Yes, uploading your labs to an AI tool can help, as long as you go in with your eyes open. Fishtown Medicine sees AI like ChatGPT as a strong first-draft assistant. It organizes years of scattered records, translates jargon into plain language, spots patterns across time, and helps you build sharper questions for your visit. What it cannot do is examine you, know your medications and symptoms, or take responsibility for a wrong answer, and it sometimes invents reference ranges that look real. General chatbots are also not covered by HIPAA, so remove your name, date of birth, and medical record number before you upload anything. The safe move is to use AI for the first draft, then bring it to Dr. Ash, a physician who reads all of it and signs his name to the interpretation.
Yes, you can upload your labs to ChatGPT or another AI tool, and it can help you a lot, as long as you understand where it helps and where it misleads. More and more patients tell me they ran their blood work through an AI before our first call, and I think that is a good instinct. A person who wants to understand their own numbers is already doing the hard part. The trick is knowing what these tools do well, where they quietly get things wrong, and how to protect your private health information while you use them.
This page is my honest read as a physician. I use these tools myself, I welcome the questions and uploads my patients bring, and I also see the specific ways they mislead people. Here is how to get the good without the harm.
Should you upload your labs to ChatGPT?
Yes, with your eyes open. Uploading a lab report to ChatGPT or a similar tool is one of the smartest habits I see in new patients, because it turns a confusing page of numbers into something you can actually think about. A good question is the start of good medicine, and AI is very good at helping you build one.
What it is not is a doctor. It cannot examine you, it does not know the medications you take or the symptoms you have been having, and it cannot take responsibility if it gets something wrong. So the right way to think about it is this: AI writes the first draft of understanding your labs, and a physician turns that draft into a plan that fits your actual body and life.
What does AI actually do well with your labs?
AI is a strong research assistant, and used well it does a handful of things very well.
- It organizes years of scattered records. If you have blood work from 3 different labs across 5 years sitting in old PDFs and patient portals, AI can pull it into one readable timeline. That kind of cleanup used to take me part of a visit. Now you can walk in with it already done.
- It translates the jargon. It explains what ApoB, fasting insulin, ferritin, or a slightly high MCV mean in plain language that respects your intelligence without burying you in Latin.
- It surfaces patterns over time. A single high value is easy to read on your own. A slow drift, like fasting glucose creeping up 2-3 points a year for 4 years, is the kind of pattern AI is good at pointing out so you can ask about it.
- It generates better questions for your visit. This is the part I value most. When you arrive with 4 specific questions instead of a vague worry, we skip the slow warm-up and get to the real work faster.
None of that replaces a doctor, but all of it makes the visit better.
Where does AI get your labs wrong?
This is the part the excitement skips over, and it matters. In my practice I see 4 recurring mistakes.
It hallucinates reference ranges. AI tools sometimes state a "normal range" with total confidence when the number is simply made up or pulled from the wrong lab. Reference ranges vary between labs and between the methods they use, and when you paste values without the report, a chatbot often has no reliable way to know which lab ran your test. A confident-sounding range is not the same as the correct one.
It is missing your clinical context. The AI does not know that your slightly low sodium is from a medication you started last month, or that your high creatinine follows a hard workout and a day of not drinking enough water. It reads the number without the story, and in medicine the story is often the whole answer.
It over-alarms and under-alarms in the wrong places. AI tends to sound the fire alarm on a mildly out-of-range incidental value that means very little on its own, which scares people badly. At the same time it can miss a subtle pattern across several results that a trained eye would catch, like a quiet trend toward insulin resistance while every single value still reads "normal." It gets the drama and the danger backward.
No one is accountable for the answer. If AI tells you something wrong, there is no physician who owns that mistake, no license behind it, and no one who follows up when the plan does not work. That accountability is a real part of what you are paying a doctor for.
Is it safe? What about privacy and HIPAA?
Here is the plain truth about privacy. General chatbots like ChatGPT, Gemini, or Perplexity are not covered by HIPAA the way your doctor's office is. HIPAA governs "covered entities" such as clinicians, hospitals, and health plans, and a consumer AI tool you signed up for on your own is not one of them. That does not mean these companies are careless, but it does mean your lab data does not get the legal protection you might assume.
So protect yourself before you upload. A few simple habits cover most of the risk.
- Strip your identifiers first. Remove your name, date of birth, address, and medical record number from the file or the text. A lab result reads the same to the AI whether or not your name is on it.
- Turn on the tool's privacy mode. Most major tools now let you turn off model training on your chats, or offer a temporary or private chat that is not used to improve the model. Turn that on before you paste anything medical.
- Assume anything uploaded may be kept. Even with training turned off, data can be retained for a period for safety and abuse review. Treat every upload as something that could persist somewhere, and do not upload what you would not want stored.
None of this should scare you away from the tool. It should just make you a careful user of it. De-identify, use privacy mode, and upload only what you need to get your question answered.
Reference range vs optimal range: what AI usually misses
This is the single most useful distinction I can give you, and most AI answers blur it.
A reference range is the band your lab prints next to your result, usually built from the middle 95% of a reference population. It tells you where most people fall. It does not tell you where you are healthiest. Plenty of common values sit inside the reference range and still carry risk, because the "normal" population that range came from can include people who are not actually well, depending on how the lab built it.
An optimal range is where the research suggests your risk is lowest, which is often narrower and tighter than the lab's reference band. Fasting glucose is a clean example. A value can read as "normal" on the lab report and still sit in a zone where I would want to watch it and act early. AI will usually tell you a value is "within normal limits" and stop there, because it is reading the printed reference range. The more useful question, and the one a physician works with you on, is not "is this normal" but "is this optimal for you, given your goals and your risk."
How do you use AI as a first draft, then bring it to a doctor?
Think of AI as your prep work, not your verdict. The workflow that gets my patients the most value looks like this.
- Gather and de-identify. Pull your lab PDFs together, remove your personal identifiers, and turn on the tool's private or no-training mode.
- Ask it to organize, not to diagnose. Good prompts are things like "build me a timeline of these values and flag anything that changed a lot," or "explain what each of these tests measures in plain language," or "give me 5 questions I should ask my doctor about these results."
- Bring the whole draft to your visit. Bring the summary, the questions, and your uncertainty. Say "the AI flagged this, is it real?" That is a great use of our time together.
- Let a physician sign the interpretation. A person with a license reads your results against your medications, your symptoms, your history, and your goals, and then owns the plan and the follow-up.
I do this gladly. When a patient shows up with an AI-organized history and a printed list of questions, we do better work in less time. Fishtown Medicine is a mobile-first direct primary care practice, so I see patients on home visits across Greater Philadelphia and stay reachable by secure message, phone, and video, and I will read every page you send before we talk. The AI writes the first draft, and my name goes on the final read.
Key Takeaways
- Yes, upload your labs to AI, knowing its limits. It organizes scattered records, translates jargon, spots patterns over time, and helps you build sharper questions for your visit.
- Know what it gets wrong. AI hallucinates reference ranges, misses your medications and symptoms, over-alarms on incidental values while under-alarming on subtle patterns, and holds no accountability for a wrong answer.
- Protect your privacy first. General chatbots are not covered by HIPAA, so strip your name, date of birth, address, and medical record number, turn on the tool's private or no-training mode, and assume anything uploaded may be retained.
- Reference range is not optimal range. A value can read "normal" from the printed band and still carry risk worth acting on early, and AI usually stops at "within normal limits."
- Use AI for the first draft, then bring it to a physician. Dr. Ash reads every page you send, confirms what is real, and signs his name to the interpretation and the plan.
Related Services and Reading
- How to Use AI to Find the Right Doctor - what AI does well when you are looking for care, and where it runs out of road
- Digital Health Literacy - the full library on reading health information in 2026 without getting fooled
- At-Home Health Monitoring - how to turn Oura, Whoop, and Apple Watch data into better care
- Advanced Lipid Testing in Philadelphia - what ApoB and particle testing add that a standard cholesterol panel misses
Fishtown Medicine | Digital health literacy
Frequently Asked Questions
Common Questions
Deep-Dive Questions
Ready when you are
Dr. Ash reads every intake himself, and answers questions personally - usually within a few hours.
