A healthcare AI company announces an enormous audience. Somebody posts a traffic estimate or an unimpressive app ranking underneath it. The screenshot looks devastating. The company replies that outsiders do not understand its business. Everyone acquires a stronger opinion; the number often remains unresolved.
Public analytics can make corporate claims more testable. They can reveal an implausible trajectory, identify a change worth investigating or challenge a narrow statement about a visible website. They can also produce confident accusations from a comparison that never measured the claimed activity.
The standard should be demanding in both directions. A supplier does not get to replace evidence with a growth graphic. A critic does not get to replace evidence with a different graphic and call the job finished.
What the traffic service actually observes
Similarweb's methodology describes a combination of directly shared analytics, a contributory network, partners and public data, followed by processing and modelling. Its outputs are therefore not universally equivalent to opening a company's complete authenticated activity database.
That distinction changes the question a reader can answer. An estimate of visits to a domain concerns the activity represented by that product and measurement process. A claim about professionally verified people performing a defined task requires identity, qualification and task information as well.
A visit may belong to a clinician, a patient, an investor or somebody reading a news article. A professional account may contribute several visits. A user can also perform several tasks without beginning several new visits. The traffic estimate can be informative while remaining incapable of independently resolving the company's chosen headline.
An analyst should describe the observable quantity before evaluating its plausibility. Calling every visit a doctor does the supplier's marketing work for it. Calling every difference proof of fabrication makes the opposite error.
App rankings are an ordering, not a population census
A ranking says where a product stands in a specified chart, territory and period. It does not by itself supply the number of unique practising clinicians who used every version of that product. The missing quantity cannot be recovered simply by looking harder at the chart position.
Even a publisher's own analytics require interpretation. Apple's app-usage documentation distinguishes installations, sessions and active devices. It also explains that usage information depends on users agreeing to share analytics. Those are different observations, each with a reporting boundary.
The practical implication is that device activity, downloads and professional adoption should not be silently substituted for one another. A clinician can have two devices. An installation can remain unused. A well-established app could have substantial returning activity without an equivalent surge in new installations.
None of these possibilities proves that a particular company has a large hidden audience. They explain why a chart position needs a narrower interpretation than either its enthusiastic promoters or its harshest critics may prefer.
First-party dashboards can disagree too
It is tempting to imagine that an internal dashboard always contains one definitive number against which all external estimates can be judged. The dashboard still embodies choices about identity, activity, filtering and reporting periods.
Google's explanation of differences between Analytics reports and event exports describes several reasons counts can differ, including identity handling, modelling and collection timing. Access to first-party data therefore improves the evidence available; it does not eliminate the need to explain the calculation.
For a clinical product, the meaningful reconciliation should begin with the operational account system and the relevant completed activity. The analyst then needs to understand how web and app events map to those people and tasks. A screenshot without those definitions is still only a screenshot, even when supplied by the chief executive.
iatroX's guide to defining an active clinician addresses that identity-and-activity problem separately from traffic measurement.
When the arithmetic really can bite
Public measurements become more powerful as the company's statement becomes more specific. Consider an illustrative claim that 50,000 distinct clinicians use a web-only product every day in one country. Suppose an independently verified, complete measurement of that same product, country and day identifies only 10,000 distinct people.
Those two statements require reconciliation. At least one count, definition or assumption about completeness is wrong. The useful challenge is exact: identify the missing people or correct the claim. The critic does not need to speculate about motives.
Now change the second measurement to an estimate of monthly visits to the company's marketing homepage. It no longer establishes the same contradiction. The portal might sit elsewhere; the estimate may not cover the claimed activity; visits are not people. These are questions to resolve, not permission to accept any number.
The strength of the criticism comes from closing alternative explanations. A large discrepancy becomes much more informative after dates, regions, domains and populations have been aligned.
Hidden channels need evidence, not imagination
Suppliers sometimes have several routes into a product: a website, a mobile app or an interface embedded in another service. Whether a particular platform has material activity through each route must be established from its actual deployment and reporting arrangements.
A hypothetical integration cannot explain a real discrepancy. Nor should every mention of an application programming interface become an escape hatch from scrutiny. If a company says an omitted channel accounts for much of its activity, it should quantify that channel under the same definition and reporting period.
The same applies to international use. A global total might explain why a country-specific estimate is lower, but it cannot rescue an explicitly country-specific headline. The explanation has to fit the original words, not a more convenient claim invented after somebody asks a question.
Trends can be useful without yielding exact totals
A sustained change in independently observed activity can justify a question even when the absolute count is uncertain. If a public channel becomes substantially quieter while a company reports accelerating use, an analyst can investigate whether use moved elsewhere, whether the audience changed or whether the measures diverged.
The comparison should use a consistent method over time and account for changes in coverage. A one-day surge after a launch and a sustained pattern across ordinary weeks are different observations. Neither should be treated as a complete explanation of clinician behaviour without further evidence.
This is where public analytics are most useful: generating precise, answerable challenges. They can narrow a dispute from "those numbers feel ridiculous" to "your stated web-only activity does not reconcile with these observations over this period".
Publish the reconciliation, not the victory lap
For readers, the best conclusion is a visible account of what matches, what differs and what remains unexplained. Where definitions reconcile the numbers, say so. Where an inconsistency survives, identify it directly. Where the evidence only supports a question, do not promote that question into a finding.
Companies should meet that standard too. A detailed explanation strengthens a credible growth story more than dismissing every outside estimate. If the public number depends on a narrow event rule, an unusual population or a material excluded channel, that belongs alongside the headline.
This discipline applies to OpenEvidence, other large platforms and iatroX. The published adoption comparison should be read in the units each source actually reports. The aim is to make the strongest defensible statement, even when a weaker argument would make a more satisfying LinkedIn post.
Frequently asked questions
Can a low app ranking disprove a large clinician-user claim?
Not by itself. The chart may concern a different period, territory or activity from the company's claim, and an ordering does not directly reveal the total audience. A stronger challenge needs comparable measurements and an account of their coverage.
Are third-party traffic estimates useless?
No. They can reveal trends, help compare similarly measured channels and identify discrepancies worth investigating. Their value increases when the analyst preserves the metric's scope instead of relabelling it as verified professional use.
What would settle a dispute about active users?
A reproducible definition, a dated count and a reconciliation across the relevant account and activity systems would provide much stronger evidence. Independent examination can test that calculation without making clinicians' private questions public.