BMJ Best Practice

The lowest centrality grade in this category and the record says so rather than dressing it up. This is a human authored, editorially maintained evidence resource. Clinicians write the topics, editors update them, and the value is the judgement embedded in that work.

Independent comparison published in 2026 states directly that its conversational features are a layer rather than the core, and that it does not offer a generative question and answer capability comparable to the three competing subscription products, each of which shipped one between early 2024 and late 2025.

It is indexed for completeness of the reference set rather than for its artificial intelligence, and a buyer evaluating on that axis specifically should look elsewhere in the category. The same honest treatment this index gives a practice management system that has added a scribe applies here.

None, by design and correctly. A clinician reads guidance and decides; nothing is automated, nothing acts, and there is no model output to accept or override.

The one place the product does more than present information is the comorbidities feature, which adjusts a recommended plan for a patient's other conditions. That is rule based reconciliation of guidance rather than inference, and the clinician still owns every decision it informs.

This axis fits awkwardly on a product with little model to disclose, and the grade reflects what is actually disclosable rather than penalising an absence of machinery.

Where the product is transparent is methodological rather than technical: evidence is graded in evidence based medicine style with links through to primary literature and Cochrane reviews, so a reader can follow a recommendation back to the studies underneath it. Independent comparison describes that grading as a genuine strength alongside international guideline alignment.

What is not published is any description of the artificial intelligence features that do exist, their scope or their limits, which leaves a buyer unable to tell how much of the product they touch.

The exposure is small by construction and the reason is worth stating precisely, because it distinguishes this product from its generative competitors on a dimension a buyer might otherwise miss. A reference tool holds no patient record, so what it sees is the query, and clinicians asking about a real patient type in enough detail to make the question answerable. That is true across this whole category.

The difference is the interface: a search against structured topics reveals less than a free text conversation about a case, because the structured route constrains the input toward a clinical concept while a conversational one invites narrative, and narrative is where age, comorbidities, social circumstances and identifying detail arrive together.

So the same unanswered question, what happens to queries, carries materially less risk here than for the conversational products graded alongside it. No retention or query handling position was located, and no model, hosting arrangement or sub processor list was identified for whatever artificial intelligence features exist. Ask whether queries are retained and for how long, whether they inform product development, and what changes about query handling when a generative feature is used rather than the structured search.

The strongest external marker on this record is one the company did not commission.

When researchers at Google DeepMind built an evaluation of conversational artificial intelligence for disease management, published in Nature in 2026, they chose this product as one of only two core clinical guideline resources against which management decisions would be judged, alongside a national guideline body. The ground truth plans for 100 multi visit patient scenarios were derived from those sources. Being selected as the standard a model is measured against is a statement about the authority of the content, made by people with no commercial interest in it, and it is the same shape as the instrument pattern this index credits elsewhere.

Held at B rather than A because it establishes authority rather than effect. No study shows that using this product changes what clinicians do or what happens to patients, which is the gap across this whole category. Daily updating, peer review and long editorial standing corroborate quality without measuring it.

Graded on an honest basis, and the exposure is small by construction. A reference tool holds no patient record; what it sees is the query, and clinicians asking about a real patient type in enough detail to make the question answerable.

No retention or query handling position was located. The exposure is smaller here than for the generative competitors in this category, because a search against structured topics reveals less than a free text conversation about a case.

Graded on an honest basis. No compliance documentation was located in this pass.

The frame fits loosely. This is a British publisher whose largest single access route is a national health service arrangement covering staff in three countries, so the governing instruments are United Kingdom data protection law and institutional licensing rather than United States health privacy law. A United States institution should establish what applies to it specifically.

Recorded honestly and provisionally: the dedicated trust and security search this index requires was not run in this pass, and no attestation was encountered incidentally.

Widespread institutional licensing through national health service access implies procurement review at scale, and nothing about it was retrieved.

No device authorisation and none needed. The same exclusion that covers the other reference products in this category applies with more room to spare here, because a clinician can always review the basis for a recommendation: the graded evidence and the links to primary literature are the basis, presented on the page.

That is worth noting as the counterpart to the observation this index made about a competing reference tool. Where citation design is what keeps a generative product outside device regulation, here the product is outside it by its nature rather than by design, because it presents evidence rather than generating an answer.

Nothing published on evaluation, monitoring or limits for the artificial intelligence features that exist, and with little model in the product there is correspondingly little disclosed.

The governance question that does apply to a reference resource is editorial rather than algorithmic: whose evidence is included, how quickly guidance changes when the literature does, and whose populations the underlying trials studied. Guidance derived predominantly from trials in one set of health systems carries that history into every recommendation it makes elsewhere, and international alignment is claimed without a statement of how it is achieved.

The transparency here is methodological rather than technical, and this axis should grade what is actually disclosable rather than penalise a product for having less machinery to describe. Evidence is graded in the established evidence based medicine style with links through to primary literature and systematic reviews, so a clinician can follow a recommendation back to the studies underneath it and form their own view of the strength behind it.

That is contestability of the most durable kind, because the grading scale is external to the vendor and the sources are readable by anyone, and an independent comparison describes the grading as a genuine strength alongside international guideline alignment. What is absent is any description of the artificial intelligence features that do exist, their scope or their limits, which leaves a buyer unable to tell how much of the product they touch.

That matters more than it might appear on a reference product, because the evidence grading a clinician trusts belongs to the editorial process, and if a generative layer sits between the reader and that graded content the reader needs to know where one ends and the other begins. No evaluation, accuracy figure or limitations statement for those features was located, and no warranty, indemnity or remediation commitment. Ask which parts of the product are model driven, what they are permitted to alter or summarise, and whether the evidence grade survives that step intact.

Not described. The product is used through the web and a highly rated mobile application rather than from inside the record, which is the traditional model for this category and is the thing its generative competitors are actively moving away from.

That has the same consequence recorded elsewhere in this index: a reference tool outside the record knows only what the clinician types, so it cannot ground guidance in the actual medication list or results. No record system integration was located.

A hosted service reached through the web and mobile applications, with institutional access commonly mediated by a federated identity system. No hosting region or retention position was located.

The access route is worth noting because it is the practical determinant of who can use it: the largest single population reaches it through a national health service identity arrangement rather than through an institutional login.

The access model is published even though the price is not, which is worth something.

Access is free to National Health Service staff in England, Scotland and Wales, a defined and checkable population, and paid elsewhere, usually through institutional subscription. Individual paid access is described by independent comparison as a barrier for clinicians outside institutional coverage.

No rate is published for institutional or individual subscription, which is the norm in this category: of the four major subscription products, only one publishes individual pricing. A buyer should establish per seat cost, whether the comorbidities feature is included at all tiers, and what artificial intelligence features, if any, sit behind a higher tier.

Broad and hospital oriented. Coverage spans more than 30 specialties and, by the publisher's account, over 90 percent of the conditions that commonly present in a hospital setting, structured around the consultation rather than around the disease entry.

The comorbidities capability widens it further in a way competitors do not match, because real patients arrive with several conditions at once and most topic reviews address one. Reach is international with content aligned to United Kingdom practice, which is a strength in that market and a consideration everywhere else. Held at B because it addresses the clinician at the point of decision only, with nothing extending into documentation, the record or patient facing use.