A clinician is allowed to be impressed by more than one piece of software. This apparently unremarkable freedom creates a considerable problem for the companies announcing that they have captured medicine.
One doctor can use a reference tool during clinic, a documentation system afterwards and a question bank in the evening. They can also compare two answers to the same question. Each supplier may legitimately count that person as a user. None has acquired exclusive rights to their attention, judgement or future purchasing decisions.
The trouble begins when a company moves from a defensible statement about reach to a much grander statement about ownership. A large user base may establish that a product has found an audience. It does not, by itself, establish that competitors have lost that audience.
Healthcare AI needs a less possessive understanding of the people using it.
The overlap is the story
Consider a wholly hypothetical survey of 100 clinicians. Seventy used Tool A during the previous month. Fifty used Tool B. Forty used both.
There are 80 individual users across the two products: 30 using only A, 10 using only B and 40 using both. Twenty used neither. Adding the vendors' totals produces 120 product-user relationships, not 120 clinicians.
Both vendors could truthfully claim substantial reach. A could say it reached 70% of this particular sample, while B reached 50%. Those percentages can sum to more than 100 because the categories overlap. The arithmetic is entirely respectable. Presenting those figures as slices of an exclusive market would not be.
Now suppose the same survey finds that most people using both tools choose B for one demanding task and A for routine lookups. A still has greater reach. B may have a stronger position in the demanding task. Neither conclusion establishes which tool earns more revenue, receives more attention or contributes more useful work.
This is why published adoption figures resist a simple league table. Before ranking competitors, decide what competition actually means.
Four different markets hiding inside one phrase
Reach answers whether someone used a product within a stated period. Preference asks which product they would choose for a specified job. Activity measures how much they actually did. Revenue records sales recognised during a stated period.
These can move in different directions. A free tool might reach many occasional users. A specialist product might serve fewer people and earn more from each organisation. A bundled service might have enormous availability while a small fraction of eligible staff actively choose it.
There is nothing inherently suspicious about any of these models. The analytical failure is treating success in one as proof of dominance in all four.
Even the phrase "share of clinical AI" needs a boundary. Does it include transcription, literature searching, education, administrative coding and patient messaging? Are we counting clinicians, tasks, institutions or money? A claim can become almost meaningless when the category expands to contain every commercially convenient application.
The AMA's physician survey examines adoption, perceptions and anticipated impacts of AI in practice. That broad professional question is useful. An individual supplier's competitive position requires additional, product-specific questions.
Using both does not explain why
Overlap can signal complementary products. It can also signal uncertainty. A clinician might use a second system because it does something different, because it retrieves different material, because it is easier to access on another device or because the first answer did not inspire confidence.
Those explanations have very different implications. Complementary use could support durable demand for both products. Repeated cross-checking could expose an unresolved trust problem. Convenience could make apparent loyalty vulnerable to a small change in access. Curiosity could disappear after a short trial.
The overlap count alone cannot choose between them. Neither can a screenshot showing several familiar apps on a clinician's phone.
This is where interviews and observation earn their keep. Ask what prompted the second tool, which part of the first answer remained unresolved and what ultimately completed the task. A useful competitive analysis follows the work through the products. It does not stop at the moment a user opens them.
For the clinician, the practical question is whether the additional tool earns its place. Two subscriptions and two conflicting summaries are not automatically better than one well-understood workflow.
Measure preference at the level of a job
The AMA's AI evaluation guide begins its evaluation domains with clinical use case and user. That is also a sensible starting point for interpreting market claims.
Ask clinicians which tool they would first open for a clearly defined task, which they actually used during a stated period, whether another tool was required and which part of the work each completed. Allow multiple answers where the behaviour is shared. Ask for one answer only where the question is explicitly about first choice.
Keep the population stable. A survey of trainees revising for examinations and a survey of consultants searching during clinical work should not be blended into a single unexplained preference score. Both groups matter; their jobs differ.
Good measurement can acknowledge switching without pretending every switch is a defection. A doctor trying another service for a week has not necessarily abandoned the first. Conversely, an account that remains registered after habitual use has stopped is weak evidence of continuing preference.
Identity still needs care within each product. Google's analytics guidance explains how identity handling, modelling and reporting choices affect user counts. Even a properly deduplicated vendor total, however, says nothing about how many of those people also use competitors.
Availability is not a vote
An institution can purchase access for a workforce. That establishes a commercial relationship and an opportunity to use the service. It does not establish that every eligible clinician has selected the product over alternatives.
Procurement, availability and individual preference are separate decisions. They may eventually reinforce one another, but the connection has to be observed. A supplier should receive credit for winning a contract without converting the entire eligible population into enthusiastic regular users by implication.
The reverse mistake also matters. A product with modest individual visibility might be deeply embedded in an institutional process. Its value could lie in an integration clinicians rarely think about. Popularity surveys and app rankings can miss that role.
Companies therefore have a choice. They can describe the specific position they have earned, or use a larger category to make that position sound universal. Clinicians should prefer the former because it tells them something they can apply.
A product does not need to own the doctor
For iatroX, Ask-iatroX and the Tutor represent different activities within the same product family. Their presence is a useful reminder that even one platform can contain several distinct reasons to visit. Combining those activities into a single popularity number would not explain which job users came to do.
The strongest competitive position may be narrow and valuable: being the dependable first choice for a particular task among clinicians who understand its strengths. That is a credible ambition even when those clinicians use other tools elsewhere.
Market leadership can be real. It simply needs a defined market and a measure that matches the claim. The industry should stop treating every shared user as captured territory. Clinicians are assembling ways to work, learn and check their reasoning. They are not being allotted to technology companies in mutually exclusive parcels.
Frequently asked questions
Can two companies both reach a majority of doctors?
Yes, if the same doctors use both products. Majority reach does not imply exclusive use. The populations, time periods and qualifying activities must also match before the figures can be compared meaningfully.
What is the best measure of medical AI market share?
It depends on the question. Revenue can describe a commercial market; first-choice use can describe preference for a task; active reach can describe access and participation. A useful comparison names its measure instead of treating all three as interchangeable.
Does using several tools mean clinicians do not trust any of them?
Not necessarily. The tools may perform complementary jobs. Multiple use can also reflect comparison, experimentation or unresolved questions. Understanding the reason requires evidence about the workflow, beyond a list of installed products.