Last week we argued that the missing variable in most clinical AI benchmarks is the person doing the invisible work before the model ever answers: assembling the patient.
We all know that job. Open the last note, scan the medication list to make sure it’s accurate, chase the outside lab, find the specialist recommendation buried in a PDF, and try to remember which follow-up never happened. Only then can we decide what matters today.
On September 1, OpenAI went directly at that problem. ChatGPT for Healthcare can now bring authorized Epic record context into ChatGPT, or appear inside a supported Epic layout.
That is a meaningful step. It is also half a workflow. The integration is read-only. It can help us understand the chart, but it cannot write anything back or place an order. And the value of clinical AI depends on more than finding the right answer. It depends on whether the tool can move a clinician-approved decision safely through the rest of the work.
What do clinicians need to know about the ChatGPT and Epic integration?
The short version:
It can review authorized Epic information such as notes, labs, medications, and specialist documentation, then summarize what changed and point back to the supporting chart entries.
It runs in a standalone ChatGPT conversation or embedded in a supported Epic layout.
A separate Healthcare Public Data plugin connects ChatGPT to nine official sources, including PubMed, DailyMed, RxNorm, ClinicalTrials.gov, and CMS Coverage.
The Epic integration is read-only. It does not write to the chart or place orders.
It is an organizational product. An individual clinician cannot switch on an Epic connection in a personal ChatGPT account.
One capability asks what happened with this patient. The other asks what the official sources say. Good clinical reasoning needs both, and neither, as announced, closes the loop inside the chart.
Our take: this is an announcement about context more than about a chatbot. For those of us who do not use Epic, the useful question is how much of this context-to-action workflow is available in tools for independent practices.
Where does the Epic integration stop?
OpenAI confirmed to Becker's Hospital Review that the integration is read-only and does not write information back to the patient record.
So it can help us reach a decision, but it cannot carry that decision anywhere: no note update, no medication reconciliation, no lab or imaging order, no referral, no follow-up task, no patient instructions. Being embedded in the chart and being able to act in the chart are different things.
We think clinical AI is moving through three layers:
Context. Turn the visit into a transcript, summary, or note.
Understand. Assemble the patient across time, identify changes and open loops, and show the source for each claim.
Execute. Move the clinician's approved decision into documentation, orders, referrals, tasks, patient instructions, and follow-up.
Most early AI tools lived in context. This announcement is a major move into understanding the story. The next competitive boundary is safe execution.
Why should clinicians outside Epic care?
Because Epic just validated a workflow problem that independent clinicians have been solving with smaller, more flexible systems: stop making the clinician reconstruct the patient by hand.
An AI-native private-practice EHR will already have more workflow depth, and needs a layer of connectors to pull in as much context as possible from other sources that live outside the chart. Today, Ultralight integrates AI across charting, visit preparation, longitudinal synthesis, care plans, and follow-up to patients. The advantage is where the AI sits: inside the system where the work continues.
The potential for where an EHR supports downstream actions, including orders, the safe standard is simple: AI prepares or routes the action, the clinician verifies and signs it.
Before buying another AI tool, ask your EHR vendor six questions:
How much additional context can be collected on the patient outside of just the traditional notion of a chart? And how much can the AI actually see?
Can it compare information across visits and over time?
Can it show the source and date behind every clinical claim?
How does it handle missing, stale, or contradictory information?
What can it write back, and which downstream actions can it prepare or initiate?
Where are clinician review, signature, permissions, and audit logs required?
Those answers tell you more than a feature labeled "AI assistant."
Does "99.1% safe" mean the output is accurate?
No. OpenAI reports that physicians evaluated responses across 27 clinical use cases, including pre-visit review, clinical timelines, medication review, and handoff summaries. Across 4,363 ratings, 99.1% of responses were rated safe.
Encouraging, and still a different measure from a 0.9% error rate in our own practices. A summary can omit a relevant trend, attach the wrong date to a result, repeat a copied-forward medication, or flatten a contradiction without producing an immediately dangerous recommendation. Read-only access removes one category of risk. It leaves untouched the risk of a clinician acting on an incomplete picture. The standard we hold is that we can see where every care-changing fact came from.
The Playbook: build a context-to-action workflow in your practice
Start with one stable, non-urgent follow-up and build a repeatable workflow around it. Use only a tool your organization has approved for protected health information, with a Business Associate Agreement in place; otherwise test on a synthetic chart.
Step | Do this |
|---|---|
1. Choose one bounded use case | A visit type you see repeatedly: a metabolic follow-up, a hormone-management visit, a medication check. For one to three comparable visits, record how long you spend preparing and how many open loops you find. |
2. Define the decision and the interval | Give the AI a bounded job instead of "review the chart": Prepare me for today's thyroid follow-up. Review the interval from the last signed visit through today. Focus on symptom change, medication and supplement changes, thyroid-related labs, outside recommendations, and unresolved follow-up items. |
3. Generate a source-linked delta | Ask for changes since the last visit, new results with dates and units, medication and outside-care changes, unresolved orders and referrals, and missing or conflicting information. Require a source and date for every claim, fact separated from inference, and "not documented" where the record is silent. No diagnosis, no treatment recommendation, no orders, no patient message. |
4. Verify the facts that could change care | Open the original source for every fact that could alter your plan and mark it confirmed, corrected, or unresolved. Then ask the patient: "I reviewed what changed in your record since our last visit. What has changed in your symptoms, medications or supplements, outside care, or priorities that may not be documented here?" |
5. Convert your decision into an action map | After you, and only you, make the clinical decision, have the system list the downstream work for confirmation: chart update, orders requiring signature, medication reconciliation, staff tasks with an owner and due date, patient instructions, monitoring threshold, next review date. |
6. Test five cases before you scale | Track preparation time, verification time, unsupported facts caught, relevant facts missed, and open loops recovered. Expand only if preparation plus verification beats your baseline, every care-changing claim is traceable, and the open loops it recovers outnumber the false positives. |
What does this integration tell us about the future of clinical AI?
For two years, clinical AI competed on how well it listens and writes. The next race is whether it can assemble the patient before the visit and safely move the clinician's decision through the work that follows.
Large health systems will pursue breadth: more of the longitudinal record, more connected sources, more institutional controls. Independent practices can redesign one workflow end to end without waiting for a health-system committee. The trap is recreating Epic's complexity by wiring together a larger pile of disconnected tools. One clinical decision, one reliable context layer, one clinician-approved path to action.
That is the standard we build Ultralight against. Context and execution live in the same system: the AI prepares the visit, synthesizes labs, diagnostics, and wearables into trends, drafts the note and the care plan, and sets up the follow-up, and the clinician stays in control of every final decision. Nobody has closed every downstream action yet, us included. But the tool that sees the chart and the tool that acts in the chart should be the same tool.
So the question to ask of any EHR: can it tell me what changed, show me where that came from, and move my approved decision through the chart without hiding uncertainty?
That is the workflow worth building. This week, ask your vendor the six questions above, then run the workflow on one stable follow-up and track total time for preparation and verification. If you reply and tell us what happened, please leave out anything that identifies a patient.
Be a modern clinician with the help of Ultralight, the AI-native EHR built specifically for functional, integrative, and longevity medicine. We've recently launched wearables integrations and improved AI-native clinical workflows. Get in touch to see the latest updates.
Join us in Coronado: MVMNT Longevity Medicine Summit
On September 22 and 23, Sunita Mohanty, co-founder and CEO of Ultralight, joins the faculty at the MVMNT Longevity Medicine Summit at Loews Coronado Bay Resort. Two days of evidence-graded longevity science, hands-on labs, and physician-built protocols, capped at 300 clinicians. If you are going, come find us.
In the news
72% of Americans want to be told when AI is used in their care, and only 16% say a provider has used it on them. A Pew survey of 3,488 US adults published August 25 found 72% call disclosure extremely or very important, including 72% for AI note-taking during appointments and 81% for AI reading scans, while 63% want more say over whether AI is used at all. Attitudes, not behavior, and a general population rather than a cash-pay panel, but the gap between 16% awareness and actual scribe adoption suggests a lot of patients are not being told. If you run an ambient scribe, a plain, up-front disclosure line is now table stakes.
The FDA authorized the first wearable that continuously reads ketones and glucose together. Abbott's Libre Duo received De Novo authorization on August 25: a 10-day sensor with readings every minute, indicated for people aged two and up with diabetes to catch rising ketones ahead of DKA, supported by six studies in more than 600 participants. The FDA says ketone readings must be interpreted alongside glucose and symptoms. Your ketogenic-diet and fasting patients will ask for it within weeks, and any use outside diabetes is off-label, so decide now how your practice handles the request and reads data the label says is not standalone.
Imaging found silent atherosclerosis in 57% of adults aged 18 to 70, including 1 in 13 aged 18 to 29. The REACT study, presented at ESC and published in NEJM on August 29, used 3D vascular ultrasound and coronary CT in 16,808 adults in Denmark and Spain; 9 in 10 of those aged 60 to 70 had plaque, men ran five to ten years ahead of women, and women's sharpest rise came between 40 and 60. Conventional SCORE2 flagged only a small minority as high risk. Cross-sectional prevalence, not outcomes, so it does not prove that imaging-guided treatment changes events. It is still the strongest population-level case yet for the image-early approach many prevention practices already take with CAC and carotid ultrasound, with a specific argument for looking earlier in perimenopausal women.
Upcoming Conferences & Events
Sept 18-19, Dr. Kara Fitzgerald Masterclass “Functional Medicine Is Longevity Medicine”· Online · Led by Kara Fitzgerald, this year’s Masterclass will continue under the theme that functional medicine is longevity medicine. Not as a trend, but as the most science-backed, clinically effective path to extending healthspan. Ultralight will be there!
Sept 22–23, MVMNT Longevity Medicine Summit · Coronado, CA · Evidence-graded longevity science, hands-on labs, and clinical frameworks you can implement the week after. Capped at 300 clinicians. Ultralight will be there!
Oct 8–10, A4M Women's Health Summit · San Antonio, TX · The best clinical education on hormone, metabolic, and midlife women's health you will see this year. The room to be in if you are growing the perimenopause and menopause side of your practice.
Oct 17-18, Roundtable of Longevity Clinics · Buck Institute, Novato, CA · Some 250 longevity clinic leaders, physicians, and researchers working toward shared standards for longevity testing and interventions. In-person and virtual tickets are open. Ultralight will be there!
Oct 21-23, DOC (Living Room Lab) · Sonoma, CA · Salon-style sessions on longevity science and medical AI with faculty from UCSF, Stanford, and the Buck, plus validated diagnostics in the Living Room Lab. Ultralight will be there!
Oct 21–24, NAMS Annual Meeting · San Diego, CA · The single most practice-changing meeting of the year for midlife women's health. Your protocols will look different after this one.
Nov 5–8, Eudēmonia Summit · West Palm Beach, FL · One of the most talked-about longevity gatherings in the U.S. Experientials, hands-on demos, and the best place to try the emerging frameworks your patients will ask you about next year. OvationLab and Ultralight will be there!
Nov 5-7, Private Physicians Alliance Annual Meeting · St. Petersburg, FL · The gathering for independent, cash-pay, and concierge physicians navigating practice independence. Practical and peer-driven. Ultralight will be there!
Nov 8-11, American College of Lifestyle Medicine Conference · Orlando, FL · Lifestyle medicine's main annual event — evidence-based approaches to behavior change, chronic disease, and healthspan. Growing overlap with the longevity medicine community.
Dec 11–13, A4M Longevity Fest · Las Vegas, NV · The biggest longevity event in the U.S. The room spans clinicians, industry, founders, and the people building next year's platforms, and the connections from this one tend to compound through the rest of your year. OvationLab and Ultralight will be there!
Know of an event we should add? Reply and tell us.
Until next week
Epic and OpenAI just proved the context problem is real and worth solving. The half they left open, moving the approved decision through the chart, is the half we can build in our own practices now. Forward this issue to the colleague who is still assembling the patient by hand.
Reply and tell us what your vendor said to the six questions. The best ideas in this newsletter come from clinicians doing the work.
Until next week, keep building the practice you imagined when you started.
— Dr. G and Sunita