Healthcare AI companies ask clinicians to trust numbers all day. Accuracy scores, evidence grades and probabilities are supposed to mean something precise. It is reasonable to expect the same precision when those companies describe their own popularity.
OpenEvidence provides a particularly revealing case. Its reach is plainly an important part of its commercial story. Yet the public language around daily physician use leaves a surprisingly basic question unresolved: what, exactly, has been counted?
This matters because a product used at some point by half a profession is a different proposition from a product used by half that profession on an average day. Both would be substantial achievements. They imply different habits, different exposure to clinical decisions and different expectations about commercial durability.
The current wording deserves a current answer
The question is not confined to an old funding announcement. The 13 October 2026 Johns Hopkins Medicine announcement describes OpenEvidence as used daily, on average, by a majority of US physicians. Its company description separately refers to millions of verified US healthcare professionals.
Those statements concern different populations. Healthcare professionals include people who are not physicians. A claim about one population cannot supply the numerator for a percentage of another. Equally, the phrase "on average" needs an object: an average across people, days, working days, or a selected group of active users?
The sensible response is a short methodological explanation. The reader should not need to solve a grammar puzzle to understand the scale of clinical adoption.
Why the distinction first attracted criticism
In December 2025, physician and clinical informaticist Ken Nepple examined the adoption wording on LinkedIn. He questioned whether the published consultation totals supported the widely repeated interpretation that 40% of US physicians used OpenEvidence every day. He also wanted clearer unique-user reporting.
The underlying July 2025 company release reported more than 8.5 million monthly physician consultations alongside daily-use language and 65,000 monthly clinician registrations. Those are three distinct measures, even when they appear in the same paragraph.
The criticism is interesting precisely because it can coexist with enthusiasm for the product. A clinician can find an application useful and still ask whether its marketing accurately describes its use. Scrutiny of the denominator is not a referendum on the quality of every answer.
Four different meanings of daily
Consider four hypothetical statements about an imaginary clinical reference service:
| Statement | What it would establish |
|---|---|
| Half the profession has registered | Cumulative account reach, subject to identity and duplication rules |
| Half the profession used it last month | Monthly reach under a specified activity definition |
| Registered users average one question per day | Mean interaction frequency across a defined user population |
| Half the profession uses it on an average day | Daily unique-person penetration across the whole denominator |
The third and fourth statements can sound similar in a funding story. They are not interchangeable. The average in the third may be driven by a relatively small number of heavy users. The fourth requires identifying distinct people each day and comparing that number with a defined physician population.
A fifth formulation is weaker still: a platform used by many doctors is also used every day. That establishes continuous activity somewhere on the platform. It says little about how regularly any particular doctor returns.
Careful reporting removes this ambiguity before it becomes part of the product's reputation.
The arithmetic can test a story without proving dishonesty
A simple calculation can expose which questions need answering. It cannot reconstruct private telemetry.
Imagine a service with 100,000 registered doctors and three million questions in a 30-day month. Dividing gives one question per registered doctor per day. That does not establish 100,000 daily active doctors. Ten thousand doctors could account for all the questions by asking ten each day, while the other 90,000 never returned.
Alternatively, almost everyone might ask one question on most days. The aggregate total accommodates radically different patterns. Without the distribution of activity, the attractive average conceals the behaviour that readers actually want to understand.
There is another trap. Logins and questions are separate events. Someone can log in, reopen an earlier answer and submit nothing new. Someone else can stay authenticated for weeks while asking questions daily. A question count therefore cannot serve automatically as an upper or lower bound for every possible definition of active use.
The correct challenge is to request reconciliation between the measures. An apparently awkward ratio is a reason to inspect definitions, reporting windows and exclusions. It is not a licence to declare that the underlying records must be invented.
Working patterns make the denominator harder, not optional
Medicine is not a profession in which everyone performs the same activity every weekday. Clinical sessions, research, teaching, leave and shifts make usage opportunities uneven. A surgeon's useful reference pattern might differ substantially from that of a general physician or a trainee preparing for examinations.
That variation creates several legitimate measures. A vendor could report all-physician daily penetration, use among clinicians who worked that day, or active days per individual clinician. Each would answer a worthwhile question. The problem arises when the easiest measure to produce inherits the strongest interpretation.
The workforce denominator also needs a date and scope. Does it contain currently practising doctors, all licensed doctors, trainees, or a particular registration category? A numerator based on individual people must be compared with a people-based denominator. Full-time-equivalent workforce estimates cannot simply be substituted.
The detailed framework in what counts as an active clinician provides a useful starting point: define the person, event and period before calculating the percentage.
A survey can corroborate popularity without auditing daily use
Asking doctors which tools they use can demonstrate that a product is familiar and valuable within the surveyed group. It can also reveal differences between specialties and career stages. However, a survey and a server-side activity report measure different things.
A voluntary poll recruited through professional social media may attract people already interested in AI. A question asking respondents to name tools they use does not establish that each tool is used daily. A respondent who selects three applications is not necessarily allocating equal time to all three.
The strongest conclusion follows the actual question and sampling method. It should not acquire a wider population, a shorter activity window or a stronger implication simply because the finding is commercially convenient.
What would settle the daily-use question?
In a later public post, Nepple again requested the missing daily-user methodology. His criticism points towards an uncomplicated remedy: a definition that another analyst could implement consistently.
A useful disclosure would state the number of deduplicated US physician identities performing a qualifying action, the qualifying action itself, the reporting window, and the denominator source. It would distinguish calendar-day from working-day averages and explain whether staff, test, automated and duplicate activity is removed.
It should also show enough history to separate sustained behaviour from a launch spike. Publishing a 30-day average alongside the median and a weekly trend would be more informative than selecting the busiest day. Historical changes in definitions should be visible, so a cleaner counting method is not mistaken for a sudden collapse or acceleration in demand.
An independent reviewer could examine that process without receiving identifiable patient questions. The audit target is the counting logic and its application, not the contents of a consultation.
Popularity still cannot do the job of clinical evaluation
Even a fully verified daily-use figure would establish exposure and habit, not clinical benefit. Repeated use can reflect convenience, easy access, reliable performance, professional curiosity or a mixture of them. The number cannot tell us which answers changed care appropriately.
This is why consultation counts require a separate clinical interpretation. Adoption and outcomes belong beside each other, with distinct definitions. Combining them into a single claim about trust makes both less useful.
The same expectation applies to iatroX. Its methodology and observable product behaviour should be assessed on their own terms. Neither a large audience nor a smaller audience substitutes for inspecting sources, intended uses and performance.
OpenEvidence does not need a more impressive adjective. It needs a public explanation of an already impressive claim. The more influential a clinical platform becomes, the less acceptable it is for its most repeated numbers to depend on interpretation.
Frequently asked questions
Has OpenEvidence's daily-user claim been proved false?
The public criticism identifies ambiguous wording and missing methodological detail. It does not provide access to the company's underlying activity records. The unresolved question is whether a clearly defined daily unique-physician measure supports the published description.
Can monthly consultations reveal daily active users?
Not by themselves. Multiple questions can come from one person, and people can use parts of a service without submitting a new question. The calculation needs a person-level activity definition and an appropriate time window.
Would an independent audit prove that OpenEvidence improves care?
An adoption audit would test the accuracy of adoption reporting. Clinical benefit would still require separate evaluation of decisions, workflow or patient outcomes against an appropriate comparator.