How to Write Hard Quiz Questions (That Actually Test Knowledge)

TL;DR. A hard quiz question requires analysis, evaluation, or application — not just recall. The five techniques that reliably produce hard questions: scenario-based stems, partial-overlap distractors, "best answer" framing, multi-step reasoning, and disconfirmation framing. Trick questions are not the same as hard questions; trick questions are bad design.

What makes a question hard

Hard ≠ obscure. A question about an obscure fact ("What was the population of Paris in 1746?") is hard to answer, but it's not testing anything useful. A hard question requires *cognitive work*, not just memory.

Bloom's Taxonomy gives the cleanest hierarchy:

To write a hard question, the cognitive verb in the stem must be at the upper levels.

Five techniques for hard questions

Technique 1 — Scenario-based stems

Bury the concept in a context that requires the student to recognize it.

Easy version (recall):

Hard version (apply):

The student must extract the relevant feature ("produces ATP through oxidative phosphorylation"), match it to mitochondria, and explain the demand-driven increase. Three cognitive moves instead of one.

Technique 2 — Partial-overlap distractors

Distractors that share most properties with the correct answer, except one critical difference.

Stem: "Which of the following is the best example of a [specific concept]?"

Bad distractors: random unrelated terms.

Good distractors: four examples that all *almost* meet the criteria — three fail on one specific element.

The student can't pattern-match; they have to evaluate each option against the definition. This is Bloom's "Analyze".

For more, see Multiple Choice Distractor Design.

Technique 3 — "Best answer" framing

Multiple choice questions assume a single correct answer. "Best answer" questions have multiple defensible answers, with one that's *most* defensible.

Example:

"A patient presents with [symptoms]. Which diagnosis should the clinician investigate first?"

Two of the four options might be reasonable; one is the highest-yield first investigation. The student has to weigh trade-offs (severity, prevalence, test cost) — Bloom's Evaluate.

This is the standard structure of clinical board exams and is borrowable for any field.

Technique 4 — Multi-step reasoning

A question that can't be answered in one mental step.

Easy: "What is 15% of 80?" (one step)

Hard: "A retailer marks an item up by 25% from cost, then runs a 15% off sale. The sale price is $85. What was the original cost?" (work backward, multi-step)

For non-quantitative material, multi-step reasoning looks like:

Technique 5 — Disconfirmation framing

Most questions ask "which is true?". Hard questions can ask "which is *not* true?" or "which is *least* consistent with…".

Disconfirmation requires the student to verify all options, not just spot a pattern. It's harder by construction and prevents lucky guessing.

Use sparingly — overuse becomes a "gotcha" pattern that students learn to defeat.

What hard questions are NOT

Trick questions. A question with deliberately misleading wording isn't hard; it's bad. The student isn't being tested on the topic, they're being tested on careful reading. If a question's difficulty disappears once explained, it was a trick, not hard.

Obscure recall. "What year was X?" is hard to answer if you don't know the year. It's not testing anything beyond memory.

Overload. A question with five sentences of irrelevant context isn't hard, it's tedious. Hard questions are dense, not long.

Ambiguous. A question with two defensible correct answers (and you grade only one as correct) isn't hard, it's flawed.

A simple test

Before publishing a hard question, ask: *can a student answer this correctly without understanding the topic?*

How many hard questions per quiz

For a summative quiz, the standard mix:

Going higher than 20% hard makes the quiz exhausting and reliability drops (students start guessing). Going lower means the quiz fails to differentiate top performers.

For exam prep, mirror the real exam's distribution. Most certifications publish their cognitive distribution.

Using AI to draft hard questions

AI tools default to easy questions unless you tell them otherwise. The fix is explicit instruction:

"Generate 10 questions at Bloom's Taxonomy levels 4 (Analyze) and 5 (Evaluate). For each, the correct answer must require multi-step reasoning that cannot be derived from a single sentence in the source. Distractors must represent plausible alternative interpretations, not random wrong answers."

This produces dramatically harder output than "make 10 hard questions". For more on AI prompt design, see AI Quiz Generator Explained.

Frequently Asked Questions

Should every quiz have hard questions?

For learning, yes — even a daily formative quiz benefits from one or two stretch questions. For pure speed drills (vocabulary, math facts), no — those are about fluency.

How do I know my "hard" questions are actually hard?

Look at the data. After 30+ students take a quiz, item-level analytics show the percentage who got each question right. Hard questions should land in the 40–60% correct range. Below 30% is too hard or broken. Above 75% is medium, not hard.

Are short-answer questions automatically harder?

Not always. A short-answer question that asks for definition recall is just as easy as the same question in multiple choice. The cognitive level is what matters, not the format.

Can hard questions backfire?

Yes. Too many in a row erodes confidence and motivation. Always sandwich hard questions between easier ones, especially for younger students.

Should I tell students which questions are hard?

Generally no — the difficulty should be implicit in the assessment. The exception is high-stakes prep, where students benefit from learning to allocate time across difficulty levels.

---

Want a quiz tool that generates Bloom's-aligned hard questions on demand? Try SimpleQuizMaker free. Back to the [How to Make a Quiz pillar guide](/blog/how-to-make-a-quiz-step-by-step).