How to build a randomized quiz question bank that is fair to every learner

A question bank is a reusable set of questions stored separately from any single quiz. A quiz then says "draw 5 from this bank" instead of listing five fixed questions. Each time a learner starts an attempt, a fresh set is drawn.

Three problems go away. Retakes stop being memory tests, because the second attempt asks different questions. Answer sharing loses value, because the screenshot a friend sends covers a different set. And maintenance gets easier, because a question you fix in the bank is fixed in every quiz that draws from it.

Start with the ratio of pool size to draw size. If a bank has 10 questions and the quiz draws 5, two attempts will overlap a lot. If the bank has 15 and the quiz draws 5, a learner can take it three times and still see mostly new questions. A pool of three times the draw is a sensible floor; five times is comfortable for high-stakes quizzes with several retakes.

The number of possible quizzes grows quickly. Drawing 5 from a pool of 15 allows 3,003 different combinations. That sounds like plenty, but combinations are not the point. What matters is how many questions a learner sees in total across their attempts, and whether they could pass by memorizing them. A big pool of near-duplicate questions does not help.

Randomness is fair only when the questions it chooses between are equivalent. If one bank mixes easy definition questions with hard applied problems, one learner might draw five easy ones and another five hard ones. They would get different scores for the same knowledge.

The fix is to use several small banks instead of one big one. Make a bank per topic and per difficulty, then build the quiz as a recipe: 2 from "Topic A, recall", 2 from "Topic A, applied", 1 from "Topic B, applied". Every learner gets the same shape of test, just different questions within each slot.

Within a bank, write each question to test the same skill at the same level. A good self-check: could you swap any two questions in the bank and be equally happy with the result? If not, they belong in different banks.

Multiple choice and multi-select questions are the easiest to write in parallel versions, because you can keep the structure and change the scenario. True/false questions are quick to write but easy to guess, so use them sparingly. Fill in the blank works well for terms and figures. Ordering and matching are good for processes and vocabulary and are hard to guess by chance.

Mixing types inside one bank is fine as long as difficulty stays level. A matching question with eight pairs is harder than a true/false question on the same idea, so do not put them in the same slot.

1. List the three to six skills the quiz must check. 2. For each skill, decide how many questions the quiz should ask. 3. Write three times that number per skill, varying the scenario, not the difficulty. 4. Write a one-line explanation for each correct answer, which learners see after submitting. 5. Put each skill in its own bank. 6. Build the quiz as a recipe of draws per bank, or, if your tool draws from one bank per quiz, give each skill its own short quiz. 7. Take it yourself three times and read every question you get.

Step 7 catches most problems. You will notice a question that gives away the answer to another, an option that is obviously wrong, or a scenario that only makes sense with context from the lesson. Fix those before learners see them.

If you already have questions in another system, import them instead of retyping. Many tools export GIFT, QTI or CSV, which most course platforms can read.

Randomizing is not always right. A short practice quiz at the end of a lesson is often better fixed, so learners can retake the same questions until they get them right. That is how practice works. Randomizing it can feel like moving the goalposts.

Randomizing also makes it harder to compare learners question by question. If you want to see that a large part of a cohort misses one specific idea, a fixed final quiz gives cleaner data. A good compromise is randomized banks for graded and certificate quizzes, and fixed quizzes for practice.

HiveLearn quizzes can draw questions from question banks, with a fresh draw for each attempt. Banks hold any of the six question types: multiple choice, multi-select, true/false, fill in the blank, ordering and matching. Each question can carry points and an explanation.

A bank-draw quiz pulls a set number of questions from one bank. That makes the fairness rule above concrete: keep each bank to one skill at one level of difficulty, and if a course needs to test several skills, give each its own bank and its own short quiz. A learner can only have one attempt in progress per quiz, so starting and abandoning attempts to fish for an easier draw does not work.

Grading happens on the server, so the drawn questions arrive in the learner's browser without the answer key, and explanations come back only after they submit. That matters more with banks than without them: a randomized quiz with a readable answer key is still a leaked quiz.

Banks can be imported and exported as JSON, CSV, GIFT or QTI, so you can bring questions from Moodle, Canvas or a spreadsheet. Quizzes sit inside the same course builder as your lessons, next to community, events and payments. See the pricing page for which plan includes them.