Novant Health's Doximity Ask rollout brings clinical evidence answers into Epic. The announcement confirms a change in where clinicians can access the tool; it does not establish that Ask automatically reads the patient's chart, writes into the record or places orders. Those are separate capabilities that should not be inferred from the word "integration".
Doximity announced the partnership on 3 September 2026. Its announcement identifies Novant's Clinical Informatics Institute and Institute for AI & Innovation as leading the work. This article assesses what that announcement supports as of 6 September 2026, rather than treating a launch statement as a completed outcomes study.
What has actually been announced?
Doximity describes answers appearing within the clinical workflow in Epic, drawing on medical literature, guidance and its physician-review system. It does not provide a technical specification of the integration, a denominator for active adoption or a measured reduction in time spent checking evidence.
That is still a meaningful product development. Its immediate subject is access and placement. The claims requiring further evidence concern what information moves between systems, how the answer is used and whether the implementation changes care or working time.
A useful distinction is: embedded access tells you where the tool appears; contextual access tells you what information it receives; write-back tells you what it can place into the record. None should stand in for the others.
A consultation question that exposes the distinction
Consider an entirely fictional follow-up appointment. The clinician wants to understand whether a published recommendation applies to someone with a relevant comorbidity and a previous treatment intolerance. The appointment is open in Epic, and an evidence tool is available nearby.
Before interpreting an answer as patient-specific, the clinician needs to know whether the tool has received those facts. An interface positioned beside the record does not answer that question. A generic answer may be appropriate for a generic prompt but incomplete for the actual consultation.
A useful opening would explicitly state the clinical learning question, the relevant context permitted by local policy and the jurisdiction. The clinician should then inspect the cited source and check whether its population and recommendation support the proposed interpretation.
This is a proposed workflow, not a report from Novant's deployment. No real patient information or measured product performance is used in the example.
Questions an implementation team should answer
The first question is scope. Which roles have access, and in which parts of the workflow? A tool available to a selected group in one clinical setting should not be described as universally available across the organisation without evidence.
The next question is data flow. What does the tool receive automatically, what must the user enter, and what does it retain? Staff need an understandable explanation, not merely reassurance that the product is integrated.
The third question is action. Can the clinician only read an answer, copy text manually, or initiate another supported function? The answer affects training and review. A reference paragraph and a signed clinical note should not be treated as the same output simply because both contain medical language.
Finally, establish how a clinician reports an unsupported answer or a workflow problem. Feedback needs enough context to investigate the issue without becoming an informal channel for unnecessary patient information.
Measure the checked answer, not just the click
A local evaluation could begin with a small set of synthetic questions and a defined reference standard. Record whether the answer addresses the question, whether its sources support the relevant statements and whether important qualifications are missing.
For workflow measurement, distinguish time to open the tool, time to receive an answer and time to finish checking it. An answer that appears quickly but requires extensive correction may not reduce the overall task time.
For adoption, distinguish access-enabled staff from people who actually use the tool, and occasional use from repeated use. These are proposed reporting categories, not figures supplied for Novant's rollout.
Include a way to record when the tool was not useful. A clinician who decides that a local policy, specialist discussion or primary source is more appropriate has made an important workflow decision. That should not disappear from an evaluation because it is not a successful chatbot interaction.
No such local evaluation was conducted for this article. Its results would need an actual implementation study, with the methods and denominator reported alongside the findings.
Integration does not settle the evidence question
A claim can be well sourced yet poorly matched to the person or setting under discussion. Check whether the supporting evidence concerns the same population, whether the recommendation is current and whether local arrangements affect implementation.
It also matters what the model was asked. An answer to "summarise the evidence" does not necessarily answer "what should happen next in this specific setting?" The latter question may require information that has not been supplied or cannot be resolved by a published paper.
The useful standard is not that every answer contains references. It is that the references support the statements the clinician intends to rely on.
Where a standalone clinical reference still fits
This article is published by iatroX and includes Ask-iatroX as a different access model, not as an Epic integration. Per its September 2026 product specification, Ask-iatroX is free clinical reference with linked sources from NICE, CKS, SIGN and SmPC information from emc. Its free access has no trial expiry or professional verification gate.
For a UK clinician with a general knowledge question, that provides an alternative route to source-linked information. It does not imply access to Novant's systems, automatic chart context or a documentation function.
A health system choosing embedded evidence access is making a workflow and governance decision as well as a content decision. An individual clinician choosing a reference tool should judge source applicability, access and the effort required to verify an answer. Those are related decisions, but they are not identical purchases.
What to watch next
The most informative next publication would explain the actual integration scope, user groups, data flow and evaluation method. Verified use and the time to a checked answer would be more meaningful than an undifferentiated count of people who could access the tool.
For now, Novant's announcement supports a clear, limited conclusion: Doximity Ask has been brought into the Epic workflow. Broader claims about automated patient context or clinical outcomes need their own evidence.
Frequently asked questions
Does Doximity Ask automatically read the Novant patient's Epic chart?
The 3 September 2026 announcement does not establish that capability. Clinicians should check the local integration specification rather than infer automatic chart access.
Does access inside Epic mean the answer is ready to copy into a note?
No. Source support, patient applicability and the intended documentation use still need review, and any write-back capability must be separately confirmed.
Has the rollout demonstrated improved patient outcomes?
The announcement reviewed for this article is not an outcomes study. It should not be presented as evidence of a measured improvement in clinical outcomes.