Copilot Health, ChatGPT Health & Amazon Health AI, Compared

Between January and March 2026, Microsoft, OpenAI, and Amazon each shipped a consumer AI health assistant. Our guide to AI agents in family health covered why that happened all at once and what “agent” actually means for your data; our piece on AI agent vs. chatbot covered the access questions every one of these tools raises. This piece answers the question those two left open: how do Copilot Health, ChatGPT Health, and Amazon Health AI actually differ, once you get past the marketing pages? (Anthropic’s Claude for Healthcare and Perplexity Health, the other two products from the same early-2026 wave, are outside the scope of this comparison and worth a separate look.)

What each one actually does

All three connect to outside medical records and answer questions about them, but the plumbing behind that differs. Copilot Health lives inside Microsoft Copilot and connects to records and wearable data through a network Microsoft calls HealthEx, which it says reaches more than 50,000 US hospitals and providers, plus 50-plus wearable and health-app platforms including Apple Health, Oura, and Fitbit. It surfaces trend insights, helps interpret results, and helps prepare questions for a clinician, and Microsoft is explicit that it isn’t meant to diagnose or treat. It launched March 12, 2026, and as of Microsoft’s May 29, 2026 update, linked above, remains in preview rather than general availability.

ChatGPT Health is a dedicated space inside ChatGPT that connects to medical records through b.well’s network of roughly 2.2 million US providers, along with wellness apps like MyFitnessPal and Peloton. OpenAI has said 230 million people, about 29% of ChatGPT’s weekly users, were already asking health and wellness questions before this feature existed, according to TechCrunch’s coverage of the launch.

Amazon Health AI is the most “agentic” of the three: beyond answering questions, Amazon says it can help book appointments, manage prescriptions, and connect users directly to a licensed One Medical clinician, built on Amazon Bedrock and reviewing records pulled from state health information exchanges. It launched first for One Medical members in January 2026, then expanded to the general Amazon.com and Amazon app audience in March, per Fierce Healthcare.

HIPAA status, compared

This is the single biggest difference between the three, and it rarely shows up on a marketing page. Amazon Health AI is the only one of the three built on top of an actual HIPAA-covered relationship: it operates inside One Medical, a licensed medical practice, backed by a published Notice of HIPAA Privacy Practices. ChatGPT Health is the opposite case, and the clearest of the three: HIPAA Journal’s analysis confirms OpenAI does not sign a consumer business associate agreement for ChatGPT, and Paubox’s health-privacy reporting put it plainly in its headline: medical records shared with ChatGPT Health are not HIPAA-protected. A separate enterprise product, ChatGPT for Healthcare, is HIPAA-BAA-eligible, but that’s sold to healthcare organizations, not to a family using the consumer app.

Copilot Health sits in the middle by default, not by design: it simply isn’t confirmed. Microsoft has published HIPAA business associate agreement terms for enterprise products like Copilot for Security and Copilot Studio, but no official Microsoft page states that consumer Copilot Health carries the same coverage. That absence is worth treating as a “not currently confirmed” answer, not a “probably fine” one, until Microsoft states otherwise directly.

Does it train on your data, and how long is it kept?

Microsoft and OpenAI both say no. Microsoft’s own Copilot privacy FAQ states users can opt out of model training while keeping personalization, and Axios reported that Copilot Health data specifically won’t be used to train Microsoft’s models. OpenAI goes further for this feature specifically: connected medical records and any conversation that uses them are excluded from training regardless of your general ChatGPT training setting, according to BleepingComputer’s reporting on the policy. That’s a stronger, more specific commitment than the general opt-out toggle that applies to the rest of ChatGPT.

Amazon is the exception, and says so directly. As Yahoo Tech reported, quoting the company’s own statement: “Amazon trains Health AI models on abstracted patterns without directly identifying information—for example, if multiple patients ask about medication interactions, those patterns may be used… without patient names.” TechRadar has separately pushed back on whether de-identification alone is a sufficient safeguard, pointing to research showing re-identification is sometimes possible even from patterns that don’t include a name.

On retention, all three come up short in the same way. A April 2026 analysis from the International Association of Privacy Professionals reviewed all five major 2026 health AI launches, including these three, and found that only Perplexity Health partially specified a retention duration for uploaded health data. Copilot Health, ChatGPT Health, and Amazon Health AI don’t publicly state how long they keep what you share, which means “how long is this kept” is a question you have to ask directly rather than one any of the three answers up front.

Pricing and family access

Copilot Health requires an individual Microsoft 365 subscription: Personal runs $9.99 a month or $99.99 a year, and Premium runs $19.99 a month or $199.99 a year, available to US subscribers 18 and up while the feature is in preview. The catch for families: even on the Microsoft 365 Family plan, which covers up to six people, AI features including Copilot Health aren’t automatically shared across everyone on the plan. Each person needs their own individually qualifying subscription to actually use Copilot Health, according to Microsoft’s own Microsoft Q&A support channel (a Microsoft-hosted forum, not an official policy page, so treat this specific point as the best available answer rather than a formal Microsoft statement). The word “Family” in the plan name describes the billing group, not shared AI access.

ChatGPT Health is available on Free, Go, Plus, and Pro tiers for logged-in US users 18 and older, so there’s no hard paywall to try it, though full record-connection functionality appears to skew toward the paid tiers in practice. Like Copilot Health, it’s a single-account product; there’s no documented way for one account to see another family member’s connected records.

Amazon Health AI has no standalone subscription fee of its own. It sits on top of Amazon Prime plus a discounted One Medical membership: $99 a year for a Prime member’s first membership versus $199 a year without Prime, and $66 a year for each additional household member, or pay-per-visit at $29 for a message-only consultation and $49 for video without any membership at all. That per-additional-member pricing is the closest thing to an actual family structure among the three, though it’s a membership discount, not a confirmed shared view inside the AI chat itself; no source we reviewed documents whether a logged-in family member can see a linked parent’s or child’s records through the assistant. For a sense of what that spending adds up to over time, our guide to the hidden retirement cost of caregiving looks at the broader financial picture beyond any single subscription.

What the critics are saying

None of the three has escaped scrutiny. MIT Technology Review interviewed six academic experts about all three products in March 2026, and every one of them raised safety-testing concerns. The piece cites a Mount Sinai study finding that ChatGPT Health sometimes recommends excessive care for minor issues while missing genuine emergencies, and notes that newer model versions performed worse at seeking patient context than older ones did. Separately, a study reported by WISH-TV found ChatGPT-style tools failed to correctly identify a meaningful share of both non-emergency and emergency conditions in clinical vignette testing, and the patient-safety nonprofit ECRI named misuse of AI chatbots the top health technology hazard of 2026 in the same report.

Copilot Health drew a different kind of scrutiny: a Forbes contributor piece framed its launch as a shift toward AI offering more actionable medical guidance, raising scope-of-practice questions about where a general-purpose assistant’s advice should stop. Amazon has faced the sharpest privacy pushback of the three: both Yahoo Tech and TechRadar published pieces questioning whether a company that also runs a retail and pharmacy business should hold health conversations at all, citing concern that a mention of poor sleep could plausibly surface an Amazon product recommendation. Amazon’s response, on record with Fierce Healthcare: “This isn’t a chatbot with a healthcare skin. It’s a system designed from the ground up to be safe, trustworthy, and useful.”

The gap between asking and sharing

Adoption and trust aren’t the same thing, and the data on older adults specifically shows that gap clearly. AARP Research fielded a survey of 1,031 US adults 50 and older in mid-February 2026, published in June 2026, and found that while a meaningful share have tried asking an AI tool a casual health question, far fewer have taken the bigger step of actually sharing a document.

That last number is the one worth sitting with. More than half of older adults who’ve engaged with AI health tools at all don’t actually know where their information goes once they’ve shared it. Given that none of the three products in this comparison discloses a retention period, that uncertainty isn’t a knowledge gap on the reader’s part. It’s a disclosure gap on the product’s part.

What to check before you connect one to family health data

The same checklist applies whether you’re looking at one of these three or a different product entirely, and it’s worth pairing with the broader access questions covered in our AI agent vs. chatbot breakdown.

Where Kaizen Health fits

Copilot Health, ChatGPT Health, and Amazon Health AI are each a health feature inside a much larger, general-purpose product, and the comparison above shows what that trade-off costs: unclear or absent HIPAA status in two of the three cases, no disclosed retention period in any of them, and no product genuinely built for a family to share. Kaizen Health starts from the opposite direction. Kai, Kaizen’s AI assistant, summarizes the records a family chooses to upload and answers questions about them on a HIPAA-compliant platform, with business associate agreements signed with the AI providers used in specific app features, and personal health data is never used to train the underlying model.

None of this means the three products compared here are unsafe to use for a quick, one-off health question; each is transparent about at least some of what it does and doesn’t do, which is more than can be said for plenty of consumer software. It means the comparison changes once a parent’s full record, or a shared family view of it, is what you actually need. If you haven’t organized what records exist yet, our guide to organizing medical records for aging parents is a reasonable place to start before connecting any AI tool, including this one.