagnostic shows a 30 percent accuracy rate at 4:00 per item, the bank is too hard. Each diagnosis points to a different tactical move, and each is a more efficient use of 90 minutes than any amount of product-page reading.

TestPrep Europe's diagnostic assessment is a natural starting point for candidates building a sharper preparation plan, and the question-bank audit above is the next layer once the diagnostic is in hand.

Frequently asked questions

How many questions should a complete GMAT Focus question bank contain?

A serious bank should expose you to at least 1,500 items across Quant, Verbal, and Data Insights, with a balanced share of roughly 300 Data Insights items spread across the five families. Banks below 1,000 items are usually thin in two-part analysis and multi-source reasoning, which are the families that drive the most score movement in the upper score bands.

Are AI-generated GMAT question banks good enough for serious preparation?

AI generators are useful as a low-stakes filler for warm-up sets, but they should not form more than a small share of total practice volume. The most reliable score movement still comes from retired test items written to the live rubric, because their distractors, stem phrasing, and difficulty gradients are calibrated by the test maker. Use AI items in clearly labelled subsets and keep retired items in the main rotation.

How do I know whether a question bank is calibrated to the Focus edition?

Check the product page for the strings 'GMAT Focus' and 'Data Insights', scan the index for Sentence Correction, and confirm the on-screen calculator is present on every Data Insights item. A bank that omits these markers, or that publishes items in legacy question types, is a re-skin of older material and will train methods the live engine does not reward.

Is it better to use one bank end-to-end or to mix two or three publishers?

One bank used across the full cycle is almost always the right answer. Bank-hopping splits your exposure across difficulty curves, which makes it hard to read your score trajectory across mocks. If you must mix, treat each publisher's items as a separate score family and use one publisher's mocks throughout, so that your calibration signal is consistent.

How should I spend the 200 to 300 items I hold back for the final week?

Treat them as a calibration set, not a learning set. Sit them under timed conditions, scored, with no review until the end of the set. The point of the final-week block is to confirm that your pacing and accuracy have stabilised at the target score, not to teach you new content. If the calibration set lands inside your target band, the bank has done its job; if it lands outside, that is information you want 10 days before the exam, not 30.

How to read a GMAT question bank description without being sold to