Scormbee

How to Write Good Quiz Questions for E-learning

9 min read Czytaj po polsku

A quiz is the only part of an e-learning course that produces evidence. Completion says someone reached the end; the quiz says whether they can do what the course was for. Most weak quizzes fail for the same reasons: questions written from the text rather than the objectives, wrong answers nobody would choose, feedback that says only "Incorrect", and a pass mark picked without arithmetic. This guide goes through each of those, then shows how the result reaches your LMS.

Start from the objective, not the text

Every question should trace back to an objective, and the objective's verb tells you what kind of question to write.

The objective says the learner can… Write a question that… Example stem
Recall a fact or term Asks for it directly, ideally in context "Which number do you call to report a lost badge?"
Recognize an example Shows a case and asks what it is "Which of these emails is a phishing attempt?"
Apply a rule Describes a situation and asks what to do "A customer asks what data you hold on them. What do you do?"
Analyze or judge Gives a messy case with competing factors "Two incidents arrive at once. Which do you escalate first, and why?"

Most compliance and onboarding courses need the "apply" row. Most quizzes are written in the "recall" row, because recall questions are the easiest to lift from the slides.

Before: "According to the policy, within how many hours must data breaches be reported to the supervisory authority?"

After: "You realize you emailed a client list to the wrong address an hour ago. What do you do now?"

The first question tests a number the employee will never use. The second tests the action the course exists to teach: report it internally, immediately.

Anatomy of a good multiple-choice question

A question has four parts: the stem, the correct answer (the key), the wrong answers (distractors) and the feedback.

  • The stem is complete. A learner should be able to answer it before reading the options.
  • One clearly best answer. If experts argue about it, rewrite it.
  • Options of similar length and grammar. A key that's always the longest and most careful option gets spotted.
  • No "all of the above" or "none of the above". They reward partial knowledge and test-taking tricks.
  • Positive stems. If you must use NOT, put it in bold.
  • No trivia. Exact dates and article numbers rarely matter on the job.
  • Shuffled options, unless they have a natural order, such as numbers.

Distractors are where quality comes from

A distractor is good when a learner who hasn't mastered the objective would pick it. Find them in real mistakes:

  • Help-desk tickets and incident reports: what people actually did wrong.
  • Subject-matter experts: "What do new people always get wrong?"
  • Partial truths: the right action at the wrong time, or the right idea applied too broadly.
  • The right answer to a different situation: correct for a customer, wrong for a supplier.

Before:

What should you do with a suspicious email?

A. Report it with the Report button. (correct) B. Print it out. C. Reply with your password. D. Ignore all emails from now on.

Nobody would choose B, C or D. The question tests reading, not judgment.

After:

An email from "IT Support" asks you to confirm your password via a link. The sender address ends in @it-support-desk.net, and your company uses @example.com. What do you do?

A. Reply and ask whether the message is genuine. B. Open the link to check whether the page looks real. C. Report it with the Report button, then delete it. (correct) D. Forward it to your team as a warning.

Each distractor is something people really do, and each has a specific reason it's wrong.

How many options? Fewer than you think. A well-known 2005 meta-analysis by Michael Rodriguez, covering 80 years of research on multiple-choice items, concluded that three options are optimal in most cases. A fourth option that nobody picks adds reading time and nothing else. Use three good options rather than four with a filler.

Feedback that teaches

Feedback is the cheapest teaching moment in a course. Write it for each option, not just "Correct" and "Incorrect":

  • A: "Replying confirms your address is active and reaches the attacker, not IT."
  • B: "Opening the link is the attack. Fake login pages often look identical to the real one."
  • C: "Right. Reporting lets the security team block the sender for everyone."
  • D: "Forwarding spreads the link further. Report it instead, and the team warns everyone safely."

In practice questions, show feedback immediately. In a graded final test, many authors show results only after submission, so the answers can't be passed along question by question.

Question types and when to use them

Type Measures well Watch out for SCORM interaction type
Single choice Decisions, recognizing the right action Weak distractors choice
Multiple response Knowing every item in a set Decide on all-or-nothing vs partial credit, and say how many to pick choice
True/false A specific misconception 50% chance of guessing; ambiguous statements true-false
Matching Relationships: term to definition, symptom to action With equal lists, the last pair is free; add one extra option matching
Ordering Procedures and sequences One swapped step marks the whole answer wrong sequencing
Fill in the blank Exact terms or values Spelling variants, case, diacritics fill-in
Numeric Calculations Rounding; accept a range numeric
Branching scenario Judgment over several steps Effort to write; keep endings meaningful Varies by tool

Guessing: the arithmetic behind question types

Question type changes how easy the test is to pass by luck. For a 10-question test answered by pure guessing, the chance of reaching the pass mark is:

Format Pass mark 70% Pass mark 80%
True/false (2 options) 17% 5.5%
3 options 2% 0.3%
4 options 0.4% 0.04%

These follow from the binomial distribution: for example, at least 7 of 10 true/false answers right by chance is 176 of 1,024 equally likely outcomes. The lesson is that true/false items shouldn't dominate a graded test, and that three-option questions are already very hard to pass by guessing.

Choosing a pass mark

The pass mark has to make sense with the number of questions, because each question is a large step on a short test.

Questions Mistakes allowed at 70% at 80% at 90%
5 1 1 0
8 2 1 0
10 3 2 1
12 3 2 1
20 6 4 2

With five questions, 80% and 70% are the same test, and 90% means perfect. If the difference matters, add questions. Then:

  • Use one value everywhere. The course's pass mark and the manifest's (adlcp:masteryscore in SCORM 1.2, imsss:minNormalizedMeasure in SCORM 2004) must match, or learners see "Passed" while the report says "Failed".
  • Decide on retakes. Allowing retakes with feedback turns the quiz into practice; allow them, but draw from a larger pool or shuffle so learners don't just memorize letters.

How scores and answers reach the LMS

The course reports the result through the SCORM API. In SCORM 2004, a graded quiz typically sends one interaction per question and then the score:

api.SetValue("cmi.interactions.0.id", "q3-sender");
api.SetValue("cmi.interactions.0.type", "choice");
api.SetValue("cmi.interactions.0.description", "Which sender address is suspicious?");
api.SetValue("cmi.interactions.0.correct_responses.0.pattern", "c");
api.SetValue("cmi.interactions.0.learner_response", "b");
api.SetValue("cmi.interactions.0.result", "incorrect");
// …after the last question
api.SetValue("cmi.score.raw", "8");
api.SetValue("cmi.score.min", "0");
api.SetValue("cmi.score.max", "10");
api.SetValue("cmi.score.scaled", "0.8");
api.SetValue("cmi.success_status", "passed");
api.Commit("");

Three details matter:

  • Order. The id must be set first and the type before the responses; otherwise SCORM 2004 returns error 408, "Data Model Dependency Not Established".
  • Stable IDs. q3-sender stays the same across versions and languages, so the LMS's interaction report remains comparable. Number-only IDs that shift when you insert a question make reports meaningless.
  • Scaled is 0 to 1. 0.8, not 80. In SCORM 1.2, send cmi.core.score.raw with min and max and set cmi.core.lesson_status to passed or failed.

What an LMS records from a SCORM course explains how to read the resulting interaction reports, which are the best source for improving the questions themselves.

Review checklist before publishing

  • Every question maps to an objective, and every objective has at least one question
  • Most questions describe a situation and ask for a decision
  • Each distractor is a real mistake, with its own feedback
  • Options are similar in length; no "all of the above"
  • A subject-matter expert has confirmed every correct answer
  • Pass mark checked against the number of questions, identical in course and manifest
  • Interaction IDs are descriptive and stable
  • Tested in the LMS: pass, fail, and read the interaction report

Writing quizzes in Scormbee

Scormbee's quiz block supports single and multiple choice with shuffled answers, in practice or graded mode, and the editor also has true/false, matching, fill in the blanks, ordering, sorting and branching scenarios. Every checked answer is sent to the LMS as an interaction, and the pass mark you set is written to both the course and the manifest. The AI assistant can draft quiz questions from your course content for 2 credits; treat its distractors as a first draft and replace weak ones with the real mistakes your experts know about. You can try it in the editor without an account, or start from a course in the template library.

Frequently asked questions

How many answer options should a multiple-choice question have?

Three plausible options are usually enough. A well-known 2005 meta-analysis by Michael Rodriguez, covering 80 years of research, concluded that three options are optimal in most cases. A fourth option is worth adding only if it's a mistake real learners make.

What is a good pass mark for an e-learning quiz?

Base it on the consequences of getting it wrong and on the number of questions. 80% is a common choice for compliance training; with 10 questions it allows two mistakes. Use the same value in the course and in the LMS.

Should I use true/false questions?

Sparingly. Pure guessing gets 50% per question, so a graded test built mostly from true/false items is easy to pass by chance. They work well for exposing a specific misconception, with the explanation in the feedback.

Should quiz feedback show the correct answer?

In practice questions, yes, with an explanation. In a graded final test, many authors show results only after submission, so learners can't pass answers to each other question by question. Either way, explain why, not just what.

How does the quiz score get to the LMS?

The course sends it through the SCORM API: cmi.core.score.raw and lesson_status in SCORM 1.2, cmi.score.scaled and success_status in SCORM 2004. Each answer can also be sent as an interaction record with the question ID, the response and the result.

More articles

12 November 2026 · 4 min read

How to Update a Dynamic SCORM Course

Publish a change to a Dynamic SCORM course in three steps: edit it, click Publish update, and every LMS shows the new version at the next launch.

Read →