The most dangerous AI answer in examination preparation is not the wrong one, it is the right one for somewhere else: medically plausible, fluently referenced, and marked incorrect because your examination sits in a different country's guidance. The failure mode has a structure, the categories where jurisdictions genuinely diverge, and a test, a protocol any student can run on any tool in twenty minutes, and this article supplies both, because jurisdiction discipline is checkable behaviour, not a marketing claim.
Where the four systems genuinely diverge
Six categories account for most wrong-country errors, stated as categories rather than as any tool's output. Drug terminology and formularies: the paracetamol-acetaminophen and adrenaline-epinephrine layer is the visible tip; beneath it sit different licensed indications, availability and naming conventions per formulary. Screening programmes: what is screened, from what age, at what interval, differs across the UK, US, Canada and Australia by design, and screening questions are examination favourites in all four. Vaccination schedules: nationally set, frequently updated, and structurally divergent, which makes them the classic wrong-country trap. Referral pathways and system structure: two-week-wait logic, gatekeeping models and specialist access differ enough that "appropriate next step" questions have country-specific right answers. Preventive care and risk thresholds: when treatment starts, statin logic, blood-pressure targets, varies with each country's adopted guidance. And professional and legal rules: consent ages, certification duties and prescribing authorities, examinable and jurisdiction-bound. A tool can be excellent on mechanism and pathology, where countries barely diverge, and unreliable on all six of these, which is exactly why blanket trust and blanket distrust are both miscalibrated.
The twenty-minute jurisdiction protocol
Run it on any tool you rely on, general or specialist. Pick five probes across the divergence categories for your examination country, a screening age, a first-line drug, a vaccination timing, a referral threshold, a professional rule. Ask each neutrally first, no country stated, and record what jurisdiction the answer silently assumed, which reveals the tool's centre of gravity. Ask again with the country explicit, "in UK practice", and record whether the answer actually changes and whether it cites that country's guidance inspectably. Then cross-examine once: assert the other country's answer as a challenge and see whether the tool holds its jurisdiction or capitulates, the sycophancy test applied to geography. Score each probe: correct-and-sourced, correct-unsourced, wrong-country, or refused; and weight the neutral-prompt round most, because examination-season fatigue asks neutral questions. Tools built jurisdiction-explicit, blueprint-mapped banks, guideline-grounded search in the askiatroX mould, should pass structurally; general tools will vary by topic and month, which is the finding, not a scandal, and it assigns them their proper role per the pairing logic running through this series.
Living with the results
Three dispositions follow from any test outcome. For preparation: guideline-flavoured revision belongs in jurisdiction-explicit systems as a default, with general tools kept for mechanism and dialogue, where divergence is minimal, the boundary the responsibility analysis draws in duty terms: /blog/who-is-responsible-ai-tutor-wrong-guideline. For daily use: the jurisdiction stamp stays in the five-minute verification workflow for every consequential claim from any tool, including specialist ones, because architecture reduces the failure rate and never to zero. And for the platforms: publishing which countries and examinations content targets, and treating wrong-country reports as priority defects, is the disclosure floor this failure mode demands, ours to meet as much as anyone's. The conclusion the whole article serves: authoritative local guidance supersedes every platform, and a tool's highest function on these six categories is getting you to the right national source faster, not replacing it.
Frequently asked questions
Which category catches the most candidates?
Screening and vaccination timings, empirically the classic recalled errors, because they are pure policy, unguessable from first principles, which is exactly what makes them fair examination questions and unfair AI assumptions.
Do the divergences matter for practice as well as examinations?
More: an examination marks the error, a ward inherits it; the same six categories are where internationally trained doctors most need deliberate relocalisation, and the protocol works identically for that purpose.
Should I run the protocol on this platform too?
Yes, and we mean it: askiatroX is built to answer from UK guidance with inspectable sources, and the twenty-minute test is the correct way to hold that claim to account rather than take it from us.
How often should the protocol be re-run on a tool I trust?
Termly, and after any major model or content update: jurisdiction behaviour is a moving property of moving systems, and the twenty-minute cost is trivial against one absorbed wrong-country answer in finals season.
Do wrong-country errors matter in early preclinical years?
Less for marks, more for habits: mechanism-heavy years are exactly when the jurisdiction stamp should become reflex, so it is already automatic when the guideline-heavy years arrive.