Free clinical AI can be sustainable when someone other than the individual searching clinician pays, or when a free reference service supports a separate paid product. Advertising, enterprise agreements and subscriptions for professional learning are different ways of funding the work. None removes the need to pay for evidence access, engineering, evaluation and support.
The commercially important question is not simply whether a tool charges the doctor. It is which activity creates revenue, whose interests the product serves and whether those incentives are explained clearly enough for clinicians and healthcare organisations to assess them.
Free access is a distribution strategy, not a complete business model
A free product can reduce purchasing friction and help clinicians establish a habit before an organisation considers a contract. Professional verification may also define the audience a supplier wants to serve. That does not mean every free product has identical eligibility, privacy arrangements or permitted uses.
OpenEvidence's official US App Store description, checked on 15 October 2026, presents its clinical evidence service as free for eligible verified healthcare professionals. OpenAI separately launched ChatGPT for Clinicians in April 2026 with US professional eligibility requirements. These are different products with their own terms, not a general permission to enter patient information into any free chatbot. OpenEvidence's listing and OpenAI's clinician product documentation.
Free access can therefore be highly valuable without being universally available or suitable for every task. A clinician should first establish the product's intended use and their eligibility, then consider what funds it.
Advertising monetises attention, not necessarily the answer itself
Sacra's September 2026 account describes advertising as central to OpenEvidence's business and estimates approximately $300 million in annualised revenue. This is a reported run rate, not audited annual revenue or disclosed profit. It does not reveal the exact margin on each clinical query. Sacra's OpenEvidence profile.
An advertising-supported service could maintain a meaningful separation between its clinical answer and its commercial placements. The appropriate questions concern how that separation works: whether promotional material is labelled, how sponsors are selected and whether commercial relationships can influence ranking, retrieval or recommendations.
It would be wrong to infer from the presence of advertising that clinical answers are necessarily distorted. It would be equally wrong to treat a statement of editorial independence as evidence that every possible influence has been excluded. The useful standard is transparent policy, inspectable product behaviour and appropriate evaluation.
A related distinction concerns advertising the platform itself. OpenEvidence's privacy wording, reviewed on 15 October 2026, discusses limited information sharing to measure advertising that promotes OpenEvidence. That is not, by itself, a complete description of the economics or targeting of advertisements displayed within its clinical product. OpenEvidence's privacy policy.
Enterprise products can charge for the work around the answer
A hospital may pay for deployment, identity management, access controls, integration, support and agreed service arrangements even when an individual evidence-search experience is free. Those organisational capabilities address a different customer problem from making a single clinical answer available.
OpenAI's current documentation illustrates the distinction. Its individual clinician product and organisation-controlled healthcare offerings have different arrangements, while the Epic plugin requires an eligible organisational deployment rather than becoming available automatically with every free clinician account. These are published product boundaries as checked on 15 October 2026. ChatGPT for Clinicians and Epic plugin documentation.
For a buyer, the question is whether the enterprise layer removes real implementation work or merely adds a purchasing tier around capabilities already available elsewhere. A proper comparison includes the configuration, assurance and maintenance that the organisation would otherwise have to perform.
Cheap inference does not make clinical software free to operate
The cost of generating text is only one part of a clinical AI service. Content rights, retrieval infrastructure, medical review, testing, security, customer support and incident response can remain substantial even when a model call becomes cheaper.
A fictional supplier might reduce its average inference bill while expanding access and increasing its total review workload. Another might use the same saving to test more outputs or provide more detailed feedback. Neither outcome can be inferred from a token-price announcement alone.
The useful economic unit is the cost of a satisfactorily completed task, including corrections and human work. For a reference product that may mean a clinically useful answer with inspectable evidence. For an agent it may mean an appropriately completed workflow. For a learning product it may mean a useful practice experience, not merely a large number of generated questions.
Paid education is a different proposition from charging for every answer
This comparison is published by iatroX and includes iatroX's own commercial model. As published on 15 October 2026, Ask-iatroX and free question access are genuinely free, with no trial expiry or verification gate. The paid subscription covers question banks, Socratic Tutor, the study planner, iatroX Simulations and CPD tools together. Simulations and CPD are included, not separate add-ons.
The published price is £99 paid upfront for a year, equivalent to £8.25 a month billed annually, or £29 a month. Three monthly payments total £87; four total £116. At those prices, annual billing becomes cheaper from the fourth month, although paying annually means committing the full £99 at the outset. iatroX's examination catalogue and CPD offering.
The value argument is not access to unrelated examinations a clinician will never sit. It is whether several learning methods address one relevant goal: identifying a misconception, practising its application, planning revision and retaining an appropriate learning record. A reader who needs only a general clinical reference answer should not be told that they need the paid educational bundle.
Different incentives require different questions
An advertising-supported product may benefit from attracting an engaged professional audience. A subscription product may benefit from retention. An enterprise supplier may benefit from wider deployment. Each model can support useful services, and each creates questions about what the business is encouraged to optimise.
A learning subscription should not encourage unnecessary use merely to preserve a streak or discourage cancellation after the learner's goal is complete. An enterprise contract should not count licences as evidence of benefit. An advertising model should not blur a clinical recommendation with a sponsored message.
These are standards to apply consistently, including to iatroX. Commercial transparency is more credible when it explains the limits of a model rather than claiming that one funding mechanism is inherently pure and another inherently compromised.
Choose by task, access and total cost
For an eligible clinician seeking rapid evidence retrieval, a free professional product may be sufficient, subject to source checking and its permitted-use boundaries. For an organisation connecting records or governing team access, a paid enterprise arrangement may solve a different and necessary problem. For an examination candidate or clinician pursuing a learning objective, a subscription can be justified by structured practice and feedback rather than by access to an answer alone.
A reader should also consider their own time. Several free tools requiring repeated copying, verification and reconciliation may be less convenient than one appropriately scoped paid service. Conversely, buying a subscription does not prove that its answers are more reliable or that its additional features are relevant.
Sustainable free clinical AI is possible. The more useful question is whether the revenue model, product boundaries and evidence of value remain understandable to the people relying on it.
Frequently asked questions
How does OpenEvidence make money if access is free?
Sacra describes an advertising-led model in its September 2026 reporting; its revenue estimates are not audited accounts or a complete disclosure of every revenue source.
Does an advertising-funded clinical AI necessarily give biased answers?
No. Clinicians should examine commercial disclosures and separation between advertisements and clinical information rather than infer either bias or independence from the funding model alone.
What does the iatroX paid subscription include?
As published on 15 October 2026, it includes question banks, Socratic Tutor, the study planner, simulations and CPD tools for £99 upfront annually or £29 monthly; Ask-iatroX and free question access remain free.