Prof. Valmed has done something genuinely notable in this category: it markets a clinical decision-support product built on a human-curated medical database and carrying CE Class IIb certification, with source categories publicly described as spanning guidelines, regulatory documents, medical-society publications and clinical studies. That regulatory achievement deserves respect rather than rhetorical diminishment, and this comparison will not pretend otherwise. What the comparison exists to do is unpack the word "validated", because it is doing several jobs at once in this market, and a prescriber choosing tools needs them separated: certification, curation, answer-level fidelity and local applicability are four different properties, and no single word, on anyone's product, should be allowed to blur them.
The six questions inside "validated"
Was the product developed within a regulated quality system? Certification speaks here, and Class IIb certification is a meaningful signal of process discipline that most of the category, iatroX's UKCA Class I registration included, does not claim at that level; credit where due. Is the software certified for its stated intended use? The intended-use statement is the certification's actual content, and reading it tells you what was assessed, which is always narrower than what marketing implies, the lesson the wider device-literacy article teaches: /blog/does-ukca-mean-clinical-ai-is-safe. Is the source database human-curated? Curation is a quality mechanism for the corpus, valuable and distinct from the next question, which is where the 2026 evidence bites. Does each answer cite the exact supporting source, inspectably? Answer-level transparency is a product behaviour, not a certification outcome, and the whole citation-fidelity problem, /blog/ai-citation-does-not-make-prescribing-answer-safe, lives at this layer regardless of how the corpus was built. Is the source current and correctly summarised? Fidelity again, plus recency, the properties no upstream process guarantees at the level of the individual answer you are reading tonight. And is the answer applicable to UK practice and the exact patient? Jurisdiction and product-level applicability, where a European certification and a European corpus meet the specific realities of NICE pathways, UK marketing authorisations, emc's product records and local NHS policy.
The comparison, run on those six
Structurally, the two platforms answer different subsets strongly. Prof. Valmed's public claims lead on the first three: regulated development, certified intended use, curated corpus, the governance-forward configuration, with the standing caveat this article's title demands: a CE mark does not by itself prove that every generated answer is correct, current or appropriate for UK practice, a sentence that applies to any certified product and diminishes none of them. iatroX's design leads on the last three as they apply to UK prescribing: answers grounded in UK guidance with the exact source one click away, medicines navigation ending at the current SmPC on emc, jurisdiction-first architecture, and the learning loop around it, with the honest converse caveat equally stated: source-grounding and UK-first design do not by themselves certify a quality system at Class IIb, and this platform does not claim they do. A prescriber weighing the two is therefore not choosing between validated and unvalidated; they are weighting six properties for their own practice, and for UK non-medical prescribing specifically, the applicability and inspectability questions tend to carry the clinical day while the certification questions carry the governance one.
What the category should take from this
The constructive close, because this comparison is really an argument about language. "Validated" should be retired as a monolith and itemised wherever it appears, on our pages and everyone's: which of the six, evidenced how, dated when. Buyers should ask for the itemisation by name, certification scope, curation process, answer-level sourcing, fidelity evidence, jurisdiction fit, and treat fluent conflation of the six as a finding in itself. And the benchmark this platform has committed to, /blog/iatrox-nmp-ai-reliability-benchmark-protocol, is the itemisation made operational: answer-level scoring, blinded, published whichever way it runs, with Prof. Valmed among the intended test set, because the only version of this comparison that finally matters is the one with measurements in it.
Frequently asked questions
Does Class IIb certification make a tool safer to rely on?
It makes the development process more assessed, which is real risk reduction at the system level; reliance on any individual answer still runs through the verification behaviours, which no certification class replaces.
Why hasn't iatroX pursued the same certification level?
Different intended use, different classification: iatroX's registered Class I scope reflects its decision-support-and-information positioning, and any future scope change would carry the regulatory work that honesty about classifications requires.
Which should a UK ANP trial first?
By binding need, as ever: governance-forward European curation versus UK-first source navigation with learning attached; the free layer makes one half of that experiment costless, and the six questions make either trial rigorous.
Could both products appear in one prescriber's stack?
Coherently, by floor: governance-forward curated reference where institutional policy adopts it, UK-first navigation and learning individually, with emc beneath both; the six questions were built to size tools, not to force a monogamy the stack logic never required.