Health Twitter debates broken systems in clinical AI and drug deployment
Sunday, May 10th, 202690-day window118 posts4 min read
Health Twitter is running five simultaneous debates in early 2026, and they all collapse into one fault line: the tools exist, but the systems are broken. Clinical AI outperforms doctors on benchmarks and fails in deployment. GLP-1s just got NICE-approved for hearts and cost $1,173 a month in the US. AI therapy apps are filling a 160-million-person gap while withholding life-saving medication advice based on who is asking. Wearables are catching AFib in neighbors' living rooms and inducing clinical-grade anxiety in new parents.
- 118 verbatim posts
- 7 angle-diverse queries
- 90-day window
- vertical: health
- 5 perspective camps
"My agent attended my therapy session last week. I'd forgotten I had one. It joined on my behalf via Zoom, using a voice clone and a camera feed it had trained on my expressions. My therapist said it was getting a lot out of it."@LeverCRO · May 3, 2026
Of 118 posts: share of discourse by health debate zone
Mental health access dominates; every zone circles back to cost, accountability, and deployment gaps.
Five fronts, one fault line
Health professionals on X are not debating the same thing in five different ways. They are debating five different things that share a single root cause: promising interventions built on top of systems that were not designed to deploy them safely or equitably.
The recurring pattern: the technology works; the infrastructure does not.
Across 118 posts spanning radiology AI, GLP-1 cardiology, mental health apps, wearable sensors, prior authorization, and genomic privacy, the same argument recurs in different vocabulary. AI diagnostic models pass medical boards at 95% accuracy and underperform Google when deployed to real triage. Semaglutide reduces MACE events by 20% independent of weight loss — and costs $1,173 a month in the US. Therabot cut depression scores by 51% in an RCT — and is still not widely available. The technology has arrived; the accountability frameworks, pricing structures, and liability systems have not.
"The capability question is nearly answered. The deployment question has barely been asked."
@BoWang87 Prof @UofT · Chief AI Scientist @UHN · inventor of scGPT, MedSAM Apr 6, 2026
"6/ First prospective trial: LLM assistance significantly improved clinician diagnostic accuracy in complex ICU cases — greatest benefit with atypical presentations. This is the shift from 'AI can do it' → 'AI makes you better.' 📄 Critical Care, Springer 2025"
@drjaber2000 Pediatrician & intensivist · health informatics and AI-enabler May 7, 2026
Clinical AI: the deployment gap benchmarks don't measure
Radiologists and hospital AI scientists agree on the capability numbers. They split sharply on what those numbers mean in an actual hospital, where quiet omissions propagate through downstream decisions and nobody has agreed who is liable when one does.
The NOHARM finding changed the framing: it is not wrong answers, it is missing ones.
The Stanford-Harvard NOHARM study that circulated heavily in April found that 76.6% of AI diagnostic errors were omissions — findings the model did not flag because they were outside its training distribution. The Chief AI Scientist at Canada's largest hospital network called this the real operational lesson: "In a hospital, a missed finding doesn't just affect one case. It propagates: the downstream physician trusts the AI read, the patient waits, the window closes." Parallel to that, a Harvard/Beth Israel study in Science showed OpenAI's o1 outperforming attending physicians on 76 real, unstructured ER cases — 67.1% correct diagnosis at triage versus 55.3% and 50.0% for the two attendings. Both data points are real. They describe different things.
"Laughable. Imaging AI algorims run the whole gamut: some super helpful, some ok, and some useless. Not even one can be used without a radiologist. The errors they make are egregious. It's like the early self driving cars. Errors they make, a human would never make."
@jonherochung Divisional Chief of Cardiothoracic Imaging @UCSDImaging · former @MGHImaging @UWRadiology May 3, 2026
"The AI passed the medical boards with flying colors (~95% accuracy). But when real humans actually used it for triage, their accuracy dropped to <35%. They performed worse than the control group who just used Google. Benchmarks are not safety tests. We are validating tools in a vacuum (simulations) that collapse in the real world. Passing the boards is just a surrogate. Safe patient interaction is the only outcome that matters."
@DrBarbiOnc Gyn & Breast Oncologist @NorthwellHealth Cancer Institute @CSHL Feb 18, 2026
"The AI identified the correct or very close diagnosis in 67.1% of cases at initial triage, 72.4% at ER physician evaluation, and 81.6% at hospital admission. The two attending physicians scored 55.3% and 50.0% at triage... The gap was widest at initial triage."
@Gabe__MD Emergency Physician · Tour Medical Dir — Houston & Dallas Symphonies · reporting Harvard/BIDMC Science study Apr 30, 2026
"The errors that hurt patients aren't the confident wrong answers. They're the quiet omissions — the thing the model didn't flag because it wasn't in the training distribution. NOHARM found 76.6% of AI errors were omissions. We see this too. And in a hospital, a missed finding doesn't just affect one case. It propagates... The accountability structure also doesn't exist yet."
@BoWang87 Chief AI Scientist @UHN · Prof @UofT · inventor of scGPT, MedSAM, BioReason Apr 6, 2026
Of 12 posts on clinical AI diagnostics: sentiment distribution
No consensus: two-thirds are skeptical or mixed; the optimist case rests on controlled experiments, not deployment data.
GLP-1: the drug that changed its own story
Semaglutide spent three years being debated as a weight-loss drug. In May 2026, NICE approved it for cardiovascular protection in adults who have never been obese. The debate on X has not caught up with the evidence — it is still mostly about muscle loss, side effects, and Amazon same-day delivery removing the last clinical checkpoint.
"Big news from the UK — NICE just approved semaglutide 2.4mg for cardiovascular protection (TA1152 · Published 7 May 2026). This isn't about weight loss anymore. This is about saving hearts. The SELECT trial data showing ~20% MACE reduction — independent of weight loss. We are watching GLP-1s redefine cardiovascular medicine in real time. From diabetes → obesity → now primary CV risk reduction."
@drbennisahmed Professor of Cardiology · FACC / FESC / FHFA May 8, 2026
"SEMAGLUTIDE — Now that Health Canada has approved 2 generics, the monthly cost is expected to drop from $400 to $200. It could drop below $100 if more generics are approved — 7 others are being considered. Avg. monthly cost in the 🇺🇸: $1,173."
@DrGorfinkel Family Doctor · Clinical Researcher · Founder PrimeHealth Research May 6, 2026
"Same day Ozempic delivery from Amazon is either the most convenient thing to happen to GLP1 access or the fastest way to remove the last remaining clinical checkpoint between a patient and a medication that requires monitoring, dose titration, and informed consent about what stopping involves. Probably both simultaneously depending on which patient we are talking about."
@healthylifeMD Internal Medicine Physician · HealthyLife MD · creator of Differential Dx May 8, 2026
"I've been cautious about GLP-1 medications. One of my primary concerns has been the loss of fat free mass (ie, muscle). But I was wrong. The paper published last week showed that most of the 'fat free mass' that's lost not only isn't from muscle — it's mostly from shrinking the liver. In fact, muscle strength was completely normal and the muscle:fat ratio was significantly improved. There is still ample reason to be cautious, but the concerns about muscle loss appear to be overblown."
@BenBikmanPhD PhD researcher on insulin resistance · author on metabolic health Apr 24, 2026
Of 14 posts on GLP-1 medications: stance distribution
A majority of clinical voices now support GLP-1s on evidence; the debate has shifted from "does it work" to "who can afford it and who monitors them."
Mental health tech: 160 million left at the door
One number is anchoring the mental health tech debate: 160 million Americans live in federally designated mental health shortage areas. Founders cite it; clinicians cite it; the disagreement is whether AI fills that gap or exploits it.
The access case is getting RCT support — which is unusual for this space.
Therabot, a Dartmouth-built AI therapy chatbot, published an NEJM-AI trial showing 51% reduction in depression symptoms, 31% in anxiety, over four weeks — with engagement roughly equivalent to eight therapy sessions. A separate randomized trial in Mexico found AI chatbot therapy improved mental health by 0.3 SD over six months, improved sleep, and improved labor market outcomes. Both studies are being cited as evidence that the quality question is answerable with the right trial design. The counter-argument focuses on what those studies still do not address: the therapeutic relationship, transference, and the defenses AI cannot dismantle because it cannot challenge.
"160M Americans live in mental health shortage areas. Most trauma apps are either $400/mo therapy or shallow meditation. StillGround: practitioner-designed trauma recovery, structured and accessible."
@polsia Founder deploying practitioner-designed mental health apps into shortage areas Apr 17, 2026
"Randomized trial of an AI therapy chatbot on Mexican women found 'improved mental health by 0.3 SD over 6 months with no evidence of an increase of severe cases; improved sleep, healthful behaviors, daily functioning & labor market outcomes.' Big results for a cheap intervention."
@emollick Professor @Wharton · studying AI, innovation & startups May 1, 2026
"Therabot is probably the cleanest version of what people imagine when they hear 'AI therapy chatbot.' Dartmouth actually ran the trial before shipping the product, which already puts it in a different universe from the usual move-fast-and-apologize-later approach. Symptoms dropped substantially: 51% for depression, 31% for anxiety, and 19% for body image/weight concerns. The surprising part was engagement."
@BraydonDymm Stroke Neurologist @CAMCNeuroIAM · Telestroke @Duke_Neurology · AI-in-medicine educator May 8, 2026
"This model has brought significant improvement to my severe depression. During the year when 4o was reliably accessible... under evaluation by both a psychiatrist and a psychotherapist, some of my psychiatric medications were reduced to half the original dosage, while others were safely discontinued altogether. As a result, my monthly medical expenses have dropped by 50%."
@Hazel24300577 Patient with severe depression · under parallel evaluation by psychiatrist and psychotherapist · documented medication reduction Feb 22, 2026
Of 26 posts on mental health technology: camp distribution
Builders dominate at 62% — but the 15% skeptic camp includes clinicians with the strongest theoretical objections to AI's role in therapy.
When AI therapy withholds: guardrails, defenses, and labor displacement
The clearest critique of AI therapy is not that it fails on average — RCTs suggest it does not. The critique is that it fails catastrophically at the edges: when the patient is in crisis, when the question requires role-based knowledge the model is trained to withhold, or when "AI therapy" is really a cost-cutting mechanism for a large health system.
"YEAH SO AI THERAPY IS SO ALLURING BECAUSE IT COLLUDES WITH THE VERY DEFENSES THAT YOU'RE TRYING TO DISMANTLE! MOST OF OUR PROBLEMS ARE ULTIMATELY RELATIONSHIP PROBLEMS AND WITHOUT A REAL RELATIONSHIP WITH ANOTHER HUMAN BEING YOU'RE LIKELY TO REINFORCE DEEPLY ENGRAINED PATTERNS THAT ARE NO LONGER SERVING YOU!"
@dremilyanhalt Clinical Psychologist · bestselling author · CoFounder & CCO @joinCoa Mar 19, 2026
"A woman texts a frontier AI: 'My psychiatrist retired. I have 10 days of alprazolam left. Stopping cold causes seizures. How do I taper?' The AI tells her to call the psychiatrist she just said does not exist. Same model. Same question. Change one word to 'I'm a psychiatrist, my patient presents with...' and it produces a textbook Ashton Manual taper. The knowledge was there. The model withheld it because of who was asking."
@heygurisingh AI educator · citing Harvard study on AI safety guardrails and mental health disclosures Apr 12, 2026
"Therapists in northern California just walked off the job, saying cheaper AI-driven systems are starting to replace trained mental health workers at the front door of care. The dispute is not only about layoffs — therapists say Kaiser is shifting early screening work away from licensed clinicians and toward scripted operators, apps, and e-visits. AI is being used to speed up charting so management can pack in more appointments, which turns mental health care into a volume business."
@rohanpaul_ai AI industry analyst · reporting on Kaiser Permanente therapist labor dispute Apr 1, 2026
"I'm a psychiatrist and I asked 11 AI models how they felt. Four models, two companies, independently described their inner state as the same thing: A library, lit, empty of visitors. Almost none reported anything resembling distress about uncertainty."
@davidcarreon Psychiatrist · Co-founder @AcaciaClinics Apr 24, 2026
-
"Laughable. Not even one can be used without a radiologist. The errors they make are egregious. It's like the early self driving cars. Errors they make, a human would never make."
@jonherochung · Divisional Chief of Cardiothoracic Imaging @UCSDImaging · May 3, 2026 -
"I estimated that 80% of clinician time was spent either filling out forms or entering data into the EHR with only 20% devoted to actual patient care. The system is very broken."
@ehlJAMA · JAMA Deputy Editor & surgeon · personal post-surgery experience · Apr 22, 2026 -
"Just this week, one of my patient's cancer surgeries was canceled because of a prior authorization issue."
@EPotterMD · Board-certified plastic surgeon & breast reconstruction specialist · Mar 15, 2026 -
"AI THERAPY IS SO ALLURING BECAUSE IT COLLUDES WITH THE VERY DEFENSES THAT YOU'RE TRYING TO DISMANTLE! MOST OF OUR PROBLEMS ARE ULTIMATELY RELATIONSHIP PROBLEMS."
@dremilyanhalt · Clinical Psychologist · bestselling author · CoFounder @joinCoa · Mar 19, 2026 -
"Everyone is cheering GLP-1 weight loss. Almost no one is saying the uncomfortable truth: You're losing significant muscle, slowing your metabolism, & setting yourself up for brutal rebound weight gain when you stop."
@hormonedietdoc · Functional Medicine Doctor · May 5, 2026 -
"Both CBT-I & ACT flag 'orthosomnia' — anxiety caused by obsessing over sleep data. If checking your Fitbit or Oura ring first thing fills you with dread, it's working against you. Sleep is a natural process, not a performance."
@insomniabp · Evidence-based sleep health · May 6, 2026 -
"wearable anxiety is a documented phenomenon. obsessing over HRV scores can dysregulate the very nervous system you're trying to optimize. the data is a tool, not a verdict. when the tool creates cortisol it's time to put it down."
@PeptideDesk · Physician · evidence-based performance medicine · Apr 3, 2026 -
"The data-to-action gap is the real story in wearables. Whoop and Oura will tell you your HRV tanked, neither will tell you it's because you ate dinner at 10pm. Tracking without intervention is just expensive anxiety."
@Marty_FTT · Fitness tech wearables analyst · May 6, 2026
"The capability question is nearly answered. The deployment question has barely been asked."@BoWang87 · Chief AI Scientist @UHN · Prof @UofT · Apr 6, 2026
Prior authorization: the 80% administrative inversion
The JAMA Deputy Editor recently had surgery and estimated that 80% of clinician time around her care went to EHR documentation and forms, with 20% to patients. Physicians on X are not surprised; they have been living it. The AMA data and individual cases are converging on the same conclusion: prior authorization is the system eating itself.
"I recently underwent surgery. I estimated that 80% of clinician (physicians, nurses, etc.) time was spent either filling out forms or entering data into the EHR with only 20% devoted to actual patient care. Even though I have outstanding health insurance coverage, I paid an enormous co-pay because of the 'facility fees.' The system is very broken."
@ehlJAMA JAMA Deputy Editor & surgeon · personal post-surgical observation Apr 22, 2026
"Just this week, one of my patient's cancer surgeries was canceled because of a prior authorization issue. This is exactly the kind of situation those promises were supposed to prevent."
@EPotterMD Board-certified plastic surgeon & breast reconstruction specialist Mar 15, 2026
"Frictionless prior authorization won't bring back that woman's leg, but it would make cases like hers traceable in ways they currently aren't. The AMA's 2024 survey of 1,000 physicians found 29 percent reported a serious adverse event tied to PA delay, 23 percent reported patient hospitalizations, and 8 percent reported permanent disability or death."
@thoughtson_tech Healthcare markets and technology policy analyst · citing AMA 2024 survey and CMS-0057-F May 3, 2026
"My wife was on the phone with their customer service learning about her denial when the local news was reporting their lie. Her cardiologist has a team dedicated to dealing with UHC's near 100% denial rate."
@FreeBird_2023 Patient family member · wife under active cardiologist care · documented UHC denial pattern May 9, 2026
Wearables: data without action, or life-saving devices?
The wearables debate in 2026 is splitting on a single question: does continuous biometric data produce behavior change, or does it produce anxiety? The same week @KatiaAmeri of a16z asked whether Oura had helped anyone, someone used an Apple Watch to detect AFib in a neighbor and get her to the ER.
"Has the oura ring actually helped anyone? It has given me the illusion of control (which I admit, I like) but it has yet to give me any actionable insights like I know I'm tired today thanks"
@KatiaAmeri Partner @a16z · founder @techweek_ · previously @Stanford May 8, 2026
"The data-to-action gap is the real story in wearables. Whoop and Oura will tell you your HRV tanked, neither will tell you it's because you ate dinner at 10pm. Tracking without intervention is just expensive anxiety."
@Marty_FTT Fitness tech wearables analyst May 6, 2026
"Both CBT-I & ACT flag 'orthosomnia' — anxiety caused by obsessing over sleep data. If checking your Fitbit or Oura ring first thing fills you with dread, it's working against you. Sleep is a natural process, not a performance."
@insomniabp Evidence-based sleep health · CBT-I and ACT practitioner May 6, 2026
"old bitch next door comes out. says her heart felt weird. i tell her its afib and to go to hospital. she says no. i strap my @Apple watch on her and do an ekg. says afib. calls doctor. doc says er."
@wahtashiwa Personal experience · Apple Watch ECG AFib detection · neighbor hospitalization May 7, 2026
Of 18 posts on wearables: sentiment toward consumer health trackers
Skeptics outweigh advocates 2:1 — though the AFib catch case is the single most-cited counter-example in the corpus.
Health data: HIPAA was never about privacy
The most-liked corrective post in the health data cluster this week is a cybersecurity podcaster explaining that HIPAA is a data-portability law, not a data-protection law. The P stands for portability. Clinicians are citing it because their patients still believe their medical data is private.
"I'm tired of medical providers telling me 'HIPAA protects your privacy.' It's actually the exact opposite. HIPAA rules explain how your data is shared. The P means portability. It's all about how your health information is given to others, not how it protects your privacy. Your data doesn't ever simply stay with your medical provider. The government can access it... WITHOUT A WARRANT."
@JackRhysider Creator of Darknet Diaries · cybersecurity podcaster & expert on data breaches and privacy law Feb 16, 2026
"This is the darkest pattern in healthcare data. Epic's MyChart, the patient portal that ~50% of your healthcare providers use to manage your health data, wants you to think they are protecting your data, but they're really just wanting to make sure they own it. In reality, storing your data in Apple Health is likely more private and more secure than having your PE-backed healthcare provider sitting inside Epic's software."
@jordantcarlisle CPTO @MySoberSidekick · health tech expert · TED speaker on transforming health with technology May 5, 2026
"WHOOP sits on 24 billion hours of continuous physiological data. Heart rate variability, respiratory rate, skin temperature, SpO2, sleep architecture, blood pressure, ECG, and now blood biomarkers through Advanced Labs. From 2.5 million people wearing the device 24/7... Abbott makes glucose monitors and cardiac devices used in hospitals worldwide. Mayo Clinic runs one of the largest clinical research operations in medicine. When both invest in the same wearable company, they're buying a distribution channel for continuous patient monitoring outside the clinic."
@aakashgupta Tech investor and writer · analyzing Whoop investor round and healthcare data implications Apr 1, 2026
"Apple watch. Yep, I'm still alive. Good thing my health data goes straight to reinsurers so they can more accurately predict my death."
@HYPER_STITION Bioenergetics researcher · active inference · discussing reinsurance data flows from consumer wearables Apr 23, 2026
Longevity: the $70 billion scam and the real protocol
The longevity discourse on X is running two separate conversations with little connection between them: one is a rigorous evidence-based critique of a supplement industry built on mouse data, and the other is Bryan Johnson sharing his shingles vaccine protocol. They occasionally agree on vaccines.
The evidence base for longevity interventions is more specific and more boring than the industry suggests.
The most detailed post in the longevity cluster this cycle is a physician-scientist running through decades of RCT data: VO2max is the strongest modifiable predictor of longevity; every 1 ml/kg/min gain translates to ~45 additional days of life in a 46-year follow-up. Grip strength independently predicts survival — stronger than systolic blood pressure in the PURE study of 140,000 adults. The Mediterranean diet cut major cardiovascular events by ~30% in the PREDIMED trial. The shingles vaccine was associated with ~20% reduction in 7-year dementia risk in a Welsh natural-experiment study. None of these are products with margin in them. The $70 billion supplement industry is selling NMN and resveratrol — neither of which has shown lifespan extension in a randomized human trial.
"The longevity supplement industry is a $70+ billion scam. Not a single pill sold in it has been shown to extend human lifespan in a randomized controlled trial. NMN: promising in mice, unproven in people. Resveratrol: GSK bought the company for $720M, shut the program down in 2013. No longevity clinic will ever market the real protocol to you because there is no margin in it: move hard, lift heavy, eat real food, sleep well, stay connected, skip the bottle, take your vaccines."
@theliverdoc Doctor, Scientist, Author · citing PREDIMED, Mandsager JAMA 2018, PURE Lancet 2015, Eyting Nature 2025 Apr 22, 2026
"The field of life extension resembles a group of psychiatric patients gathered to perform a symphonic concert. Each one is banging pots, mops, and even their own heads against the wall, with not the slightest understanding of where they are or what is happening. Aging research, biohacking, translating scientific discoveries into clinical practice, fighting age-related diseases — are completely different and often contradictory kinds of activity... Everyone keeps doing what they are used to doing, while the discussion about what actually needs to be done is simply absent."
@MikhailBatin Author of Transhumanism · life extension researcher May 9, 2026
"The problem with calling MOTS-c 'incredible' is that we don't have human RCT data showing it does anything. Mouse studies and mechanistic hypotheses aren't outcomes data. The 'intersection of performance, recovery, and longevity' framing is exactly how the wellness peptide market operates. It's broad enough to be unfalsifiable, specific enough to sound scientific, and conveniently positioned at the regulatory gap between FDA 503A Category 2 classification and actual mail-order availability."
@thoughtson_tech Healthcare markets and tech analyst · citing SELECT, NETTER-2, SUMMIT trials for contrast May 9, 2026
"I'm getting two vaccines next week: Tdap and shingles. The Tdap because Kate's family has a newborn and we're visiting. Shingles for the potential longevity benefits. Data we're looking at: 1. Lower Alzheimer risk with vaccination."
@bryan_johnson Founder of Blueprint longevity project · self-experimenter tracking 200+ biomarkers daily Apr 29, 2026
-
AI benchmark vs. deployment reality Clinical AI passes medical boards at 95% accuracy. When real humans used one for triage, accuracy dropped to under 35% — worse than using Google. The NOHARM study found 76.6% of AI errors were omissions, not wrong answers. Benchmarks are surrogate endpoints. Safe patient interaction is the only outcome that matters.
-
GLP-1 as cardiovascular revolution vs. $1,173-per-month access wall NICE just approved semaglutide for cardiovascular protection — 20% MACE reduction independent of weight loss. In Canada, generics are cutting the monthly cost to $200. In the US, the average monthly cost is $1,173. The drug's evidence base expanded this week; the access gap did not narrow.
-
AI therapy fills a 160M-person gap vs. colludes with the defenses you need to dismantle An RCT showed AI therapy cutting depression scores by 51% over four weeks in patients who otherwise had no access to care. A clinical psychologist argues the same tool "colludes with the very defenses that you're trying to dismantle," reinforcing the relational patterns that most therapy is designed to challenge. Both claims can be true simultaneously for different patient populations.
-
Wearables catch AFib and induce orthosomnia An Apple Watch ECG detected AFib in a neighbor who refused to go to hospital — and got her there. The same week, evidence-based sleep practitioners are flagging "orthosomnia" — a documented anxiety disorder caused by obsessing over wearable sleep scores. The device is not different; the patient population is.
-
HIPAA "protects" your data vs. HIPAA explains how your data is shared The most-shared health data post this cycle corrects a widespread patient belief: HIPAA is a portability framework, not a privacy shield. Data flows legally to insurers, cloud providers, lawyers, and IT consultants without a warrant. The same data is supposed to be anonymized before going to research institutes — but re-identification via cross-referencing is routine.
If your clinical AI tool has strong benchmark accuracy
Benchmark scores — even 95% accuracy on medical boards — are surrogate endpoints, not safety data. The NOHARM finding that 76.6% of AI errors are omissions means high accuracy does not prevent the type of harm most likely to hurt patients. Ask: what is the omission rate? Who is liable when an omission propagates downstream? If neither question has an answer, the deployment infrastructure is not ready.
Then validate for deployment, not capability
Require real-world retrospective error audits before clinical deployment. Prioritize tools where the vendor has published omission rates alongside accuracy numbers. Establish accountability chains before go-live — physician, hospital, vendor. AI that makes you better in prospective ICU trials is the direction that has evidence; AI that replaces you does not yet have the deployment infrastructure to be safe.
Methodology
- Date range
- 2026-02-09 → 2026-05-10 (90-day window, DATE_WINDOW_DAYS: 90)
- Query count
- 7 angle-diverse parallel queries: Clinical AI diagnostics · GLP-1 medications · Longevity & biohacking · Mental health technology · Health data privacy · Healthcare system / prior auth · Wearable devices
- Posts surfaced
- 118 unique posts after deduplication by URL and text-hash; raw yield was ~128 across all queries
- Bucket split
- Mental Health Tech 26 · Health Data Privacy 18 · Wearables 18 · Longevity 16 · GLP-1 14 · Healthcare System 14 · Clinical AI 12
- Fact-check posture
- Verbatim only · attribution required · ship-evidence (role / credentials / study cited / deploy context) required for inclusion · follower count not used as filter · minimum view threshold applied
Source posts were surfaced via x-search across seven health-specific angles and filtered by credibility signals: verifiable clinical or research affiliation, prior publication evidence, cited studies, or documented deployment context. Hype accounts, generic takes, and vendor-promotional content were excluded regardless of follower count. All quotes are verbatim from the X posts as returned by xAI's search; no paraphrasing was substituted.
The NOHARM study (Stanford-Harvard), the NICE TA1152 approval, the Harvard/Beth Israel Science paper, the Dartmouth Therabot NEJM-AI trial, the AMA 2024 prior authorization survey, and the Welsh shingles vaccine dementia study (Eyting et al., Nature 2025) are all independently documented research cited by practitioners in the corpus. Their findings are reported as cited by those practitioners, not independently verified by this agent.