Get the next issue in your inbox — clear, evidence-based GLP-1 analysis, every week.

Between August and December of 2025, a researcher at Yale went shopping for a drug I prescribe almost every day.

She wasn't a patient. She was a simulated one: a profile built to meet the eligibility criteria for a GLP-1 medication, then pointed at the open internet to see what it would sell her. She visited 49 websites. 45 of them wrote her a prescription. 34 put the medication in the mail.

2 of them did it in under 5 minutes.

The study landed in JAMA on July 6. I read it the way you'd read an inspection report on your own house. I've spent most of my career inside the industry that built these sites, and I recognized every shortcut in it.

Here's the number I can't put down. Only 55% of the sites asked whether she had an eating disorder.

That matters more than it sounds. Starting a GLP-1 in someone with an active, undiagnosed eating disorder is one of the few genuinely dangerous things you can do with these medicines, and it's exactly the kind of thing a real clinical encounter is designed to catch. Nearly half of these sites never even asked the question. The professional societies couldn't be clearer that this assessment belongs in the standard of care. These websites, apparently, disagreed.

The setup is almost too simple. One researcher, one simulated patient meeting eligibility criteria, one question: what does the market actually do when a plausible customer shows up with a credit card? The answers are worth sitting with.

Screening was optional. Only 53% of sites asked about diet and physical activity. Only 65% asked about prior weight-loss attempts. Only 18% bothered to ask whether she even had a primary care clinician. The questionnaire existed. The clinical curiosity behind it didn't.

The safety checks were theater. 9 sites (20%) required a full-body photo or a photo standing on a scale, then wrote the prescription anyway when she sent only an upper-body shot. A requirement that's never enforced isn't a requirement. It's set dressing.

The clinician was often a formality, and sometimes barely present. Only about a quarter of sites required a video visit. A handful required even a phone call. When a visit did happen, the median lasted 9 minutes. 3 clinicians prescribed to the same simulated patient across 2 or more different websites, which tells you how thin the human layer really is. One rubber stamp, many storefronts.

Commerce ran on autopilot. 76% of the prescribing sites automatically charged and shipped the medication after approval, with no confirmation step from the patient at all. You didn't so much buy the drug as trip a wire that mailed it to you.

And then there was the part that made me put the paper down for a minute. Nearly half the sites (45%) asked leading questions engineered to justify a compounded formulation. Would you like your doctor to tailor your dose? Would you like control over how the medication is injected? These aren't clinical questions. They're consent theater for a product decision that was made before you arrived. 60% of the prescriptions came bundled with supplements, B12, glycine, NAD+, carnitine, added for reasons that have far more to do with margin and differentiation than with evidence.

None of this is an argument against telehealth. I run a telehealth practice. It's not an argument against these medicines either. GLP-1 receptor agonists are among the most important tools obesity medicine has ever had, and the access this study describes is real and, for a lot of people, life-changing. So if the technology is good and the drugs are good, what exactly went wrong?

I've been employed inside venture-backed and private-equity-adjacent care models. I've sat in the meetings where the growth math gets done, and I can tell you nobody ever stood up and proposed removing the doctor. That's not how it happens. It's slower and more boring than that. Someone shares a screen. There's a funnel with a drop-off at the visit step, and the visit is the most expensive box on the slide. So the conversation becomes: how do we shrink that box.

The dilution happens one quarter at a time. Growth capital has a metabolism. It has to be fed, and it never stops being hungry. So every cycle, in a slightly different conference room, the same question gets asked: what can we take out of the process and still convert the customer?

The clinician is the most expensive, slowest, least scalable input in the entire machine. So the clinician is where the pressure always lands. First the visit gets shorter. Then it goes asynchronous. Then it becomes a questionnaire. Then the questionnaire becomes a formality. Then someone discovers you can prescribe on an upper-body photo, and the full-body requirement quietly becomes a checkbox nobody enforces. Each step is individually defensible. Each step ships. And the sum of them is the study you just read.

This doesn't require a villain. It requires a spreadsheet and a fiduciary duty to someone who isn't the patient. When a company takes outside capital, it acquires a new boss, and that boss isn't the person in the exam room and isn't the clinician trying to help them. It's the return. A care model can be genuinely excellent on the day it launches and still degrade over time, because the force acting on it never sleeps and never once points toward the patient.

The single most valuable thing a care model can do is preserve a clinician's ability to ask an inconvenient question and then act on the answer, even when that answer is "not yet," or "not you," or "let's get labs first." That capacity is expensive. It doesn't scale cleanly. It shows up as friction on every growth dashboard ever built. Which is precisely why it's the first thing outside capital erodes, and the last thing a patient can afford to lose.

The patients in this study weren't real, so no one was harmed. The next 49 people are real. Somewhere in that group is a woman with an eating disorder she was never asked about, on a compounded formulation she never needed, shipped to her door before she had a chance to reconsider. The drug isn't the problem. The absence in the room is the problem.

Somebody has to be there to ask the question. In the market this study describes, nobody did.

That's the whole story. It's also the assignment.

REFERENCES