The most quoted number in question-bank marketing is the least defined. "5,000 questions" can describe one examination's bank or an entire platform; it can include full mock papers, silently or loudly; it can count each response of an EMQ separately or the question once; it can include AI-generated variants without saying so; and two products advertising the same count can differ by thousands of genuinely distinct, answerable items. None of this requires bad faith, definitions simply differ and are frequently insufficiently disclosed, but the effect on buyers is identical: headline counts are not comparable as published. This page fixes what can be fixed unilaterally: it publishes the counting taxonomy we apply to our own bank, proposes it as a disclosure standard for the category, and gives candidates the questions that make any vendor's number interpretable. It is the methodology half of a larger project; a provider-by-provider counting audit will follow it, built on these definitions with right of reply.
Why headline counts mislead without anyone lying
Four structural ambiguities do the damage. Scope: a count may describe one exam's bank, a specialty family, or the whole platform, and the same vendor may use different scopes in different placements. Unit: an EMQ with five answered responses can be counted as one question or five items; a two-part stem likewise; neither convention is wrong, but mixing them across a comparison is meaningless. Inclusion: full mocks, flashcards converted to MCQs, "past-paper-style" recalled material and generated practice variants may or may not sit inside the number. And freshness: counts accumulate, and a figure can include retired or archived items a current subscriber never sees. A candidate comparing two headline numbers is therefore usually comparing two undisclosed definitions, which is why cost-per-question arithmetic built on headline counts inherits every ambiguity it started with.
The iatroX counting taxonomy
The definitions we hold ourselves to, published so they can be checked and borrowed. One scored single-best-answer response equals one item. Each separately answered response of an EMQ equals one item, with the shared stem recorded separately and never counted as an item itself. Mock-paper questions are tagged and reported as mocks, not silently folded into the practice count. Flashcards, notes and explanations are content, not items, and are excluded from any question count. Generated variants are disclosed as generated and counted separately from the curated bank. And near-duplicate variants of one clinical decision are identified as variants where observable, because two stems testing the identical discrimination are one unit of preparation value wearing two costumes. Raw count is what a vendor reports; normalised count is what remains after these definitions are applied; the gap between them is the disclosure gap, and closing it costs a provider one paragraph.
What candidates should ask before trusting any number
Five questions convert a headline into information. Does the count cover my exam specifically, or the platform? What is one "question", and how are EMQs counted? Are mocks inside or outside the number? Is any of it AI-generated, and is that counted separately? And how many items will I actually meet as a current subscriber, excluding retired material? A provider who answers all five quickly has a real number; a provider who cannot has a marketing number, and the difference matters more than the digits. The companion discipline, remembering that volume is not value, closes the loop: beyond the coverage a blueprint requires, additional near-duplicate volume adds little, and a smaller bank with strong discrimination, honest explanations and a live corrections route routinely outteaches a larger one; how to judge that quality is its own article: /blog/can-ai-generated-medical-questions-be-trusted.
Frequently asked questions
Is a bigger bank ever the right choice?
Yes, where blueprint coverage genuinely requires it and for high-volume retrieval practice; the point is that size decides between adequate options, it does not define adequacy.
Will you publish counts for other providers?
That is the audit this taxonomy exists to enable: provider-by-provider, evidence-labelled, with a seven-day right of reply before publication; definitions first, numbers second, in that order deliberately.
How does counting affect cost per question?
Directly and misleadingly: cost per headline question rewards loose definitions, which is why our price methodology pairs every per-question figure with the counting basis behind it: /blog/how-to-compare-qbank-prices-90-day-method.
Why do EMQ counting conventions matter so much?
Because the multiplier is large: a bank with heavy EMQ content counted response-by-response can report several times the figure the question-level convention would give, from identical material; neither number is a lie, and comparing them across vendors without knowing the convention is arithmetic theatre.
Should regulators or colleges standardise counting?
A voluntary disclosure norm would do most of the work: one published paragraph per provider stating scope, unit and inclusions; this taxonomy is offered as exactly that template, free to borrow, and providers who adopt it will make their own numbers stronger.