The appeal is obvious. You have a lab report written for a doctor, not for you, full of abbreviations and reference ranges. You would like to ask a plain question - what does this actually mean, is it different from last year, should I ask someone about it - and get a plain answer.
If you live in the United States, 2026 is the year that became an ordinary feature rather than a demo. Several of the biggest AI companies now let you connect your actual hospital records and ask questions of them. Most of those features are US only, so Americans have more choice here than anyone else. Everything below was checked November 2026, with a source for each claim.
Disclosure
I am a doctor, and I founded exora, which lets you ask questions of your own health record. So I am describing a category I compete in, and you should weigh what I say accordingly.
So I say what each product does well, source it, and am specific about what mine does not do. I have also treated the option most people use - pasting documents into an ordinary chatbot - seriously, because it works.
The short answer
- ChatGPT Health, Claude, Perplexity Health and the Google Health app can all connect to US medical records, each with its own plan requirements. Microsoft’s Copilot Health can too, but is paused at the time of writing.
- If you are a One Medical member, Amazon’s assistant inside the One Medical app already works from the records your own care team keeps.
- Pasting documents into an ordinary chatbot still works, and is genuinely good for decoding jargon. It has real limits for anything about your history over time.
- HIPAA mostly does not follow your records into a consumer AI app. The app’s own terms do.
- The one thing worth judging any of them on: can you check the answer against the document it came from, in one step? If not, you are trusting fluency.
The products that connect to your records
ChatGPT Health
OpenAI announced ChatGPT Health in January 2026 and opened it to everyone in the US in July. It is available to “U.S.-based users over 18 across all plans”, on the web and iOS, and you can now draw on connected health information in any chat, not only in a dedicated Health space (TechCrunch, July 23, 2026, checked November 2026).
You can connect Apple Health and medical records from hospital systems on Epic and Oracle Health, plus participating providers including One Medical and Function Health (Digital Health News, updated July 29, 2026). The record connections run through b.well, which says its network reaches “more than 2.2 million U.S. healthcare providers” (b.well, January 8, 2026).
OpenAI says connected records, Apple Health data and the conversations that use them are “not used to train OpenAI’s foundation models or serve targeted advertisements”, that it works with “hundreds of physician advisors across 60 countries” to test its health answers, and that the feature is “intended to support, not replace, professional medical care” (Digital Health News). OpenAI’s own announcement is at openai.com.
What it does well: it works on every plan, and its record connection reaches a very large share of US providers.
Claude
Anthropic launched health features in January 2026: “In the US, Claude Pro and Max plan subscribers can choose to give Claude secure access to their lab results and health records”, through connectors from HealthEx and Function, with Apple Health and Android Health Connect integrations on its mobile apps (Anthropic, January 11, 2026). Anthropic’s current guide says the HealthEx connector is available to “Pro, Max, and Ultra users in the United States”, requires identity verification before any records are retrieved, and that chats using it “are never used to train our models” (Claude tutorial, checked November 2026).
What it does well: it is one of the few that works on Android as well as iPhone, and it is designed to fetch only what a question needs. The guide says “your full health record isn’t accessed unless it’s relevant to what you’re asking.”
Perplexity Health
Perplexity launched Perplexity Health in March 2026, rolling out first to Pro and Max subscribers in the US. It connects Apple Health, electronic health records “from over 1.7 million care providers” through partners including b.well, and wearables including Fitbit, Ultrahuman and Withings. Perplexity says health information is “never used to train AI models or sold to third parties”, and that answers carry citations to clinical guidelines and peer-reviewed journals (Perplexity, March 19, 2026, checked November 2026).
What it does well: citing the medical literature behind an answer is the right instinct for a health question, and it pulls records and wearables into one view.
Google Health app
In May 2026 Google relaunched the Fitbit app as the Google Health app. “In the US, you can sync your medical records to the app”, and you can get summaries of them from its Gemini-based Coach, which requires a separate Google Health Premium subscription (Google, May 7, 2026, checked November 2026). Google’s announcement repeats its commitment not to use Fitbit health and wellness data for Google Ads. It does not, as far as I could find, address AI training, so check the current terms before you connect anything.
What it does well: if you already wear a Fitbit or Pixel Watch, your records and your activity and sleep data sit in the same app.
Microsoft Copilot Health
Microsoft launched Copilot Health in preview for US adults on Microsoft 365 Personal, Family and Premium, able to connect wearables and health records, and states that what you share “is not used to train AI models”. But its support page currently says that “during the transition to the unified Copilot experience, Copilot Health will be temporarily unavailable” (Microsoft Support, last updated October 2026, checked November 2026).
Amazon One Medical
This one is different in kind. Amazon’s Health AI assistant is “now live for all One Medical members” inside the One Medical app, and draws on the records, lab results and medications that One Medical itself holds. Amazon says this is protected with HIPAA-compliant privacy and security safeguards, and that chats are not automatically added to your medical record (Amazon, January 2026, checked November 2026).
What it does well: there is nothing to connect or upload, and it sits inside the provider that treats you. The flip side is scope: it knows what One Medical knows, not your care elsewhere.
Apple Health Records
Apple’s Health Records feature downloads your clinical records from participating providers to your iPhone, and is available in Canada, the United Kingdom and the United States (checked November 2026). It is not a chatbot, but several products above can connect to Apple Health on iPhone.
The other option: paste documents into an ordinary chatbot
ChatGPT, Claude and Gemini all accept an uploaded PDF or a photo, and all do a genuinely good job of explaining what a discharge summary means in plain English. It would be dishonest to pretend this has no value. If your goal is “translate this paragraph of clinical shorthand”, it is hard to beat.
Three caveats, in order of how much they matter.
It has no memory of your history. Each conversation works from what you give it. That is fine for one document, and weak for the questions that matter over time: is this trending up, when did this first appear, have I taken this medication before, how does this compare with the same test two years ago at a different lab. Answering those needs a record, not a chat.
You cannot easily check the answer. More on this below, because it is the central issue.
Your training settings are your job. General-purpose assistants usually have a setting that controls whether your conversations help improve the model, and the dedicated health features above make separate, stronger promises than ordinary chat may. Find the setting and decide deliberately. Turning it off is generally not retroactive, and a training opt-out is not deletion. Check the current terms of whichever service you use, because these settings change.
The HIPAA point most people miss
HIPAA protects the copy of your records that your doctor, hospital or health plan keeps. It does not travel with the data. HHS is explicit that once your information is sent, at your direction, to an app that is not a covered entity or business associate, “the information is no longer subject to the protections of the HIPAA Rules” (HHS guidance, checked November 2026).
Consumer AI apps are generally not covered entities. So when you connect your records to a chatbot, what protects them is mainly the company’s own privacy commitments, plus the Federal Trade Commission, whose Health Breach Notification Rule applies to “health apps and similar technologies not covered by HIPAA” (FTC, April 26, 2024). An assistant run by your own provider, like One Medical’s, is a different situation, because the provider is covered.
That does not make the products above unsafe; their stated commitments are good ones. But a privacy policy is a promise a company can change, not federal health privacy law.
The thing actually worth judging these on
Every product in this article will give you a confident, fluent, well-organized answer. That is what these models are good at. The problem is that a fluent wrong answer and a fluent right answer look exactly the same, and in a health context you are the least equipped person in the room to tell them apart.
An illustration from medical publishing: fabricated references in biomedical papers - citations to work that does not exist - have risen sharply as AI writing tools spread: from roughly one paper in 2,828 in 2023 to one in 277 in early 2026, according to an audit published in The Lancet and reported by Retraction Watch in May 2026. The instructive detail is not the rate. It is that the fake references were in published, PubMed-indexed papers. Nothing about them signaled “check me”.
That says nothing about clinical accuracy. But the failure mode is the one that matters: the wrong output looks exactly like the right one.
Which leads to the only test I would apply. Can you get from the answer to the source in one step? Not “does it cite medical literature” - useful, but that tells you the general claim is sound, not that it read your record correctly. The question is whether it will show you the actual page and line in your actual lab report, so you can see what the answer was based on.
If it can, you can verify anything that matters. If it cannot, treat every answer as a prompt to go and check, not as a finding.
Three questions to ask any of them
- Where does my data go, and under whose rules? In most cases, the company’s privacy policy rather than HIPAA. Read the health-specific section, not just the headline promise.
- Is it used to train models? Most of the products above say no for connected health data, and the exceptions are noted in their sections. Check that the answer covers what you are actually uploading or typing, and check the setting yourself.
- Can I check an answer against my own document? The one above. If a product cannot do this, it is a chatbot with your records in it rather than a record you can question.
Where exora fits
exora is a personal health record app, on iOS, Android and web. You upload documents, such as PDFs, photos and scans, and exora reads them for you, with a reading system built for medical records. It pulls out the clinical details into one record, and you can then ask questions of your own history in plain language. exora is free to start, with an optional membership.
The part I would judge us on is the verification problem above. Every fact links back to the exact page and text of the document it came from, and chat answers cite the document they drew on, so checking is one tap rather than a project. Your health data is never used to train AI models. The security page explains how your data is protected.
Where it does not help, honestly:
- No connection to MyChart, any portal, TEFCA or any US record network. You download your records and upload them. That works, and it is a manual step. If you want automatic syncing, the products above, or the apps in this comparison that connect directly, are a better fit.
- The reading system is ours; the language models underneath it come from established providers, as with every product in this article.
- It can get things wrong. That is why every fact links back to the exact page and text it came from, and chat answers cite the document they drew on. Check anything that matters.
- Your data is stored in Sydney, Australia. exora is an Australian company. It is not a HIPAA covered entity, holds no certifications, and using it does not create a HIPAA relationship with your providers.
- It cannot read documents you do not have. You have a legal right to your records from every provider who has treated you, and this guide explains how to ask. If you are unsure how a portal and your own record differ, this explains it.
- It is not medical advice. It organizes your information and helps you understand it. It does not diagnose, it does not recommend treatment, and it is not a substitute for talking to a clinician.
A sensible way to use any of this
Use it to prepare, not to decide. These tools are genuinely good at turning clinical shorthand into English, at reminding you what happened when, and at helping you arrive at an appointment with a clear question instead of a vague worry.
If a lab result appears in your portal before anyone has explained it, these tools can help you understand the words while you wait. They should not be the last word. If an answer suggests something has changed, or something looks wrong, the next step is your family doctor or the clinician who ordered the test - ideally with the actual document in front of you both.
Product availability, plans and privacy statements are sourced to each vendor’s own pages or to reputable reporting, and claims about US law to HHS and the FTC, all linked inline and checked on November 9, 2026. This category changes very quickly, so treat the dates as load-bearing and check anything important against the vendor’s current documentation. This article is general information about software, not medical or legal advice. It does not describe or recommend any treatment. Talk to a qualified health professional about your health.