A usable question bank needs consistent tags, stable question IDs, checked answers and a clear approval process. Start with a small tagging scheme and a reviewed collection. A larger bank is not automatically a better bank if teachers cannot find a suitable, correct question when preparing a test.
Define the minimum record
Keep question ID, subject, topic, learning objective, format, source reference, reviewer, version and status. Store the answer key and explanation with the same version as the question. Label difficulty as an estimate until you have appropriate response evidence. A spreadsheet can hold your review log if your platform does not support a particular field.
Download the question-bank review log. This is an editable, ungated worksheet from ThePrepLab. Examples are illustrative, not customer results. Use anonymous IDs instead of student personal information.
| Field | Example | Rule |
|---|---|---|
| Subject | Mathematics | One agreed spelling |
| Topic | Linear equations | Avoid near-identical duplicate tags |
| Objective | Solve a two-step equation | Describe observable work |
| Format | Single-answer MCQ | Match the stored marking configuration |
| Status | Draft / Review / Approved / Retired | Only approved items enter the release set |
| Source | Original faculty item, ALG-014 | Record permission and provenance separately |
Detect duplicates without deleting useful variants
Exact duplicates have the same wording and options. Near duplicates may change only numbers, while genuine variants may test different reasoning. Compare the stem, intended objective, answer and source before deciding. Do not delete a question used in a past assessment until you understand how that platform preserves historical attempts. Retirement or a separate archive may be safer than removal.
An example of version control
Question ALG-014 originally says “Solve 3x + 2 = 14” with key x = 4. A teacher later changes the constant to 5 but forgets the key; the correct answer is now x = 3. Treat this as a new revision requiring a fresh check. Record the old and new values, reviewer and affected tests. Do not silently change a released test while students are attempting it.
Use a repeatable review queue
- Draft: confirm a permitted source or record the generation instruction.
- Academic review: solve the question, check ambiguity and inspect every option.
- Presentation review: check equations, images and student-view formatting.
- Release: save the approved revision and test membership.
- Maintenance: investigate learner reports and retire unsuitable items with a recorded reason.
The review owner should be identifiable to your team, but the public website does not need their personal details. When two reviewers disagree, hold the question rather than treating an AI answer as the deciding authority. The AI question review guide provides a separate checklist for generated drafts.
How should you measure bank quality?
Useful measures include the proportion of release-ready questions with current review, unresolved error reports, retrieval time for a specified blueprint and coverage gaps by objective. Define the denominator: approved items divided by all stored items means something different from approved items divided by this month’s draft intake. These are suggested operating measures, not published PrepLab benchmarks.
Can software do the whole process automatically?
Software can assist selection and organisation, but this guide does not claim automatic duplicate detection, immutable version history or approval gates in every PrepLab package. Demonstrate the required workflow before purchase. The question-bank solution and bank-to-test tutorial show the preparation use case. Use a test blueprint to choose items and item analysis to review later evidence.