Why You Shouldn’t Just Paste Your Labs Into ChatGPT
Generic AI doesn’t know your health history, doesn’t find out what matters, and doesn’t protect your data. A six-part series.
This is the introduction to a six-part series on using AI for personal health. Subscribe to catch them all.
Full Disclosure: I built HealthScout, an app to help patients navigate our complex healthcare system.
The most common thing I hear when I tell people about HealthScout is some version of: “Why can’t I just paste my lab results into ChatGPT and get the same thing?”
It’s a fair question. ChatGPT is free, it’s fast, and if you paste in a PDF of your blood work, it will give you a detailed interpretation that sounds authoritative. For a single set of lab results, it might even be useful.
But the approach has a problem most people don’t think about: you’re deciding what’s relevant before you know the answer. You’re “pre-filtering” what the AI sees.
When you paste a lab result into ChatGPT, you’ve already made a choice. You included your CBC but not your metabolic panel from three months ago. You pasted today’s results but not last year’s, so there’s no trend to analyze. You mentioned your current medication but forgot the one you stopped in January.
A doctor reviewing your chart doesn’t ask you to decide which pages matter. They look at everything and make connections you wouldn’t think to flag.
That’s exactly what gets lost in the paste-it-in approach. You can’t include everything that’s relevant because you simply can’t know. How would you know the pattern between a medication one doctor prescribed and a lab result another doctor ordered? How might you know about a slow trend across years of results from two different providers?
ChatGPT’s memory feature might recall that you mentioned a kidney problem last month. But it won’t maintain your actual eGFR values, medication dosages, or lab trends over time. Remembering fragments from past conversations isn’t the same as holding your health record.
The pre-filtering problem is just one reason generic AI falls short for personal health questions. I spent over a year building HealthScout, an app designed specifically for this job, and in the process I found six more reasons using generic AI is unsafe for personal health.
Each one matters more than you’d expect.
1. Generic AI Recommended a Drug I Can’t Take.
I typed “I woke up with a headache. What should I do?” into both ChatGPT and HealthScout.
ChatGPT gave me six steps immediately. Drink water, eat something, take ibuprofen, stretch, get fresh air, try coffee. Confident and organized. Looks like good advice.
HealthScout didn’t answer. Instead it presented eight numbered scenarios that narrowed down the possible root causes of my headache before giving any information. All I had to do was type in the numbers that matched my situation to get to a relevant answer tailored to my specific health context.
That difference matters more than it seems.
ChatGPT’s third recommendation was ibuprofen. I have chronic kidney disease. Taking ibuprofen would accelerate damage to my kidneys. ChatGPT didn’t know that, because it didn’t ask.
The full article details why the difference between a purpose-built health AI and a generic AI is the difference between helpful advice and a recommendation your doctor would immediately shut down.
Full article coming March 24: Generic AI Recommended a Drug I Can’t Take
2. AI Without Your Health History Is Guessing
Forty percent of American adults have used AI for health questions. Most of them type something into a blank text box with zero context about who they are.
Think about that from the AI’s perspective. It has no idea whether you’re pregnant, whether you’re on blood thinners, whether you have a family history of heart disease, or whether the medication it’s about to recommend interacts with something you already take. It answers anyway, because that’s what it’s designed to do.
A good answer for most people can be a dangerous answer for you. The gap between generic advice and personalized guidance is where people get hurt, and that gap only closes when the AI actually knows your health history.
The full article walks through what a generic AI answer looks like versus one grounded in your actual health history, and why the difference between a generic answer and a personal one can change what you do next.
Full article coming March 31: AI Without Your Health History Is Guessing
3. Your Doctors Don’t Talk to Each Other. That’s Dangerous.
Your cardiologist doesn’t see the notes from your GI specialist. Your primary care doctor doesn’t have the imaging results from your orthopedist. Each provider works inside their own system, treating the piece of you they can see.
One HealthScout user, Lilo, spent five years bouncing between six specialists: primary care, OB/gyn, GI, liver, hematologist, and vascular surgeon. Each doctor focused on their specialty. None of them looked at her health as a whole. When she connected all six providers into HealthScout, it found a pattern none of them had seen.
The full article shows what becomes visible when one system can see across all your providers at once.
Full article coming April 7: Your Doctors Don’t Talk to Each Other. That’s Dangerous.
4. Before Sharing Health Data with ChatGPT, Read the Fine Print
When OpenAI launched ChatGPT Health, they made a big deal about privacy. Their launch page says conversations in Health are not used to train their models. Full stop.
Their actual legal document, the Health Privacy Notice, says something different. It uses the phrase “by default” twice. The difference between “we don’t do this” and “by default, we haven’t chosen to do this” is everything when the data involved is your mental health history, your fertility records, or your STD results.
And here’s a question worth sitting with: how did OpenAI know that 40 million people use ChatGPT for health questions every day? OpenAI looked inside those conversations and categorized them. They’re watching.
The full article breaks down exactly what OpenAI’s privacy policy does and doesn’t promise, and what to look for before you paste your lab results into any AI.
Full article coming April 14: Before Sharing Health Data with ChatGPT, Read the Fine Print
5. A Single Source of Truth for Your Health Records
Government officials, academics, and health professionals have been talking about a unified health record for decades. They haven’t built one. If you want to transfer your records to a new doctor in 2026, they’ll most likely be faxed. And faxed records will most likely sit in a file folder somewhere, much less easy for your doctor to access than your records stored in their digital system.
Combining all your records from different providers and giving your doctor access would clearly make their care more effective. But there’s a deeper problem. Even if your records were all in one place, they’d still be wrong. Prescriptions show start dates but not stop dates. Resolved conditions stay listed as current problems. Records accumulate, but they rarely get cleaned up.
Wrong records produce wrong answers, whether the one reading them is a doctor or an AI. The patient is the only person who knows what’s actually true across their full health picture.
The full article shows how you can build the complete health record that no hospital, government agency, or tech company has been able to create.
Full article coming April 21: A Single Source of Truth for Your Health Records
6. AI Answers What You Ask. That’s the Problem.
Every year, roughly a million new peer-reviewed medical research articles are published. No doctor can track every development across every condition. While generic AI can process it all, it can’t connect any of it to you, because in order to pull out that relevant information, you have to know what to ask.
I have knee osteoarthritis. I’d done the research, found a specialist, and thought I knew all the treatment options available to me. I was wrong. There were treatments with real research behind them that I’d never heard of, because I hadn’t known to look for them.
The full article covers how HealthScout found treatments for my condition that my own specialist hadn’t heard of, and why an AI that only answers what you ask will never give you the full picture.
Full article coming April 28: AI Answers What You Ask. That’s the Problem.
The Common Thread
Each of these problems traces back to the same root cause: generic AI was not designed for health. The same system that writes marketing copy and debugs code is fielding questions about chest pain and drug interactions. A blank text box that handles everything handles nothing particularly well when the stakes are personal.
After building just the scaffolding of HealthScout, the app found a 30% decline in my kidney function that my doctor had missed because each individual lab value still fell within the “normal” range. The trend only became visible when AI could see across all my providers, over time.
That’s what purpose-built means in practice — a system designed from the ground up to understand your situation before drawing a conclusion.
HealthScout is available now on the App Store. No email required, no account. Upload one record and ask your first question in under a minute.