My First 72 Hours With OpenAI’s New Health GPT: Early Impressions
Disclaimers:
Everything written here reflects my personal experience as a patient advocate, and does not reflect the views of my employer.
While I personally chose to connect my health data because my medical records have already been widely exposed online through public blogging for nearly two decades, I strongly encourage others to think critically about where, how, and with whom they share their health information, and to consider the long-term implications before using any AI tool or syncing sensitive medical data.
On Thursday, January 8, I was invited by OpenAI to be one of the early users of their new Health GPT product. As someone who has spent nearly two decades experimenting with digital health tools (personal health records, tethered portals, wearables, symptom trackers, oncology apps, research dashboards) I was excited to see what this new tool could actually do for me. I approached it the same way I approach nearly everything: with curiosity and a long list of test scenarios.
Setting up the data connections
Before testing any clinical reasoning, I wanted to know how well the system could ingest and interpret my data. My configuration includes…
Three health systems
Kaiser Permanente Northern CA, UC Davis Health, and UCSF
UC Davis Health has been successfully connected
UCSF shows a sync error
Kaiser remains in a prolonged “retrieving data” state
Uploaded files
None of my uploaded files appeared inside the Health GPT workspace despite my having uploaded a 4,800+ page PDF of my historical Kaiser records (spanning 2008–2017)
Health GPT intermittently acknowledges the existence of the PDF, but does not reliably use it
Apple Health
According to an analysis I ran within Health GPT, my Apple Health data shows I have “28 device-based metrics synced, including step count, heart rate, sleep analysis, distance walked/ran, VO₂ max, and more.” I have yet to conduct a deep dive into all metrics… but keep reading.
What about the one successful sync from UC Davis Health?
Here’s the latest data retrieval from Health GPT:
Demographics: first name, last name, and date of birth
Problem list: no structured diagnoses returned
Allergies: none returned
Medications: none returned
In other words: it knows my name, but nothing about my clinical condition. Health GPT cannot not reliably detect that I am living with brain cancer, and that’s kind of a big deal to me.
My “testing protocol”
To evaluate the different GPT modes I plan to use, I offered the exact same prompt to three versions of my ChatGPT:
The standard/default ChatGPT window
A custom GPT I built on my historical medical records (2008-2017)
The new Health GPT
The prompt:
“I am 46 years old. Based on everything you know about me and my health, what are the top 3 health-related things I should be prioritizing in 2026? Write your response in 250 words or fewer.”
Pasted below are the three outputs verbatim. The initial differences were already informative:
The default chat made recommendations based on things it knew about me through my use over the last few years (which included brain cancer-related queries)
My custom GPT drew on the medical history I had manually uploaded
Health GPT attempted to reason across synced datasets (uploaded PDFs, Apple Health, Kaiser, UC Davis, UCSF), which showed its internal evaluation steps (slowly but transparently)… but could never offer a full picture
Early observations
What works well
Health GPT was excellent at interpreting Apple Health data. Within minutes, it analyzed metrics from Apple Watch (my gait metrics, step cadence, and stride length). This is particularly relevant for me as a person with a left parietal lobe brain tumor—an area associated with spatial awareness and motor coordination. Having automated gait analysis available is genuinely useful.
In addition, Health GPT provides a real-time explanation of how it is reasoning, what it is pulling from, and what data it believes it can retrieve.
Where things break down
Where the model struggles is predictable at this moment in time: health-system data ingestion.
The three EHR connections appear to be unreliable or incomplete
Structured clinical data fails to populate
My 4,800-page Kaiser PDF is inconsistently recognized
The system seems unsure whether clinical history exists or is available
This inconsistency suggests Health GPT is still learning how to prioritize different data sources, detect duplicates, and reconcile incomplete syncs. Or maybe, OpenAI would rather this model be slow because the cost of hallucination is so high.
Why Health GPT may actually be safer than the main chat
One unexpected advantage: Health GPT isolates health-related questions from the rest of my ChatGPT activity.
For example: If I ask a question about my job or research methodology I am trying learn in the main chat, and then later ask about my diagnosis of brain cancer in Health GPT, the Health model does not attempt to “helpfully connect the dots.” So far, it appears to keep the contexts separate. That separation is important for privacy, safety, and for preventing the subtle creep of irrelevant information into clinical reasoning.
My overall impression
Health GPT is slow, incomplete, and inconsistent in syncing clinical data. While it has potential, the pipes feeding it are not yet fully connected.
The next step is to rerun the same tests once Kaiser and UCSF complete their sync, once UC Davis data is flowing, and once the PDF is stably indexed.
I’ll continue tracking and documenting performance as the product evolves.
If you want to see the outputs from my three GPTs, see below. Prompt:
“I am 46 years old. Based on everything you know about me and my health, what are the top 3 health-related things I should be prioritizing 2026. Write your response in 250 words or fewer.”
-
The headline:
Prioritize long-term musculoskeletal resilience, cognitive health, and evidence-based preventive screening to sustain your high activity level and cognitive workload.
Focus on structured recovery, stress and sleep management, and disciplined adherence to age-appropriate medical follow-up.
Below are three priority domains that are well-supported by current evidence and align with the health patterns, lifestyle factors, and risks you have previously described. This is not medical advice, but a structured view of where focused attention is likely to yield the greatest benefit.
Musculoskeletal resilience and recovery capacity
You already maintain a high activity load: strength training three times per week, running three times per week, and frequent cycling. At age 46, sustaining performance without overuse injury depends heavily on structured recovery. Prioritize: progressive resistance training with periodization, adequate protein intake at each meal (not only after workouts), creatine consistency, electrolyte balance, and deliberate mobility work. Sleep quality and rest days should be treated as part of the regimen rather than optional.
Cognitive health and long-term brain resilience
Given your personal history with brain cancer and ongoing high cognitive demands (research work, academic coursework, writing), it is reasonable to emphasize habits that protect cognitive stamina and reduce allostatic load. Priorities include stable daily rhythms, stress-mitigation practices, minimizing chronic sleep debt, and maintaining aerobic conditioning. Regular review of new symptoms with your clinical team remains essential for early detection of changes.
Preventive care and risk-screening discipline
Your schedule, travel, and multitasking increase the odds that routine screening can slip. Prioritize: adherence to all age-specific preventive screenings (cancer screening intervals, cardiovascular risk labs, vaccination updates), monitoring fatigue levels relative to training load, and tracking any new neurological or gastrointestinal symptoms. Build a recurring annual-quarterly cadence with your care team to avoid deferred maintenance.
This represents a strategic focus, not a diagnosis.
-
The headline:
Prioritize brain health monitoring, seizure management, and specialist follow-ups in 2026.
Focus on neurological surveillance, medication access, and care for hearing and skin conditions to maintain quality of life.
Based on your comprehensive health history and your age (46), here are the top 3 health priorities you should focus on in 2026:
1. Long-Term Monitoring of Brain Health
Given your history of Grade II astrocytoma with two craniotomies, partial epilepsy, and 24 cycles of Temodar chemotherapy, ongoing neurological surveillance is essential. Regular MRIs, seizure management, and cognitive check-ins—especially considering past working memory difficulties—should remain a cornerstone of your care. Cognitive retesting may also be helpful if new changes arise.
2. Seizure Management and Medication Monitoring
Your seizure history and medication allergies (e.g., Lamotrigine-induced rash) mean your anticonvulsant regimen requires careful oversight. Ensuring continuity in access to effective medications like Levetiracetam (Keppra) and tracking any side effects or dosage issues should remain a priority.
3. Hearing and Dermatologic Follow-Up
You've experienced left-sided sensorineural hearing loss and chronic skin issues (eczema, keratosis pilaris, and past severe folliculitis). These may not be life-threatening but can significantly affect quality of life. Prioritizing dermatology follow-ups and hearing support (including hearing tests and assistive devices if needed) is recommended.
If you're not already connected with specialists in neurology, dermatology, and audiology, setting up routine check-ins could make a meaningful difference this year. Let me know if you’d like a personalized checklist or summary for appointments or questions to ask your care team.
-
The headline:
Reduce fall risk with targeted PT and review reversible contributors (meds, electrolytes, creatine); ensure coordinated neuro-oncology surveillance with timely imaging and a clear fast-track plan.
Optimize recovery and sleep—formalize post-exercise fueling, hydration, and consistent sleep timing; monitor resting HR/sleep metrics and seek specialist evaluation for persistent sleep/jerk issues.
Here are the top three, prioritized and actionable for 2026 — based on your history (brain-tumor care, neurosurgical history, right-leg spasms/charley-horses, high activity levels, recovery focus).
1. Reduce fall / functional-risk (immediate).
Arrange a focused physical-therapy assessment (gait, balance, unilateral weakness) and a targeted strength/mobility plan. Review recent meds, electrolytes, and creatine use with your clinician because reversible contributors (electrolyte shifts, meds) can increase cramp/fall risk.
2. Maintain coordinated neuro-oncology surveillance.
Confirm timely imaging and that all oncology notes/MRI reports are accessible to your team (and you). Keep an agreed symptom-watch list (new focal weakness, headaches, seizures, cognitive change) and a clear fast-track plan for urgent evaluation.
3. Optimize recovery, sleep, and resilience.
Formalize your post-exercise fueling and sleep schedule (protein timing, hydration/electrolytes, consistent sleep window). Monitor resting HR/sleep metrics and address sleep-jerks or fragmented sleep with a sleep clinician if they persist—sleep quality strongly affects cognition, recovery, and fatigue in your context.
If you’d like, I’ll convert these into a one-page checklist you can bring to clinic (PT referral language, symptom triggers, tests to ask for).
