Doximity Ask

The model is the engine of the product, and a reference library and a network sit around it. Doximity Ask answers clinical questions, reads uploaded documents and drafts letters with a language model. Beside the model it offers thousands of hospital-grade drug monographs grounded in FDA labeling, full-text PDFs from more than 2,000 peer-reviewed journals, and PeerCheck, a panel of more than 12,000 physicians who evaluate and improve its answers.

It is reached inside Doximity's professional network alongside messaging, fax and Dialer. Take the model away and a drug and literature reference remains that a clinician could still use, so the model is the engine of the core module rather than the whole of what is offered.

Each answer shows its basis, and the limits of the tool are not described. Doximity Ask surfaces citations inline so a clinician can see which guideline, study or reference supports each statement and open the source in one click. PeerCheck has more than 12,000 physicians continuously evaluating and improving answers, which reviews the product over time rather than each answer before it is read.

The clinician remains the one who acts: answers support dosing, differential diagnosis and treatment choices, and letters and notes are drafted for the clinician to send. Nothing published says what the tool does when the evidence is thin or conflicting, whether it declines to answer, whether answers about an uploaded document draw only on that document, or how a preference for particular journals changes what a clinician sees.

The approach is described, and the model beneath it is not named on the current pages. Doximity Ask grounds answers in guidelines, studies and full-text journal articles with inline citations, draws dosing and prescribing answers from drug monographs grounded in FDA labeling, offers an instant mode and a slower Thinking Mode for complex cases, reads uploaded documents, and remembers each clinician's preferred journals, formats and letter style.

An earlier Doximity page for the product said its business associate agreement with OpenAI included zero data retention; that page now redirects to the Ask overview, which names no model provider. No model, version, retrieval architecture or evaluation result is published. The product has also been renamed from Doximity GPT to Doximity Ask.

No model provider is named on the current pages, and the agreement governing subcontractors is public. An earlier Doximity product page said its business associate agreement with OpenAI included zero data retention; that page now redirects to the Doximity Ask overview, which names no model provider. Doximity's member business associate agreement requires any subcontractor handling protected health information to accept restrictions no less strict than Doximity's own. Hosting is described as servers colocated in US facilities, with AWS Key Management Service for encryption keys. No subprocessor list is published.

Scale and named clinician testimony, with no published evaluation of the answers on the product pages. Doximity says its clinical AI suite has been reviewed, approved and deployed at more than 165 health systems, including 10 of the top 20 US hospitals, and names physicians who use Ask for research and for letters of medical necessity.

PeerCheck's 12,000-plus physician reviewers evaluate answers continuously, but no result from that review is published on the Ask or security pages: no accuracy rate, error rate or peer-reviewed study of Ask. Deployment counts measure adoption rather than whether answers are right.

The terms that govern patient data are published, and they permit more than the product page suggests. Doximity's member business associate agreement allows it to de-identify protected health information to the HIPAA standard and says de-identified data 'may be used, disclosed, and commercialized ... for any lawful purpose, including product development'.

It also says return or destruction of protected health information at termination is not feasible, so Doximity may retain it under the agreement's continuing protections. The Ask page says patient data and uploaded documents are handled within a secure framework, and Doximity encrypts protected health information at rest with AES-256.

No retention period for prompts or uploads is published, and the current Ask pages do not repeat the earlier commitment that prompts are never used to train models. Safety engineering, meaning guardrails, hallucination mitigation and how a safety event is handled, is described only through citations and physician review.

The business associate agreement is published, and whether it reaches Ask is left for the buyer to confirm. Doximity's security page says it enters a business associate agreement with each individual member on registration and offers an institutional agreement to health systems.

The agreement is public and incorporated into the terms of service; it flows its restrictions down to subcontractors that handle protected health information and sets reporting duties for breaches and security incidents. It applies to the 'Doximity Tools' the company identifies as appropriately secure for protected health information. The Ask page calls the product HIPAA-compliant and says patient data and uploads are handled securely, but no published list names Ask among the covered tools.

Doximity names a SOC 2 Type 2 attestation as its own and links a public SOC 3 report. Its security page says its platforms are covered by SOC 2 Type 2, describes recurring risk assessments, penetration testing by internal testers and external firms, white-box testing and a private HackerOne vulnerability program, and links a SOC 3 report as a downloadable PDF. It also calls the company 'HIPAA/HITECH certified', which is not a certification any body issues. The SOC 3 report's period and system scope, and whether Ask sits inside it, are not stated on the page.

No regulatory position is stated, and the product answers questions that sit close to clinical decisions. Doximity Ask answers questions on drug interactions and dosing, differential diagnoses, guideline summaries and treatment options, reads uploaded labs, imaging summaries and notes, and offers a Thinking Mode for complex cases such as ICU patients with multiple comorbidities.

Every statement carries an inline citation to its source and the user is a clinician, so the basis of each answer can be checked before it is acted on; that is support to a clinician rather than a decision made for a patient. Nothing published states an intended use, a limitation, or a position on clinical decision support rules. Drafted letters and notes go out under the clinician's signature.

A named review process exists, and nothing from it is published. PeerCheck has more than 12,000 physicians continuously evaluating and improving Ask's AI-generated answers, and Doximity says its clinical AI suite has been reviewed and approved by more than 165 health systems. That is a governance process with a named body behind it.

No results are published: no error categories, no accuracy by specialty, no bias or subgroup evaluation, and no account of how reviewer findings change the model. One design choice needs governing: clinicians can save preferred journals and references, which then shape later answers, so two clinicians may get different answers to the same question, and nothing says whether an answer discloses that a preference was applied.

Mechanisms let a clinician check an answer, and nothing stands behind a wrong one. Every statement in an Ask answer carries a citation to the source that supports it, and PeerCheck's physician reviewers evaluate and correct answers over time. No accuracy figure, error rate, warranty, indemnity or remediation commitment is published on the Ask or security pages. The costly errors are a wrong dose or interaction answer acted on without opening the citation, and a letter that misstates a patient's history.

No record system integration is published for Ask. Clinicians bring material in by uploading lab reports, imaging summaries, notes and letters, and send drafted letters out through Doximity's own secure fax; drafted notes are copied into the record by hand. Doximity says its clinical AI suite is deployed at more than 165 health systems, but no named record system, marketplace listing, interface standard or write-back path is described for Ask. A chart note or discharge summary that has to be pasted into the record moves work rather than removing it.

Residency is stated for the platform, and the model processing path is left open. Doximity's security page says it runs on servers colocated in US facilities with physical access controls, encrypts protected health information at rest with AES-256, and manages keys for databases holding it with AWS Key Management Service. Doximity Ask is available only to verified US clinicians, through web and mobile, as one hosted service.

The company does not say whether model inference for Ask runs in those facilities or with an outside model provider and where, or how clinicians' sessions and uploads are separated, and it describes no institutional deployment option.

The individual price is published and zero; the institutional terms are not. Doximity Ask is free on desktop and mobile to verified US clinicians, with unlimited access and no paid tier described, so the full cost to an individual is clear without speaking to anyone.

Doximity also says its clinical AI suite has been deployed across more than 165 health systems, and offers an institutional business associate agreement through its hospitals team, but publishes no institutional price, contract structure or governance terms. The free product is funded elsewhere: Doximity's fiscal 2026 annual report says its revenue-generating customers are primarily pharmaceutical manufacturers and health systems, buying its marketing, hiring and workflow solutions.

Coverage is broad across roles and tasks, with validation behind deployment rather than any specialty. The product serves verified US physicians, nurse practitioners, physician assistants and students, and covers clinical questions, drug information, literature, document review, clinical notes, patient communications and administrative letters.

Doximity reports deployment at more than 165 health systems; named testimony comes from individual physicians, and no evidence of performance is published for any specialty. Availability is limited to the United States by design. It also handles authorization paperwork on the requesting side: Ask generates prior authorization requests, insurance appeal letters and letters of medical necessity, which clinicians can send by Doximity's secure fax.