An AI doctor,
available right now.

Most consultations are fifteen minutes. Most people spend the first five describing something they have not had time to organise, and leave without asking the thing they came for. The Health Copilot is for the hours before that - understanding the result, working out what is worth raising, and arriving with three specific questions rather than a general worry.

That is a genuinely useful job and it is not the same job as diagnosing. It will not tell you what you have, and a tool that offered to would be doing something both unsafe and unlawful.

Clinicians tend to be in favour of this, for the same reason attorneys are: a patient who arrives organised gets more out of a short appointment, and so does the person seeing them.

A copilot that answers everything is dangerous on this topic. This one is built to recognise the categories where the correct output is "see someone now, not later" - chest pain, sudden neurological change, breathing difficulty, anything suggesting sepsis, and every mental-health presentation with risk attached.

On those it gives you the escalation path and the emergency number, and it does not attempt the analysis. That is a deliberate refusal, and it is the most important behaviour on the page.

Health is the domain where a general-purpose model is most heavily constrained, and for defensible reasons - the downside of a confident wrong answer is severe and the model has no way to know whether it is talking to a patient, a nurse or a worried parent at 2am.

The result is text that is technically accurate and practically useless: "consult a healthcare provider" in response to a question about what a number means. That is not caution, it is a non-answer, and it leaves people to search the symptom instead - which is strictly worse.

A specialist configured for this domain can explain the number, give the reference range, name what was not tested, and still refuse to diagnose. Those are separable, and treating them as one thing is what produces the hedge.

People bring this copilot lab results, medication lists and things they have told nobody. That deserves a direct answer rather than a policy link.

Conversations are encrypted in transit and at rest, are not used to train models, and are not shared with third parties. You can export or delete everything at any time.

The thing to understand, and it is stated plainly rather than buried: Copilotly is not a HIPAA-covered entity. Your conversation is private, but it does not carry the legal confidentiality that a conversation with your doctor does. For most uses that distinction does not matter. For anything that might become contested - a disability claim, an insurance dispute - it might.

Primary care visits in the US average somewhere around fifteen to eighteen minutes, and a meaningful part of that goes to the computer rather than to you. That is not a failure of individual doctors; it is what the scheduling and reimbursement system produces.

The practical consequence is that a visit rewards preparation in a way nobody tells patients. A clear timeline, a specific list of symptoms, an accurate medication list and two or three prioritised questions can double what fits into the same appointment.

The opposite is also true and more common. A visit that opens with a vague description, spends five minutes reconstructing when something started, and surfaces the actual concern in the final thirty seconds - the classic doorknob question - produces a rushed answer to the most important thing you came for.

Physicians have written about this for years. The patients who get the most from the system are not the ones with the most knowledge; they are the ones who arrive organised. That is a solvable problem, and one of the few genuinely high-leverage things a patient can do.

Write it down beforehand. Bring the list. Ask the most important question first rather than last.

Lab results arrive in patient portals now, often before anyone has looked at them, which means a great many people read flagged values with no context at all. A few things make that considerably less alarming.

Reference ranges are population statistics, not health thresholds. They are typically set so that around 95% of a healthy reference population falls inside, which means roughly one in twenty healthy people sits outside on any given test by definition. Run a twelve-marker panel and a flagged value is close to expected.

Direction and magnitude matter more than the flag. A value marginally outside the range means something different from one at twice the upper limit, and the flag treats them identically.

Trend beats snapshot. A single result is a point; three results over eighteen months is information. Clinicians think in trends and portals display points, which is much of why the portal is more frightening than the appointment.

Context decides everything else. The same number means different things depending on age, sex, medication, recent illness, whether you fasted, and what the test was ordered to investigate. A copilot can explain what a marker measures and what generally moves it. Only a clinician who knows your history can say what yours means, and that distinction is worth holding onto when reading at eleven at night.

Some patients are consistently under-treated, and the research on this is not ambiguous. Pain in particular is under-recognised in women and in Black patients across a large body of studies. That is a systemic finding rather than an accusation about any individual clinician, and knowing it is useful because it changes what to do about it.

Specificity helps more than insistence. Saying it hurts is easy to discount. Saying it is a seven out of ten, it wakes you at night, it has stopped you driving, and it has been eight weeks is a clinical picture that is much harder to set aside.

Asking for the reasoning is legitimate and effective. What else could this be, and what would change your mind, is a question clinicians answer readily, and it often surfaces the differential that was never said out loud.

Asking for it in the record is the strongest available move. Requesting that a clinician note in your chart that you raised something and that the decision was not to investigate it is entirely reasonable, rarely refused, and changes the conversation - because the record is what the next clinician reads.

And a second opinion is normal. For anything serious, anything surgical, and anything where the first answer does not fit what you are experiencing, it is standard practice rather than a challenge to anyone.