On October 15, NEJM AI published a study that priced the care AI recommends. Clinicians increasingly consult AI for decision support. According to the authors, the cost of the services and treatment it recommends had not been systematically evaluated. Dvir Aran of the Technion and colleagues gave 24 AI systems the details from 200 deidentified outpatient and emergency department notes. Each system recommended tests, referrals, and medications. The researchers then priced every plan at Medicare rates, alongside the plans the treating physicians actually documented.
The AI plans averaged $314 per visit. The physicians’ plans averaged $97. That is 3.2 times more, split roughly evenly between extra testing and extra specialist referrals. The AI systems also suggested more new medications, which were counted but not included in the dollar figure.
The clearest difference showed up in two-thirds of visits, where the treating physician ordered no tests. On those visits, every AI system suggested testing anyway, usually the same basic panel: CBC, CMP, TSH, A1c, and lipids.
So whose plan was better? The study can’t tell us, because it didn’t follow patients to see what the extra workup would have found. Reviewers did check the extra recommendations against guidelines. In about one in five visits where the AI recommended more than the physician, the extra workup was guideline-concordant. Most of the time, it wasn’t.
For our practices, this raises two questions. Is the AI spending our members’ money on care they don’t need? Or was the physician too reactive, and the AI’s plan is closer to the proactive workup many of us were trained to do? A basic panel at a visit where nothing was ordered looks a lot like what a longevity practice calls a baseline.
Then there’s the bill. Would you pay three times as much for Dr. Chat’s plan? In a cash-pay practice, that question is real. Add supplements, lifestyle programs, and care team visits on top of a broader workup, and the total can climb much higher, depending on who signs off on the plan.
What does it cost to cut costs?
The visits in this study were single snapshots of care: one outpatient or emergency note, one plan. That isn’t how most of our practices work. Our patients come to us for a care journey that runs months or years. Visits are longer, a baseline workup is often the point, and we track results over time to see whether the plan is working. In that setting, whether a broad panel is wasteful depends on whether each test, referral, and therapy fits this patient, at this point in their care.
Many of our patients want more than a safe plan. They come to us because they want their care to be comprehensive and proactive, and to catch problems early rather than wait for symptoms. So trimming the AI’s list isn’t automatically the right move. A plan that saves money today but skips the assessment that would have caught a problem early isn’t saving the patient anything.
But our patients also pay for that care themselves. The study priced every plan at Medicare rates, but most of our patients pay cash, and cash prices can run higher or lower. That makes the 3.2 times figure more useful than the dollar amounts. With no insurer reviewing what gets ordered, if the AI suggests a test and we sign off, the patient pays for it, and we’re responsible for following up on the result. Cost matters, and so does what gets left out.
A functional medicine clinician on a recent onboarding call showed colleagues how she balances the two. She picks Quest, selects cash pay billed to the patient, and starts from what she normally orders, because “I can always edit that within the chart if, you know, money is an issue.” She works the list down for each patient: “start with my wish list and then I can always edit that.”
That is personalization. For each patient, she chooses the assessments, care team support, and adjunctive therapies that fit, and leaves out the rest. The AI can’t do that on its own. It doesn’t know the patient’s budget, goals, history, or what she’s already tried unless we tell it. The clinician is the one holding all of that context.
Better prompts and better design help, up to a point. In the study, a cost-aware prompt brought AI spending down to 2.5 times the physicians’ cost, and a prompt asking for parsimonious but safe care brought it to 1.4 times. Microsoft built a system that has the AI work like a panel of doctors, with one role checking cost before any test is ordered. On 304 hard diagnostic cases, it cut diagnostic costs by 20% compared with physicians. Neither approach replaces the clinician. As the same clinician told her colleagues, “your AI response is only as good as your prompt is.” We’d add that the prompt is only as good as the context the clinician puts into it.
That’s where the foundational skills come in. Precision prompting tells the AI what we need and what the patient can afford. Iterative refinement means working the draft down over a few rounds instead of accepting the first list. Hyperpersonalization is something only the clinician in the room can do: adding what we know about this patient that no chart or prompt fully captures. The AI’s first draft is a plan for everyone. What we sign should be a plan for an n of 1.
What to try this week
Build one reusable prompt that drafts the most viable care plan for each patient: the plan that fits their health picture, their goals, and what they’re willing to spend. Test it on your next five visits. The wording below is ours, not the study’s. Use it inside a HIPAA-compliant clinical tool, since consumer chatbots aren’t covered by HIPAA and anything you paste in is yours to protect.
Give it the full picture. Paste the patient’s history, goals, current plan, what they’ve already tried, and the last six months of results. Then ask which tests, referrals, or therapies would change what you do next.
Tell it what the patient wants. Say that the patient pays cash and how they like to approach their care, whether comprehensive and proactive or focused on one concern. Ask it to build the plan around those preferences.
Ask for the plan in tiers. Have it sort suggestions into what’s needed now, what can wait for the next review, and what’s optional, with the reason for each and how it would change the plan. Keep anything you’d order for safety, whatever tier it lands in.
Ask what it left out and why. This shows you its reasoning and lets you catch an assessment the patient would want that it cut.
Refine it, then save it. Go back and forth until the plan fits, then save the prompt as a snippet. As the clinician on that call put it, “Have like a good five prompts that you’re going to use with most of your patients.”
Count what you changed. For two weeks, note how many suggestions you removed and how many you added at sign-off. Your own count tells you whether the prompt is working.
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 at A4M Longevity Fest in Las Vegas
We’ll be at A4M Longevity Fest, December 11 to 13 at the Venetian and Palazzo in Las Vegas. If you’re going, use code Longevity26 to save $100 on registration, and come find us there.
In the news
Testosterone therapy in postmenopausal women was linked to more cardiovascular events in a records study, and the authors are asking for a prospective trial. The Menopause Society shared the findings ahead of its October 20 to 24 annual meeting. Among more than 6,300 propensity-matched women aged 45 and older, the composite cardiovascular outcome carried a hazard ratio of 8.08 (95% CI 4.72 to 13.82), with low absolute event rates. It is a retrospective TriNetX analysis presented at a meeting, so it shows an association and cannot establish cause, and Dr. Stephanie Faubion, the Society’s medical director, said long-term safety data on testosterone in women are lacking. For those of us prescribing low-dose testosterone to women, the practical step is to record a cardiovascular baseline and the risk conversation before starting and at each review.
A compounded tirzepatide recall is a prompt to check lot numbers against what you have dispensed. Apothecary Pharma of North Carolina is recalling 22 vials of compounded tirzepatide and one vial of NAD+ injection because sterility “cannot be assured,” per Fox Business. FDA classified it Class II on October 8, no illnesses have been reported, and a December 2025 FDA warning letter cited the company’s sterile production practices. If you dispense from a compounding pharmacy, match the lots to your records and ask for its sterile compounding documentation.
A face-photo AI estimates biological age and predicts cancer survival, and for now it is a research tool. Mass General Brigham researchers reported the FAHR-Face models, trained on more than 40 million face images, in The Lancet Digital Health on October 20. Both scores were independently associated with survival in an external cohort of cancer patients. The team runs an online research study where anyone can upload a selfie, so members may arrive having tried it. Outside experts call it a proof of concept, and the study did not test whether it adds anything to a clinician’s assessment. When members ask, we can say it was built and tested in people with cancer, and keep trending grip, VO2 max, and labs over time.
Upcoming Conferences & Events
Nov 5-8, Eudēmonia Summit · West Palm Beach, FL · One of the most talked-about longevity gatherings in the U.S. With 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, with evidence-based approaches to behavior change, chronic disease, and healthspan. Growing overlap with the longevity medicine community.
Nov 10-13, Valley Forum 2026 · Napa Valley, CA · Invitation-only forum where healthcare leaders work through the industry's hard problems, AI's role in care among them. Ultralight will be there!
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. Save $100 on registration with code Longevity26. OvationLab and Ultralight will be there!
Know of an event we should add? Reply and tell us.
Until next week
AI will draft a care plan for any patient, and the draft carries a price. What we sign should fit the patient in the room: her goals, her history, and what she is willing to pay. That context sits with us, and it goes into the prompt before it goes into the chart.
Reply and tell us how you trim an AI-drafted lab list, or send the prompt you use most. Know a colleague who is starting to use AI for care plans? Forward this to them.
Until next week, keep building the practice you imagined when you started.
Sunita & Dr. G