Should you write your own MRCPCH practice questions?

Writing your own MRCPCH practice questions can be a useful learning task when it forces you to identify one important distinction, verify the answer and explain why the alternatives fail. It should remain a small, source-checked part of revision - not a replacement for official sample material, professionally reviewed question banks or first-seen practice. Write questions to expose and repair your reasoning; do not treat an unreviewed home-made item as an authoritative assessment.

This article covers non-clinical MRCPCH examination preparation. It contains no clinical teaching, diagnostic, treatment, prescribing, dosing, safeguarding or patient-management guidance. Any medical facts used in a candidate-written question must be checked against current authoritative sources before the item is retained or shared.

The direct answer: write fewer questions, then use them properly

Candidate-generated questions are most valuable during error review. Start with a mistake, uncertain correct answer or confusing contrast. Decide what the original question was truly testing, write one clean lead-in, identify the verified answer and explain every option. Then wait before answering the item again.

Use a simple limit: one high-value question from a meaningful learning gap is better than ten rushed items. If writing the stem and distractors consumes time without improving your explanation, switch to a shorter retrieval prompt or an existing reviewed question.

The learning value is not the possession of another question. It is the work of selecting the tested idea, verifying it, predicting plausible errors and retrieving the distinction later.

Key points

• A 2025 medical-education meta-analysis found weak positive correlations between writing or answering peer-made MCQs and summative performance, but the evidence was observational, included only six studies and could not exclude negative results for question writing in future settings.
• Question generation can support learning compared with restudy, but it is not consistently superior to answering well-designed questions. Its value depends on the quality of the reasoning, verification, feedback and later retrieval.
• Treat every self-written item as a learning draft. Check the answer and explanation against current sources, remove technical clues and do not share it as authoritative without qualified review.

What the evidence actually shows

Medical-education evidence is promising but limited

A 2025 systematic review and meta-analysis combined six studies involving 1,571 undergraduate medical and allied-health students. Writing peer-made MCQs had a weak positive correlation with summative performance, as did answering them. The authors explicitly noted confounding from factors such as prior academic performance, and the prediction interval for writing questions included the possibility of a negative association in future studies.

That finding supports cautious use, not a promise that authoring questions raises MRCPCH scores.

A separate systematic review of medical students developing MCQs found mixed objective outcomes: six studies reported improved grades, two found no noticeable difference and one reported a slight fall. Guidance, peer review and expert checking often influenced question quality. The review also found that some student-written questions were poor quality.

Generating questions can outperform passive restudy

In a university study published in 2020, students who generated questions or completed tests showed better overall recall after one week than students who restudied. Bayesian analyses suggested benefits for both factual and transfer knowledge compared with restudy.

However, a three-experiment study comparing rereading, answering supplied questions and generating questions found that the outcomes depended on the test and the quality of the questions learners produced. "Write questions" is therefore not a complete method. The prompt, answer, feedback and later use matter.

Answering still has the stronger evidence base

A 2021 classroom meta-analysis integrated 222 studies and data from 48,478 students. Quizzing produced a medium overall benefit, with outcomes moderated by feedback, repetition, format and material match. If time is limited, answering and reviewing well-designed questions is a more evidence-supported default than building a large personal question bank.

The practical sequence is:

1. answer reviewed questions;
2. identify a precise gap;
3. generate a question only when construction adds useful analysis; and
4. retrieve the corrected idea later.

Why writing a question can deepen learning

Question writing requires decisions that rereading can avoid:

• What is the single learning target?
• Which information is essential to answer it?
• What tempting error would a plausible distractor represent?
• Why is the keyed answer better than every alternative?
• Which source establishes that distinction?

These decisions can reveal an illusion of understanding. If you cannot write a precise lead-in or defend the options, the underlying concept may still be unstable.

This complements MRCPCH self-explanation: the explanation becomes part of the item rather than a comment added after an answer.

When question writing is worth the time

After a high-value error

Use question generation when the error reflects a recurring distinction, misread lead-in or reasoning shortcut. Do not generate an item for every isolated slip.

The confidence-calibration guide helps distinguish a stable answer from a correct guess, recognised uncertainty or high-confidence error before you decide whether a new question is useful.

When two alternatives remain confusable

Write an item whose options make the distinction explicit. Each distractor should represent a reasoned misconception, not random vocabulary.

When a rule has conditions or boundaries

A good question can test when an approach applies, when it does not and which cue changes the decision. Keep the content non-clinical unless it has been appropriately reviewed; this workflow is about exam reasoning, not publishing guidance.

When teaching a study partner

Writing for another candidate forces clarity, but peer use adds responsibility. Label the item as learner-generated, attach direct sources and invite challenge. The study-group guide explains how to prevent confident error from circulating.

When not to write your own questions

Do not author questions when:

• you have not yet understood the source;
• the fact is time-sensitive and you cannot verify it;
• you need broad first-seen practice rather than another familiar prompt;
• distractor writing is consuming most of the session;
• you are copying wording rather than retrieving; or
• you are using question production to avoid answering difficult questions.

If the learning target is a single definition or contrast, a well-designed flashcard may be faster. If you need exam-format exposure, use official MRCPCH sample papers and reviewed question-bank material.

Use the WRITE framework

WRITE is a PaedVault editorial framework, not a validated assessment tool.

W - Write one learning target

Complete this sentence before drafting the stem:

This question tests whether I can distinguish ______ from ______ when ______.

If the sentence contains several unrelated aims, split the item. One question should have one defensible target.

R - Retrieve before researching

Draft the target, likely answer and reasoning from memory first. This exposes the gap. Then open the source and correct the draft.

Do not publish or retain the recalled version as truth. Retrieval is the diagnostic step; verification is the safety step.

I - Inspect the source and scope

Record a direct, current source URL. Check whether the claim is stable, whether important conditions were omitted and whether the question sits within the intended MRCPCH scope.

The RCPCH currently describes FOP, TAS and AKP theory questions and provides official sample papers. Use the live RCPCH structure and syllabi page for current format information rather than guessing the College's style.

T - Test every option

For each answer choice, write one sentence:

• why the keyed answer is correct;
• why each distractor is not the best answer; and
• what misconception would make that distractor attractive.

If two options can be defended from the same source, the item is not ready. If an option is obviously absurd, it contributes little discrimination or learning.

E - Edit, delay and evaluate

Remove grammar clues, unequal option length, repeated wording and unnecessary difficulty. Then wait. Answer the item after several days without opening the source, and preferably change the surface wording or context.

Track whether it helps you answer a new question - not whether you remember the letter you assigned.

A safe anatomy for a self-written best-of-five item

1. Source label

Place the source and access date above the draft. This is not part of the question; it is the audit trail.

2. Focused stem

Include only information needed to identify the tested distinction. Long does not mean authentic. Unnecessary detail can turn reasoning practice into reading endurance.

3. Explicit lead-in

The candidate should know exactly what must be selected. Avoid vague prompts such as "Which statement is true?" when a more specific question is possible.

4. Homogeneous options

Options should answer the same lead-in and belong to the same conceptual category. Mixing actions, explanations and labels can reveal the answer through form rather than knowledge.

5. Verified explanation

Explain the decision rule, not merely the correct letter. Link the source and note any uncertainty or time-sensitive element.

6. Later transfer prompt

Add a second, shorter prompt that asks for the same distinction in new wording. This prevents recognition of your own prose from masquerading as learning.

Common item-writing flaws that damage learning

A prospective crossover study of 100 paired flawed and unflawed MCQs found that the effects of individual item-writing flaws were not reliably predictable. Some flaws changed scores in unexpected ways. The conclusion was not to exploit the clues - it was to avoid flaws because they introduce irrelevant variance.

Watch for:

• the correct option being longest or most qualified;
• implausible distractors;
• grammatical mismatch between stem and options;
• two defensible answers;
• negative wording that tests visual vigilance;
• clues repeated from the source; and
• obscurity added merely to make the item "hard".

Difficulty is not quality. A question can be difficult because it tests an important distinction, or because it is badly written.

Should you use AI to generate the question?

AI can help produce variants or expose possible distractors, but it cannot verify its own output. The AI-for-MRCPCH guide explains the required source-checking workflow.

For the learning benefit of generation, draft the target and reasoning yourself before asking for assistance. If the system creates the entire item first, you may gain another question but lose the generative work.

Never assume that fluent wording, a plausible citation or a confident explanation is accurate. Check every claim against the actual source.

A 20-minute question-generation session

Minutes 0 - 3: choose the gap

Select one recurring error or uncertain distinction from a completed question block.

Minutes 3 - 7: retrieve the rule

Without notes, write the learning target, answer and reason. Mark uncertainty explicitly.

Minutes 7 - 12: verify

Open the authoritative source. Correct omissions and attach the direct URL.

Minutes 12 - 17: build the item

Write the lead-in and options. Explain every choice. Remove clues and ambiguity.

Minutes 17 - 20: schedule transfer

Create a short changed prompt for later retrieval. Do not immediately repeat the item until it feels fluent.

If the question cannot be verified within the session, quarantine it rather than adding it to active revision.

How to use self-written questions in PaedVault

Use PaedVault's notes, flags and review workflow to capture the learning target and source beside the original error. Keep learner-generated prompts separate from platform-authored questions so scores remain interpretable.

Do not count a remembered self-written item as equivalent to a first-seen question-bank result. The good question-bank score guide explains why first-seen, repeated and assisted performance should not be merged.

During the weekly MRCPCH review, archive questions that:

• duplicate an existing reviewed item;
• no longer test an active gap;
• depend on an outdated source;
• remain ambiguous after revision; or
• produce letter memory rather than reasoning.

A small maintained set is more useful than a large unreliable bank.

A quality gate before sharing a learner-written question

Do not share unless every answer is yes:

1. Is there one clear learning target?
2. Is the answer verified against a direct current source?
3. Does every option answer the lead-in?
4. Is exactly one option defensible?
5. Are the distractors plausible for a stated reason?
6. Does the explanation address every option?
7. Is it labelled as learner-generated?
8. Has a qualified reviewer checked it if others may rely on it?

Even after this gate, it remains revision material - not an official RCPCH item, prediction or recall.

Common myths

"Writing questions is always deeper than answering them"

No. Poor question generation can become slow paraphrasing. Well-designed retrieval with feedback has a stronger and broader evidence base.

"Hard distractors make a good question"

Only if they are plausible, mutually distinct and clearly inferior to the keyed answer. Ambiguity is not desirable difficulty.

"If peers agree, the answer must be right"

Agreement can spread the same misconception. Require sources and qualified review for shared material.

"A polished AI question is ready to use"

Presentation quality does not establish factual accuracy, relevance or psychometric quality.

"More personal questions mean better coverage"

Self-written questions usually cluster around what interests you or what you already noticed. Use the MRCPCH syllabus plan and reviewed banks for breadth.

Conclusion

Writing your own MRCPCH questions can turn an error into a structured act of retrieval, verification, comparison and explanation. Evidence from medical education is promising but mixed, and the association with examination performance is modest and potentially confounded.

Use WRITE: Write one target, Retrieve before researching, Inspect the source, Test every option, then Edit, delay and evaluate. Keep the activity small, source-checked and subordinate to reviewed first-seen practice.

The goal is not to impersonate an examiner. It is to understand one important distinction well enough to test it honestly - and retrieve it when the wording changes.

Frequently asked questions

Should I write my own MRCPCH practice questions?

Yes, selectively. Write a question when constructing it helps you analyse a recurring error, confusing contrast or important condition. Verify the answer and every option against a current source, then test the idea later with changed wording. Do not replace reviewed question-bank practice or official sample papers with an unreviewed personal bank.

Is writing questions better than answering practice questions?

Not consistently. Question generation can outperform passive restudy and may produce learning comparable with testing in some studies, but answering well-designed questions with corrective feedback has a larger evidence base. Use answering as the default and generation as a targeted repair tool when construction adds useful reasoning.

How many self-written MRCPCH questions should I make?

There is no evidence-based number. Start with one question for a high-value recurring gap and cap the activity by time, such as one 20-minute session. Retain only items that are unambiguous, source-checked and useful on delayed retrieval. Quality and later use matter more than bank size.

Can I share my own MRCPCH questions with a study group?

Only with safeguards. Label them as learner-generated, include direct current sources, explain every option and invite correction. If other candidates may rely on the content, obtain qualified review. Never present home-made questions as official RCPCH material, recalled exam content or verified clinical guidance.

Can AI write MRCPCH practice questions for me?

AI can suggest drafts or variants, but it can also introduce inaccurate answers, weak distractors and invented sources. Draft the learning target yourself, verify every claim against the original authoritative source and treat the output as unreviewed. AI should assist question construction, not certify correctness.

Sources

1. Royal College of Paediatrics and Child Health. Theory exams: structure and syllabi. Current page checked 26 October 2026.
2. Hwang R, Paley M, Bull D, et al. A systematic review and meta-analysis on the impact of peer-made MCQ question-bank usage on summative assessments in medical education. Journal of Advances in Medical Education & Professionalism. 2025;13(4):259-269. DOI: 10.30476/jamp.2025.104780.2084.
3. Al-Rukban MO, Munshi FM, Abdulghani HM, Al-Hoqail I. Does developing multiple-choice questions improve medical students' learning? A systematic review. Medical Teacher. 2022;44(3):270-278. DOI: 10.1080/0142159X.2021.2004305.
4. Ebersbach M, Feierabend M, Nazari KBB. Comparing the effects of generating questions, testing, and restudying on students' long-term recall in university learning. Applied Cognitive Psychology. 2020;34(3):724-736. DOI: 10.1002/acp.3639.
5. Weinstein Y, McDermott KB, Roediger HL III. A comparison of study strategies for passages: rereading, answering questions, and generating questions. Journal of Experimental Psychology: Applied. 2010;16(3):308-316. DOI: 10.1037/a0020992.
6. Yang C, Luo L, Vadillo MA, Yu R, Shanks DR. Testing (quizzing) boosts classroom learning: a systematic and meta-analytic review. Psychological Bulletin. 2021;147(4):399-435. DOI: 10.1037/bul0000309.
7. Pham H. Examining the impact of specific types of item-writing flaws on student performance and psychometric properties of the multiple-choice question. MedEdPublish. 2018;7:225. DOI: 10.15694/mep.2018.0000225.1.