Path 1
Manual PDF upload
Download the Function PDF, make a redacted working copy, upload it into ChatGPT, and add your real-world context before asking for analysis. This gives you control over exactly what you share.
Back to Healthspan
Export / ChatGPT field notes
Export / ChatGPT field notesI exported my Function Health data. ChatGPT helped turn it into better questions.
Function Health gave me the raw signal: a broad lab panel, biomarker trends, and clinician-reviewed notes. ChatGPT became useful when I added the missing context: age, training history, supplement routine, fitness level, goals, schedule, diet, sleep, and what I was actually willing to do consistently. The result was not medical advice. It was a translation layer that helped me see priorities, ask better questions, and build a plan that fit my life.
Field workflow
01Export PDF
02Add context
03Ask ChatGPT for patterns
04Discuss and retest
The more useful prompt was not "what is wrong with me?" It was "what should I ask next, what might be noise, and what deserves a professional review?"
Boundary
ChatGPT can organize health data. It should not become your doctor.
This page is personal experience and educational workflow. Use exported results to prepare better questions and cleaner follow-up, not to diagnose yourself, change medication, or avoid qualified medical care.
- Verify extracted values against the original PDF.
- Remove unnecessary identifiers before pasting sensitive data.
- Bring meaningful findings to a qualified clinician.
Two clean paths
Use the official PDF path or the Health connection. Skip the hacks.
With health data, trust matters. I would not send readers toward unofficial CSV scrapers or browser-token workarounds. The cleaner public recommendation is the Function PDF or the official ChatGPT Health connection where it is available.
Open Function DocumentsPrivacy posture
Treat the file like a medical record, not a gym note.
Health chats, files, and memories are separate from regular ChatGPT and are not used to train OpenAI foundation models.
For regular ChatGPT chats outside Health, review Data Controls, opt-out settings, and Temporary Chat behavior before uploading sensitive health data.
Third-party apps in Health are off by default unless you explicitly enable them, and connected apps can be disconnected later in settings.
Step-by-step
The desktop export workflow before ChatGPT gets involved.
The translation loop
The export only got useful after I added context.
Lab values alone can make ChatGPT sound more certain than it should. The better workflow was values plus training, sleep, food, history, goals, and a hard instruction to produce questions and levers instead of diagnosis.
Export
PDF report of record
marker
value
unit
range flag
ChatGPT lens
Organize, caveat, question
"Group this into reassuring signals, watch items, and clinician questions. Do not diagnose me."
omega-3
Action lane, not diagnosis.
fiber
Action lane, not diagnosis.
zone work
Action lane, not diagnosis.
retest
Action lane, not diagnosis.
Prompt ladder
Ask ChatGPT to slow the dashboard down.
The better prompts did not try to make ChatGPT the authority. They made it a sorting layer: extract the data, group the signal, name uncertainty, and create doctor questions.
Context I added
Training
Lifting, cardio volume, Zone 2/Zone 4 split, recent intensity
Recovery
Sleep trend, HRV/readiness, stress, illness, alcohol
Nutrition
Protein, fish intake, fiber, supplements, meal consistency
History
Family risk, medications, prior labs, clinician concerns
Goal
Prevention, energy, performance, retesting, doctor questions
Reusable prompt
The prompt mattered more than the upload.
The upload alone was not the unlock. The unlock was telling ChatGPT who the data belonged to, what constraints mattered, and where professional review still owned the decision.
Prompt to reuse
You are helping me understand my personal health data for education and planning. You are not diagnosing me or replacing a clinician. I uploaded my Function Health lab-results PDF. Use the actual values, units, reference ranges, and dates in the file. My context: - Age: - Sex: - Height / weight / body composition, if relevant: - Training: - Cardio: - Sleep: - Diet: - Alcohol: - Current supplements: - Current medications: - Main goals: - Constraints: - Family history or known diagnoses I want considered: Please produce: 1. A plain-English summary of the biggest patterns. 2. Any results I should discuss with a doctor. 3. Any results that may be affected by training, fasting, hydration, illness, timing, or supplements. 4. Lifestyle levers ranked by likely impact and ease. 5. Supplement-related considerations, without prescribing. 6. A realistic 8-12 week plan. 7. Questions to bring to my clinician. 8. What I should avoid changing until I get professional input.
Prioritization
The useful output was hierarchy, not a prescription.
This was where the workflow started to feel different. It was less like scrolling through lab values and more like walking into an appointment with the chart already organized.
The loop I would run again
01
Export the latest Function Health PDF.
02
Save the untouched original.
03
Create a redacted AI copy.
04
Upload or connect the data.
05
Add age, goals, training, diet, sleep, supplements, medications, and constraints.
06
Ask for pattern recognition, not diagnosis.
07
Ask for questions to bring to a clinician.
08
Ask for an 8-12 week plan and what not to change yet.
09
Retest later and compare against the same baseline.
My Function signal board
This is where ChatGPT helped: not panic, not gloss.
My standard lipid and metabolic markers looked mostly reassuring. The deeper picture had low omega-3 status and LDL particle caveats. ChatGPT helped hold both truths at the same time.
See the full review and narrativeMarkerResultChatGPT job
LDL-C98 mg/dLreassuring baseline
HDL-C67 mg/dLstrong
Triglycerides63 mg/dLlow
ApoB81 mg/dLin range
Lp(a)23 nmol/Lin range
hs-CRP0.2 mg/Llow inflammation
LDL-P1279 nmol/Lwatch item
Small LDL263 nmol/Lwatch item
OmegaCheck3.2%low signal
HbA1c5.2%reassuring
Insulin3.9 uIU/mLreassuring
Glucose95 mg/dLcontext marker
Action plan
The output became four action lanes I could actually test.
Guardrails
The workflow works only if you keep the boundary visible.
AI can make a pile of lab values feel more legible. That is powerful and risky. The safest use is preparation: better tables, better questions, better follow-up.
Do not ask ChatGPT to diagnose you
Ask for organization, questions, patterns, and uncertainty. Diagnosis belongs with qualified medical professionals.
Do not paste more identity than needed
If you can work from values, units, flags, age range, sex, and relevant context, avoid pasting unnecessary identifiers.
Do not change medication from a chat
Medication, treatment, and disease-management decisions need your clinician, your history, and professional responsibility.
Do not trust one clean table as the whole truth
Extraction can be wrong. Check values, units, ranges, dates, and anything that would change a decision.
Where to go next
Keep the loop moving.
Sources
Source-backed boundaries.
Function owns the account interface, so exact labels can change. This page separates PDF-sharing guidance, practical export route notes, and ChatGPT privacy/medical-boundary notes.
Source
Function: share results with your doctor
Function says ready results can be downloaded as a PDF to share with a healthcare provider.
Source
Function: how it works
Function describes reviewed results, ongoing tracking, and the ability to download and share results.
Source
Empirical Health export guide
Third-party guide documenting the practical documents-route PDF download path for Function Health results.
Source
OpenAI: ChatGPT Health
OpenAI describes Health as a separate space for health and wellness questions, with Function listed as a supported launch app.
Source
OpenAI: data controls and Temporary Chat
OpenAI documents model-training controls, Temporary Chat behavior, and deletion timing for normal ChatGPT use outside Health.
Source
OpenAI: privacy and sensitive information
OpenAI notes that users should avoid sharing sensitive information they would not want used or reviewed.
Source
Function app in ChatGPT announcement
Function announced that members can authorize ChatGPT to access a limited, high-level summary of lab results and revoke access later.
FAQ
Function Health export and ChatGPT questions, answered directly.
Can I export Function Health results?+
Yes. Function says ready results can be downloaded as a PDF and shared with a healthcare provider. The safest public guidance is PDF unless your account clearly exposes another format.
Can I upload Function Health results to ChatGPT?+
You can use ChatGPT to organize health information, but treat exported labs as sensitive personal data. Use privacy settings carefully, remove unnecessary identifiers when possible, and do not rely on ChatGPT for diagnosis or treatment decisions.
Can ChatGPT Health connect to Function?+
OpenAI lists Function as a launch third-party wellness app in ChatGPT Health. Availability depends on your account, region, and rollout status. If it appears, connect it inside Health and re-authorize Function for Health specifically.
What should I ask ChatGPT to do with bloodwork?+
Ask it to extract values, group reassuring and watch-item signals, list uncertainty, draft clinician questions, and suggest conservative lifestyle levers to discuss. Do not ask it to prescribe, diagnose, or override medical guidance.
What did ChatGPT help with in this Function Health review?+
It helped organize the pattern: reassuring metabolic and inflammation markers, low omega-3 status, imperfect LDL particle markers, and a practical loop of food, supplements, cardio structure, retesting, and clinician review.
Should I use CSV instead of PDF?+
Use CSV only if Function exposes it directly in your account. The source-backed export path I can verify publicly is the PDF report.
Is this medical advice?+
No. This is personal field-notes content and an educational workflow. Lab interpretation, diagnosis, treatment, and medication decisions belong with qualified healthcare professionals.