Your doctor sends over a lab report with 20 values, half flagged in red, none of them explained. Fifteen minutes later you've opened Claude or Perplexity and you're getting a clearer explanation than you'd get in a five-minute follow-up appointment. That's genuinely useful. But most people don't stop to ask: what happens to that health data after you hit send?
The HIPAA Gap
HIPAA - the U.S. law governing how medical providers handle patient information - doesn't cover consumer AI tools. Anthropic, OpenAI, and Perplexity are technology companies, not healthcare providers or their business associates. That means the legal protections that apply when your doctor's office stores your records don't automatically apply when you share the same information with an AI chatbot.
Anthropic's privacy policy for Claude states the company may use conversation data to improve its models, though subscribers on paid plans can opt out of having their data used for training. Perplexity has similar terms. In plain terms: the blood glucose reading, the thyroid marker, and the questions you're asking about what they mean could be stored - and depending on your settings, used to make the AI smarter for future users.
This isn't unique to health data, but health data carries more risk than most. It can reveal conditions you haven't disclosed to an employer. It can affect insurance. And once shared, you have limited ability to take it back.
What You Actually Need to Share
You usually don't need to upload the full document to get useful information. There's a meaningful difference between asking "what does an ALT level of 52 U/L mean?" - a general question that doesn't identify you - versus uploading a PDF containing your name, date of birth, and clinic information alongside the same values.
You get most of the explanatory value by describing your results in plain terms. The AI doesn't need to know it's your lab report, specifically, to explain what elevated ferritin means or what range is considered normal for TSH. If you're using an employer-provided AI tool, be particularly careful - depending on how the company licenses the product, administrators may have visibility into usage data.
The Practical Middle Ground
AI tools can be genuinely useful for health literacy - helping non-specialists understand medical terminology, what follow-up questions to ask, or whether a result is borderline versus clearly outside normal range. Claude handles these explanations well and is careful to flag when something requires professional evaluation rather than just returning alarming information without context. Perplexity tends to cite specific medical sources, which helps you find authoritative explanations.
The practical approach: use AI to understand vocabulary and context. Describe your values in general terms rather than uploading the full report. Use paid tiers where you have more control over data retention settings - free tiers tend to have the most permissive data use terms. And treat AI health explanations as preparation for a medical conversation, not a replacement for one.