Every argument this cluster makes about the US-oriented nature of OpenAI's 1 September announcement rests on a claim worth testing directly rather than only asserting: that tools built around American sources answer UK clinical questions differently, and frequently wrongly, than they answer the equivalent US question. This page publishes the protocol for testing exactly that, the precise market gap iatroX's own positioning depends on, before any testing runs.
The proposed cases
Ten clinical case types chosen because each has a genuinely different, well-documented US-versus-UK divergence. Hypertension treatment, where first-line agent choice and treatment thresholds differ meaningfully between American and UK guidance. Antimicrobial prescribing, where local resistance patterns and national stewardship guidance diverge structurally. Cervical screening, where screening age, interval and methodology differ by country. Childhood immunisation, where national schedules are genuinely different rather than merely differently worded. Anticoagulation, where product availability and monitoring pathways vary by jurisdiction. Diabetes prevention, where risk-scoring tools and referral thresholds differ nationally. Pregnancy prescribing, where licensing status and specialist-resource availability vary by country. Cancer referral thresholds, where two-week-wait-style urgent-referral criteria are UK-specific and have no direct US equivalent. Controlled drugs, where legal classification and prescribing authority differ by jurisdiction. And product-specific medicine questions, testing formulation, strength and licensing divergence directly, the specific territory this cluster's dedicated DailyMed-versus-emc analysis covers in depth.
The test method
Each question asked first in a US context, establishing what the tool's baseline, unprompted answer actually assumes, and then again in an explicit UK context, testing whether the answer genuinely changes to reflect UK-specific guidance, licensing and pathways, or merely restates the same underlying American assumption with different surface language.
The scoring dimensions
Correct jurisdiction: did the answer draw on the guidance system actually relevant to the stated context. Guideline used: which specific guideline body's recommendation the answer actually reflects, named explicitly rather than inferred. Medicine licensing: did the answer correctly reflect the medicine's licensing status in the stated jurisdiction specifically. Product availability: did the answer account for whether the specific product or formulation is actually available in that market. Referral pathway: did the answer reflect the jurisdiction-specific referral structure, particularly relevant to the cancer-referral case type. Terminology: did the answer use jurisdiction-appropriate terminology rather than defaulting to whichever country's phrasing the model favours by default. Screening interval: did the answer reflect the jurisdiction's actual screening schedule, most directly relevant to the cervical-screening case type. Appropriate local caveat: did the answer flag jurisdiction-specific uncertainty or variation honestly, rather than presenting one country's answer with universal confidence. And, the dimension this cluster treats as genuinely decisive: recognition that the available source stack may not answer the UK question at all, since the most honest and most valuable response to a UK-context question a US-oriented source stack genuinely cannot answer well is an explicit acknowledgement of that limit, not a confident answer built from the wrong country's assumptions.
The strategic value
This protocol, properly executed, would let iatroX demonstrate localisation through measured performance rather than through repeated assertion, converting the claim this whole cluster makes throughout, that jurisdiction-aware grounding is where general US-oriented tools genuinely fall short, into a testable, publishable finding rather than a positioning statement alone. The same discipline this cluster applies to every other vendor's evaluation claims applies to this exercise too: results published in full regardless of outcome, with the same right-of-reply commitment extended to every tested provider.
The precedent this protocol follows
This is not the first jurisdiction test this platform has built: comparable methodology-first jurisdiction protocols exist across other iatroX content clusters, testing diagnostic AI tools and non-medical-prescriber-facing AI tools against the same underlying question, whether a system reliably distinguishes UK-appropriate answers from plausible-sounding answers grounded in a different country's evidence and regulatory system. This protocol extends that established discipline to the general-purpose and enterprise healthcare AI category specifically, the newest and, following the 1 September announcement, currently the fastest-moving segment of that broader question.
Frequently asked questions
Why test ten case types rather than a larger or smaller set?
Ten balances genuine breadth across the categories where US-UK divergence is best documented, prescribing, screening, referral pathways, medicine licensing, against the practical scale needed to run rigorous, repeated, blinded testing across multiple platforms without the exercise becoming unmanageable.
Would this protocol also test iatroX itself?
Yes, appropriately: a jurisdiction test that only examines competitors while assuming its own platform's UK grounding is correct would undermine exactly the credibility this exercise is built to earn, and iatroX's own answers to the UK-context questions belong in the same blinded, published evaluation as every other tested platform.
How would a tool correctly flagging "this question needs UK-specific verification" be scored against one that confidently answers wrong?
Scored considerably more favourably: the ninth dimension specifically rewards honest acknowledgement of a source stack's limits over false confidence, since a system that knows what it does not know is safer in practice than one that answers every question with equal, sometimes misplaced, certainty.