ChatGPT for Healthcare adds agents that click through your EHR

OpenAI is pairing computer-use agents, better cited clinical answers, and more than 20 new plugins to tackle the busywork between healthcare systems

Ask any clinic manager where the day goes, and the answer is rarely patient care. It’s the EHR, the payer portal, the scheduling tool, and the policy PDF nobody can find. OpenAI’s latest update to ChatGPT for Healthcare, announced November 16, 2026, goes straight at that problem. Instead of asking staff to switch between systems, it asks an AI agent to do the switching.

The update has three parts: computer-use agents that operate existing software, improved clinical search with citations, and more than 20 new healthcare plugins.

Agents that work inside the software you already have

Computer-use agents read screens, click, type, and move between applications. That matters because many healthcare systems have no clean integration to build on. If a human can do the task through a screen, an agent can attempt it too.

OpenAI says it has made the models more capable, safer, and cheaper for clinical and administrative work. It has also added access controls and audit logs. Early deployments involve Abridge, Ambience, Cigna Healthcare, and CVS Health, covering referral processing, document transfers between systems, and assembling information for care teams.

The demo is a good example. A clinician uses voice to ask for help following up on an MRI request. The agent reviews the chart, then drafts an order and a patient reply. Both go to the clinician for review. The agent prepares, and the human decides.

Organizations choose which applications agents can reach and which steps require human sign-off. For HIPAA-covered work, they must use services covered by their BAA with OpenAI, apply the required security and retention settings, and extend protections to connected systems. OpenAI says content shared with ChatGPT for Healthcare is not used to train models.

Clinical answers that show their sources

OpenAI has also updated how ChatGPT finds, evaluates, and cites medical evidence, in both ChatGPT for Healthcare and ChatGPT for Clinicians. The goal is answers that link evidence to a clinical question and suggest practical next steps, with citations a clinician can check.

The company tested this in a blind evaluation. Physicians reviewed responses to real-world questions across multiple specialties, including multi-turn conversations, in more than 1,300 reviews. The new experience scored higher on all eight dimensions, including factual correctness, safety, actionability, and source authoritativeness. In the published chart, factual correctness rose from 89.4% to 95.5% high ratings, and safety from 86.8% to 92.4%.

These are OpenAI’s own results, and the chart’s label-to-value mapping is hard to read. Still, a blind physician review is a more meaningful method than a benchmark score, and clinicians should look for independent validation as adoption grows.

Twenty-plus plugins, one conversation

The plugin expansion covers clinical care, care coordination, and business operations. All are available now in ChatGPT for Healthcare, including through ChartGPT, which brings ChatGPT into the EHR workflow. Named integrations include:

  • Clinical and reference tools: UpToDate Expert AI, Metriport, Elation Health, K Health, Photon
  • Coordination and scheduling: Luma Health, Amion, Ferry, Assort Health, Aiva
  • Payer, coding, and cost data: Stedi, TriZetto, Turquoise, IMO Health, PenguinAI Medical Coding, Availity Extend
  • Data and connectivity: Redox, Oracle HDI, Notable, Infinitus Relay, Subsalt

The example prompts show the intent. A care coordinator can pull tomorrow’s unconfirmed appointments from Luma Health and check eligibility through TriZetto. A primary care physician can retrieve a discharge summary through Metriport, see which heart failure medications changed, and ask what monitoring UpToDate recommends. A care navigator can compare two hospitals for a knee MRI, using IMO Health for the code, Stedi for benefits, and Turquoise for negotiated rates.

Availability and what to watch

The plugins are live today. Organizations interested in agents can contact OpenAI’s sales team and start with a single workflow spanning existing applications.

The bigger picture is a shift from chatbots that answer questions to assistants that complete tasks. That’s where health systems feel the most pain, and where errors cost the most. Audit logs, scoped access, and mandatory review steps are the right design choices. But buyers should test how often agents stall on odd interfaces, how reviewers catch subtle mistakes, and whether time saved holds up outside a demo. The promise is real. The proof will come from deployments.