The Pulse
ChatGPT can now read your patient's medical records.
On July 23, OpenAI turned on Health in ChatGPT for US adults, letting people connect Apple Health, One Medical, Function Health, and records from Epic and Oracle Health systems.
The model now holds a longitudinal view of someone's medications, labs, visits, sleep and activity, and uses it inside ordinary conversation.
Your patient does not have to open a special screen or ask a careful question. They can type "is this normal" and get an answer built on their own numbers.
That is the biggest shift to the telehealth education layer since the category started, and almost nobody is treating it as a retention event.
Hims & Hers reported Q2 on Monday.
The print landed after the close on August 10, against expectations of roughly $699M in revenue and a small per-share loss, with the stock down about 51% over the past year.
The number worth reading is not subscriber count. It is revenue per subscriber, and what management says about retention on a base the manufacturers have repriced twice.
Subscribers in this category are getting cheaper to buy and more expensive to keep. That is a different business than the one the 2024 multiples were built on.
CMS moved against third-party remote monitoring payments.
The CY2027 Physician Fee Schedule proposed rule, published in mid-July, would tie remote patient monitoring payment more tightly to physician oversight and pull revenue away from third-party vendors.
Reimbursed monitoring has been quietly subsidising a lot of "we track your progress" features across digital health.
Narrow that subsidy and monitoring has to justify itself to the patient paying cash instead of to a payer. That turns a billing feature into a marketing one, and it lands on lifecycle.
The Deep Dive
Telehealth retention has been quietly resting on being the only informed voice in the patient's life, and as of three weeks ago that stopped being true.
The patient who plateaus at month three used to have two options: wait for your check-in email, or search something vague and get a page written for nobody.
Now there is a third. Open an app that already has their labs, their dose history and their weight trend, and ask.
1. The questions you were built to answer now get answered first.
Look at what an onboarding flow actually does in this category. It handles the side effect scare in week one, the "is this working" doubt around week six, and the plateau anxiety at month two or three.
Every one of those is a question a model with the patient's chart answers instantly, at 11pm, without waiting for a scheduled send.
Your day 14 check-in is not competing with silence anymore. It is arriving second, to someone who already has an answer.
2. Information asymmetry was doing work nobody priced.
A share of telehealth retention has never been about care quality. It has been about being the only party who could interpret the patient's own numbers back to them.
That asymmetry made a generic educational email feel valuable. It made the provider message feel like access. It kept people subscribed through the uncertain middle of a protocol.
When the asymmetry goes, everything built on top of it gets repriced at once. And the flows that were mostly information delivery stop earning their slot.
We have been treating the education gap as a retention gap for a while. The gap just moved: it is no longer whether the patient can get an answer, it is whether yours is worth reading after they already have one.
3. What survives is the part the model cannot do.
Three things stay yours, and they are narrower and more valuable than what most flows currently send:
Your record of this specific patient. Their actual dose history with you, what they reported last month, what their provider wrote down. The model has their labs, not your chart notes.
The ability to change something. A model can explain a plateau. It cannot titrate, switch the molecule, or move them to a plan that fits their budget.
A named clinician who is accountable. Routing through the provider rather than the brand has always outperformed in this category. It now carries the entire differentiation.
The practical read: any email in your flow whose whole job is to explain something general should either get specific to that patient's chart or come out of the sequence.
Takeaway: stop paying for the slot that generic education used to occupy. Rebuild those sends around the patient's own history and a clinician who can act on it, because that is the only part of your lifecycle program a free assistant with their labs cannot replicate.
Quick Takes
The manufacturer now sells the commitment device.
Novo's multi-month Wegovy subscription runs $329 a month on a three-month plan and $249 on twelve, with the price held flat through dose changes, sold through Ro, WeightWatchers, LifeMD, Hims & Hers and Sesame.
Plan length is the cheapest retention lever in telehealth. Multi-month plans retain 2-3x better than monthly.
But when the term and the discount both belong to the manufacturer, running your own "save 15% on quarterly" against it is margin you set on fire for nothing.
Sell the length on the outcome instead. "Patients who finish a full course see better results" is a claim you own. A price you do not control is not a lever.
Nobody has retention data on the pill cohort.
Oral Wegovy has passed 3 million US prescriptions since its January 5 launch: the first million in twelve weeks, the next two in ten. Citi has it running at roughly twice Zepbound's rate at the same point in launch.
These patients chose a pill largely because they were never going to choose a needle.
There is no 24-month churn curve for that cohort, because it did not exist two years ago. Running injectable-era assumptions against it is a guess, so instrument the cohort separately now and find out what its drop-off month actually is.
One Thing to Try
Open your onboarding flow and pull up the day 14 check-in.
Take the main question it answers, something like "why has my weight loss slowed" or "is this side effect normal," and put that question into ChatGPT with a realistic patient profile attached.
Read what comes back. If it is as good as your email, your email is not doing retention work, it is filler that happens to be well written.
Rewrite it around the three things the model does not have: the patient's dose history with you, their provider's note from the last visit, and their next titration date.
Send it from the provider rather than the brand. Forty minutes of work, and it is the difference between content and care.
If retention is your biggest revenue leak, that’s what we fix. growthtrigger.xyz
Reply to this with your biggest retention challenge. I read and respond to every single one.