A growing share of people now ask AI chatbots about symptoms, lab results, and medication questions before they ever call a doctor. That’s not going away, so the useful question isn’t “should you do this,” it’s “how do you do it without getting hurt.”
What AI Is Actually Good At
Large language models are pattern-matching machines trained on enormous amounts of text. Applied to your own health data, that’s genuinely useful for a few narrow things:
- Translating jargon. “Elevated ALT and AST” becomes “two liver enzymes are a bit high, here’s what usually causes that.”
- Spotting a trend across time. Paste three years of the same blood panel and ask what moved. Humans are bad at holding that much numeric detail in their head. AI isn’t.
- Drafting the questions you should ask your doctor. Not answers, questions. “What’s driving my fasting glucose creeping up over two years” is a much better use of a 12-minute appointment than starting from zero.
That’s the actual use case, and it’s a real one. It is not the same as diagnosis.
A Workflow That Works
- Upload the actual document (a PDF of your lab panel, not a screenshot of a screenshot) so nothing gets lost in translation.
- Ask specific questions, not vague ones. “What does an LDL of 142 mean at my age” beats “is my cholesterol bad.”
- Ask it to flag anything that changed meaningfully since your last panel, not just what’s out of range today. Trend matters more than a single number.
- Bring what you learned into your next doctor’s visit as a starting point, not a conclusion.
Where This Goes Wrong
This is the part most “AI for health” content skips, and it’s the part that actually matters.
AI can be confidently, fluently wrong. Language models don’t know what they don’t know. When they lack a clear answer, they don’t reliably say “I’m not sure,” they generate the most statistically plausible-sounding response, and a wrong answer can read exactly as confident as a right one. There’s no built-in tell.
Reference ranges aren’t universal. “Normal” for a lab value depends on the lab, the assay method, your age, sex, and sometimes even the time of day you were tested. A general-purpose AI tool often doesn’t know which reference range your specific result was measured against, and may quietly apply the wrong one.
It doesn’t have your full picture. A real doctor factors in your medical history, medications, family history, and what they see and hear in the room. AI only knows what you type into the box, and it has no way to catch the context you didn’t think to mention.
Nobody is accountable if it’s wrong. A doctor who gives you bad advice carries real professional and legal accountability, which is part of what keeps medical advice careful. An AI tool carries none of that. That asymmetry alone is a reason to treat its output as a first draft, not a verdict.
It is not an emergency line. Chest pain, sudden numbness, trouble breathing, a fever in an infant, that’s a call to a doctor or 911, not a prompt. No AI response time competes with an actual emergency room.
A 3-Question Trust Check
Before you act on anything an AI tells you about your health, ask:
- Can I verify this against a real source, my doctor, a lab reference sheet, a primary study?
- Would I make a different decision today because of this answer?
- Am I using this to have a better conversation with a clinician, or to skip having one?
If the honest answer to #3 is “skip one,” stop and make the call instead.
Where This Fits With What We’re Building
This is exactly the narrow lane Aliya, the AI inside SelfHealth, is built for: translating your own uploaded blood panel into plain English, nothing more. It’s not a diagnostic tool, and we don’t want you treating it like one. The caution in this article applies to Aliya just as much as it applies to ChatGPT. Read the number. Ask better questions. Still go to the doctor.