QWE AI Academy Don’t read another GLP-1 aging recap – interrogate the papers
Feeds flattened the claim into one line: GLP-1 may slow biological aging. I wanted the opposite of a summary. One weekend, ChatGPT and Claude got the actual result tables, my own blood panel, and a deliberately nasty muscle-loss scenario. What follows is that workflow – prompts you can paste, free proxies, and the gotchas headlines skip.
If you only want a vibe check, stop here. If you want a decision log you can reopen in six months, keep going.
Five prompts that replace a news thread
Free-tier ChatGPT (or Claude) is enough. Work in a fresh chat so prior hype doesn’t taint the answers.
- Paper extract (human RCT): paste the abstract + results from the Nature Communications Corley et al. post-hoc. Prompt: “Extract every epigenetic clock result, p-value, CI, and covariates. Mark pre-specified vs post-hoc. Table only.”
- Paper extract (mouse): same treatment for the Nature Feng et al. mouse lifespan paper. Prompt: “List median lifespan arms, hallmarks scored, calorie-restriction comparisons, sex/start age, and every human-translation caveat the authors wrote.”
- Free PhenoAge proxy: paste albumin, creatinine, glucose, CRP, lymphocyte %, MCV, RDW, ALP, WBC + chronological age. Prompt: “Walk Levine PhenoAge math step by step. Show intermediate terms. No coaching me toward a preferred answer.”
- Muscle stress-test: “I drop 15% body weight on semaglutide; ~30% of that loss is lean mass. Compare functional risk (grip, gait, falls) for age 55 vs 70. Use typical STEP-trial DXA ranges, not influencer anecdotes.”
- Go/no-go: “BMI 28, no diabetes, off-label interest only for clocks. Build a checklist: evidence grade, monitoring, cheaper alternatives, stop rules. Separate proven metabolic use from aging speculation.”
My labs through prompt 3 came out 3.2 years under chronological PhenoAge – free, same day. Retest the identical nine markers after any intervention; direction over six months beats one glamorous kit PDF.
Turns out the consumer methylation route is slower and pricier: TruDiagnostic TruAge lists about $499 one-time or ~$249 per test on subscription as of late 2026, finger-prick, 4-6 week turnaround. Official framing still treats these as research/wellness snapshots, not clinical endpoints – lock-in discounts don’t change that.
Ever notice how a single lab draw feels like a weather report? One storm (bad sleep, a cold, a hard training week) and the number twitches. Clocks behave the same way. That is why prompt 1 forces covariates and post-hoc flags onto the page before your brain writes a redemption arc.
What those clock moves actually were
-0.09 on DunedinPACE (~9% slower pace), PhenoAge -4.9 years/year, PCGrimAge -3.1 – those are the adjusted contrasts from the 32-week HIV-associated lipohypertrophy RCT analysis (semaglutide n=45, placebo n=39 assayed). Largest organ-system shifts sat on inflammation, brain, and metabolic clocks.
Mouse arm, started at 20 months in female C57BL/6: median lifespan 834 vs 742 days (~12%). Hallmark readouts moved in the same direction as classic calorie restriction, with extras on mitochondria and senescence markers called out by the authors.
Drugmaker presentations covered in MIT Technology Review’s ARDD 2026 write-up put proteomic clocks roughly 2-3 years younger vs placebo in large overweight/diabetic cohorts, heart-protein clock up to ~4 years; tirzepatide epigenetic slides pointed the same way. Strongest human signals still cluster in metabolically unhealthy or accelerated-aging groups – UC / study communications keep repeating the prospective-trial ask.
Prompt worth saving: “Compare DunedinPACE vs PhenoAge for short interventions. Name the original training cohorts and failure modes.”
The catch is muscle, noise, and who was studied
STEP-style DXA work shows lean mass often making up ~25-40% of total weight lost on semaglutide. Under 60, short-term strength can hold if you lift. Past 60, the same ratio can speed sarcopenia – clocks improve on paper while gait speed and fall risk do not. Pair any experiment with progressive resistance work and roughly 1.2-1.6 g protein/kg. Prompt that earned its keep: “12-week beginner strength plan for 1-1.5 lb/week loss; weekly protein targets; deload rules.”
A single methylation draw is noisy. Illness, sleep, and acute stress shove DunedinPACE around. The cleanest human epigenetic package so far is still a post-hoc slice of an HIV-lipohypertrophy cohort, not a healthy-lean longevity RCT. Community and clinician threads (plus the papers themselves) flag exactly that generalization gap.
Kits aren’t diagnoses. Compounded products add purity risk. And LLMs will invent doses if you let them – pin every number to the primary table.
Serious question to sit with before prompt 5: if a clock ticks slower but you lose stair-climbing power or facial volume you cared about, did you win? Nobody’s feed answers that for your knees.
Stack rank without the mythology
| Approach | Clock evidence (as cited above) | Cost / effort | Main catch |
|---|---|---|---|
| GLP-1 (semaglutide etc.) | Post-hoc RCT clocks + mouse lifespan + drugmaker proteomic/epigenetic slides | Rx + labs; high cash-pay | Lean-mass loss, GI effects, thin data in lean healthy adults |
| Calorie restriction | Strong animal lifespan; human clock work exists but is separate literature | Free, brutal adherence | Nutrient gaps, quit rate |
| Resistance + protein + sleep | Indirect via metabolic/inflammation markers | Low cash, real sweat | Slower cosmetic clock moves |
| Other research geroprotectors | Uneven human clock literature | Off-label variability | Different risk profile, less GLP-1-style mass data |
Ask: “Rank these four for DunedinPACE/PhenoAge using only peer-reviewed 2024-2026 sources. Short table, citations, no cheerleading.” The model usually shows where the evidence is thin within one pass.
FAQ
Does this mean I should start a GLP-1 just for aging?
No. Talk to a clinician who has your history. Aging-only use is still speculative.
Can ChatGPT really calculate my biological age accurately?
It can replay published PhenoAge arithmetic on your nine labs and show every term – a free directional proxy, not a methylation assay. I ran bloodwork through the prompt, then later compared direction to a kit; same arrow, different precision. Try pre/post lifestyle panels six months apart and ask for delta + what would invalidate the read.
What’s the biggest misconception right now?
That a slower clock automatically means a longer, better life for every body. Clocks correlate; hard outcomes and physical capacity still rule. Facial fat loss (“Ozempic face”) can make someone look older while metabolic markers improve – a trade-off viral posts skip. Mouse median gains and the ~9% pace slowdown are real signals inside their study designs. Authors still say candidate, post-hoc, needs prospective trials. Keep the caveats in the same note as the numbers when you use AI, or the tool just becomes a hype amplifier.
This week: pull your last panel, run the five prompts, paste outputs into a note titled “GLP-1 aging decision log.” Revisit after one retest cycle. Own the claim or drop it – don’t rent it from a timeline.