The 4 quiz question types

Wisteria supports four question types. You can mix and match within a single quiz. The four are deliberately small — we’ve resisted adding a fifth on the principle that more types make trainers feel like they’re choosing between similar tools rather than picking the right one.

1. Multiple-choice (MCQ)

The classic. Two to four options, one correct.

Building it

Question text — the prompt
Options — 2, 3, or 4 of them
Correct option — radio button, exactly one

Learner experience

Sees the question and all options
Taps one option to select
Sees correct/incorrect feedback after submitting all questions

When to use it

Knowledge-recall questions
Concepts with a single clear right answer
Quick to write, quick to take, no ambiguity

When NOT to use it

Questions with multiple defensible answers (will frustrate learners)
Questions where guessing is too easy (4 options + 1 correct = 25% baseline)

2. Fill in the blank

Short text answer. Learner types a word or phrase.

Building it

Question text — should hint at the expected answer
Correct answer — what the learner should type

Wisteria does case-forgiving matching: “PARIS” and “paris” both match “Paris”. Whitespace is trimmed.

Learner experience

Sees the question
Types an answer in a text input
Submits

When to use it

Vocabulary
Specific names, dates, numbers
When you want to test recall, not recognition

When NOT to use it

Questions where multiple phrasings are correct (“Paris” vs “The city of Paris”) — the learner will be marked wrong for technically-correct variants
Questions requiring multi-word answers (the matching is exact, not semantic — for semantic, use Oral)

3. Matching

Pair items from a left column with items in a right column. Up to six pairs.

Building it

Question text — the prompt explaining what to match
Pairs — 2 to 6 pairs, each with a left value and a right value
The right column is shuffled when shown to the learner; left stays fixed

Learner experience

Sees the left column (fixed order)
For each left item, picks the matching right item from a dropdown
Submits

All pairs must be correct for the question to pass.

When to use it

Vocabulary pairs (term → definition)
Causes and effects
Roles and responsibilities
Process steps and outcomes

When NOT to use it

More than 6 pairs (the UI gets cramped; split into two questions)
Pairs that aren’t actually 1-to-1 (e.g. “which of these belong to category A?” — that’s an MCQ)

4. Oral

Spoken answer. Wisteria transcribes via Whisper, then Claude grades semantically.

See Setting up an oral question for the full setup walkthrough.

Building it

Question text — the prompt
Model answer — what the ideal response sounds like (full text)
Keywords — 3–6 words/phrases that must appear in a good answer
Pass threshold — default 70% (configurable per question)

Learner experience

Sees the question
Taps mic, speaks the answer, taps stop
Wisteria transcribes the audio (Whisper)
Claude grades against the model answer + keywords
Sees the transcript + score

Scoring

40% keyword match — were the required terms present?
60% semantic similarity — does the meaning align with the model answer?

A learner gets two attempts in the moment. If both fail, the question is appended to the end of the quiz as a retry — with two fresh attempts. If those also fail, the question is marked wrong in the final results.

When to use it

Customer-facing roles: scripted greetings, handling complaints, sales objections
Procedures spoken aloud (e.g. clinical handoffs)
Soft-skills assessment
Anything where you’d judge a person on what they say, not what they type

When NOT to use it

Questions where exact wording matters more than meaning (use Fill in blank instead)
Quiet environments where learners can’t speak aloud (factory floor, open-plan office)
Languages Whisper doesn’t support well (works best in English; performance varies for other languages)

Generating questions with AI

Each module has an AI Quiz Generator button. It reads your flashcards and proposes questions across the four types. You configure type weights (e.g. 50% MCQ, 20% Fill, 20% Matching, 10% Oral) and total question count.

Two important behaviours:

Auto-detection of verbal content: if Claude sees a high density of “scripted” signal words (greeting, customer, scenario, handle), it includes oral questions even if you set 0% Oral in the slider. Override is intentional — if your content is clearly verbal, the AI insists on testing it that way.
Auto-save: AI-generated quizzes save immediately to the database. You don’t have to click Save as Draft. Ready to Submit unlocks on the next click.

Mixing types in one quiz

Most modules benefit from a mix:

MCQ for quick recall checks
Fill in blank for specific facts
Matching for relational concepts
Oral for any spoken-skill content

A 10-question quiz might be 5 MCQ, 2 Fill, 2 Matching, 1 Oral. The variety keeps learners engaged.

What learners CAN’T do

Skip a question — every question must be answered to complete the quiz.
Go back — once a question is submitted, you can’t change the answer in the same attempt.
See correct answers until they finish the whole quiz.

These are deliberate. The quiz is a test, not an exercise.