"We are not saying that semaglutide reverses aging or makes people younger."
That's Michael Corley, the UC San Diego geriatrics researcher whose name is attached to the most-cited "Ozempic reverses aging" headline of 2026. He said it twice, on the record, to his own university's press office. Almost no one repeating his finding quoted the sentence.
Here are the two headlines that actually traveled. One: Ozempic "reduces biological age by 3-5 years." The other: Ozempic "doubles sudden blindness." Both went viral in 2026. Both were built on fewer than 110 human beings.
Both went viral in 2026. Both were built on fewer than 110 human beings — one inflating benefit, one inflating harm, both off tiny samples.
Last week in this space, we walked through the heart-failure signal — the unglamorous, moderate-quality result that GLP-1s reduce HF-rehospitalization in patients with preserved ejection fraction. This week the same molecule is being sold as a longevity drug, and the evidence has inverted itself: the loudest claims rest on the thinnest data.
Let's rank the claims by the quality of the trial underneath them, because that ranking is the whole story.
Start with the biological-age finding, the one generating the "3-5 years younger" copy. It comes from Corley's Nature Communications analysis of a previously published trial: 108 adults with HIV-associated lipohypertrophy, roughly half on weekly semaglutide, half on placebo, tracked across 32 weeks with a panel of epigenetic clocks.
The headline number is a 9% slower pace of aging on DunedinPACE versus placebo. That is a real, randomized, double-blind, placebo-controlled result. It is also a signal in one small, deliberately unusual population.
The population was chosen precisely because it ages fast. People with HIV-associated lipohypertrophy show biological aging in excess of their chronological age — which makes them exquisitely sensitive for detecting any age-slowing effect, and simultaneously terrible for telling you what happens in a healthy 55-year-old.
Corley says this plainly. The HIV community "can help us identify interventions that may improve healthspan more broadly" — a generalizability hypothesis, not a generalizability result. Larger trials in non-HIV populations are underway to answer the question this one structurally cannot.
Then there is the deeper problem, the one the "3-5 years" framing erases. Epigenetic age is a predictor, not a measurement of lifespan. A three-year reduction in epigenetic age is a statement about future mortality risk, not three years added to anyone's life. Confirming the latter would require multi-decade follow-up nobody has run.
The popularizers know this and skip it. The write-up that put "3-5 years" in a headline also conceded, near the bottom, that these are "biomarkers, which are well-validated as predictors of aging-related outcomes but are not themselves clinical outcomes." The caveat was true and buried, which is roughly the fate of every honest caveat in this story.
Now the blindness scare, which is weaker still. David Sinclair told Joe Rogan's audience that GLP-1s double rates of sudden blindness.
Trace the number. As Brad Stanfield laid out, the alarm originates from 37 patients at a single eye clinic. The condition, NAION, runs at roughly 1 in 10,000 people per year at baseline.
Do the arithmetic the way one commenter did: if every adult in the United States took a GLP-1, a doubling of a 1-in-10,000 event is a large absolute number of people. But a signal from 37 patients at one clinic is not a population rate, and "double" applied to a rare baseline is exactly the phrasing designed to sound catastrophic while resting on almost nothing.
"It's difficult not to get irritated with this poor science communication," Stanfield wrote. That irritation is the correct response to both headlines, in opposite directions — one inflating benefit, one inflating harm, both off tiny samples.
Now the result nobody tweeted: the heart-failure meta-analysis. This is the one genuinely moderate-quality entry, and it is boring, which is why it stays buried.
It pooled four randomized trials — FLOW, STEP-HFpEF, STEP-HFpEF DM, and SELECT — real Phase III studies with thousands of patients. In HFpEF, semaglutide cut heart-failure rehospitalization with a relative risk of 0.41, 95% confidence interval 0.26 to 0.65. That is a solid, clinically meaningful effect on a hard outcome.
But read the sensitivity analysis, because it is honest about its own fragility. When the authors excluded two trials for incomplete baseline data, the estimates swung wildly — cardiovascular death moving to a relative risk of 0.27 with a confidence interval of 0.04 to 1.64, an interval so wide it spans "miracle" to "harmful" and crosses no conclusion at all.
So even the best evidence here is moderate, not overwhelming — genuine on rehospitalization, shaky on mortality, and utterly resistant to a viral headline. That is the shape of real data. It rarely fits in a tweet.
Here is the ranking, then. Strongest: a heart-failure rehospitalization benefit that nobody is calling anti-aging. Middle: an epigenetic signal in 108 accelerated-agers whose own author refuses the word "reverses." Weakest: a blindness doubling extrapolated from 37 patients. The public conversation runs that list precisely upside down.
And all of it — every tier — is moot against the one variable the longevity framing ignores: people don't stay on the drug.
These are indefinite-use medications. As Yale's Alissa Chen puts it, "These are medications that should be taken for the long term" — stop, and the weight, and presumably the metabolic benefit, returns. Jason Fung is building whole videos around the question of why people quit, to comment sections demanding he "just give the answer."
A longevity drug you take for eight months and abandon is not a longevity drug. It is a temporary intervention with a rebound. The Corley signal ran 32 weeks; healthspan claims run decades. The gap between those two numbers is the entire unsolved problem.
Meanwhile the market has run ahead of the science into open improvisation. About one in seven GLP-1 users now microdoses, per an Evidation survey — some to cut costs, some explicitly chasing longevity without weight loss.
Telehealth clinics like AgelessRx already sell microdoses as "a powerful new path to promoting long-term wellness." The evidence base for that sentence, per bariatric specialist Katy Williams, is that "there is no rigorous scientific data to support microdosing."
An ob-gyn in Santa Monica says 60% of her patients over 40 microdose, and that most feel better — a real-world signal: unblinded, uncontrolled, unmeasured. This is the layer beneath the headlines: not the trials, but what people actually do with a drug they can't reliably afford or tolerate.
Government is about to pour accelerant on all of it. The new Medicare GLP-1 Bridge program would set three weight-loss drugs at $50 a month — for 18 months, after which the subsidy expires and, by the drug's own logic, the weight comes back.
So we have a temporary program funding indefinite-use drugs, in a population — adults 60 and up — barely represented in the trials, where only about one in ten early participants was 65 or older, and where the side effects that don't matter much at 35 (muscle loss, falls, dehydration) matter enormously.
That is the real experiment, and it is not the one the biological-age clocks are measuring. The clocks measure a predictor in 108 people. The country is running a persistence experiment on tens of millions.
The honest summary is the least shareable one. GLP-1s show a moderate, real cardiovascular benefit; a small, intriguing, unconfirmed epigenetic signal in an unusual population; and a blindness scare that arithmetic dissolves.
None of it becomes a longevity story until someone shows people will take the drug for twenty years — and until an epigenetic clock, which forecasts mortality, is validated against mortality actually measured. Until then, the loudest claims and the strongest evidence are pointing in different directions, and the crowd is following the noise.