AI in Healthcare
HITRUST vs SOC 2: Which Does a HealthTech Company Need?
HITRUST vs SOC 2 for healthtech: assurance model, HITRUST e1, i1 and r2 levels, hospital buyer demands and the order to pursue them.
November 04, 2026Read more
Generative AI moved from novelty to everyday use in healthcare remarkably fast. This article looks at the adoption side: how quickly clinicians and organizations are taking it up, what buyers are prioritizing, what is holding adoption back and what that means for healthtech companies. It deliberately avoids repeating use cases, which we cover in our main guide to generative AI in healthcare, and it uses only sourced figures.
Physician adoption has risen quickly
The American Medical Association has surveyed physicians on AI since 2023. Its figures show how fast use has grown:
| Survey | Physicians reporting AI use | Source |
|---|---|---|
| 2023 | 38% | AMA |
| 2024 (fielded November 2024) | 66% | AMA |
| 2026 (fielded early 2026) | More than 80% | AMA |
The AMA counts any professional AI use, not only generative AI, but much of the growth has come from generative tools for documentation, summarizing research and drafting patient materials. In the 2024 survey, physicians said the features that would most increase adoption were a designated feedback channel, data privacy assurances and integration with the EHR.
Where organizations are investing first
Health systems have mostly started with generative AI for work that already has a human checkpoint and a clear time saving:
- Clinical documentation, especially ambient scribes that draft visit notes. See choosing an AI medical scribe.
- Patient messages drafted for clinician or staff review. See generative AI for patient communication.
- Revenue cycle tasks such as prior authorization requests and appeal letters. See AI in revenue cycle management.
- Knowledge search over internal policies and clinical references, with citations.
What buyers now expect
- EHR integration: tools that work inside existing workflows, not separate apps.
- Privacy assurances: signed business associate agreements covering the vendor and its model providers, and clear limits on data use and training (HHS sample BAA provisions). See HIPAA-compliant AI.
- Evidence: measured results on time saved, quality and user satisfaction, ideally from organizations like theirs.
- Transparency: clear information on how the tool works and its limits, similar to the source attributes ONC's HTI-1 rule requires from certified health IT (ONC HTI-1).
- Governance support: audit logs, admin controls and the ability to monitor and switch off features.
What is slowing adoption
- Accuracy concerns: generative models can produce fluent but wrong content, which NIST calls confabulation (NIST AI 600-1).
- Skill loss: in the AMA's 2026 survey, most physicians expressed at least some concern about loss of clinical skills.
- Integration and cost: connecting tools to the EHR and paying per user or per use.
- Regulatory uncertainty for tools that move closer to clinical decisions; FDA has sought public input on its approach to generative AI-enabled devices (FDA announcement).
What this means for healthtech companies
- Lead with workflow fit and measured time savings, not model capability.
- Have BAAs, security documentation and a model fact sheet ready before sales calls.
- Build feedback and monitoring into the product; buyers ask for them.
- Be clear about regulatory status for anything near clinical decisions.
For building these features, see building generative AI features for healthcare products. Teams adding generative AI to a product can see our approach on the AI implementation support page.
Conclusion
Physician use of AI rose from 38% in 2023 to more than 80% in 2026, according to the AMA, with generative tools for documentation and communication leading the way. Buyers now expect EHR integration, privacy guarantees, evidence and governance features. Healthtech companies that meet those expectations, and are honest about accuracy and limits, are best placed as adoption continues.
If you are bringing a generative AI product to healthcare buyers, book a consultation with Ekipa's HealthTech engineers.
Frequently asked questions
How many physicians use AI?
The AMA reported 38% in 2023, 66% in 2024 and more than 80% in its 2026 survey, counting any professional AI use.
Where is generative AI adopted first in healthcare?
Mostly in documentation, patient message drafting, revenue cycle tasks and knowledge search, where a person already reviews the output.
What do healthcare buyers expect from generative AI vendors?
EHR integration, signed BAAs with clear data limits, evidence of results, transparency about how the tool works, and governance features such as audit logs.
What slows generative AI adoption in healthcare?
Accuracy concerns, worries about skill loss, integration effort and cost, and regulatory uncertainty for tools close to clinical decisions.
Is generative AI in healthcare regulated?
Documentation and administrative tools are generally not medical devices. Tools that diagnose or guide treatment may be, and FDA has sought input on its approach to generative AI-enabled devices.
