Last week, the Texas Responsible AI Governance Act took effect. Doctors' offices in Texas now have to tell patients, in writing, when artificial intelligence might be involved in their diagnosis or treatment. Other states are queuing up similar laws. The federal direction looks like the state direction, just slower.
But you don't have to live in Texas to ask. AI has been in your doctor's office for years already — listening to the visit, drafting the after-visit summary, flagging the X-ray, suggesting the next medication, reading the EKG, scoring the diabetic retinopathy photo, generating the chart note before the doctor even gets back to their desk. In most clinics, in most appointments, none of this gets mentioned. It isn't hidden, exactly. It just hasn't been a category that comes up.
The thing about being a family caregiver is you have standing to ask. You're the one taking notes at the appointment. You're the one keeping the medication list, the appointment calendar, the chart of how your mother has been over the last three months. You're often the one a clinician asks about how it's going at home. That standing extends to asking about the tools shaping the recommendations being made.
These are the five questions worth having ready. They aren't gotchas, and they aren't anti-AI. They're the questions a thoughtful patient or family member would ask a thoughtful doctor about any new piece of clinical apparatus, except this apparatus is invisible and increasingly load-bearing.
1. "Is AI being used to help diagnose or recommend treatment for my parent?"
This is the opening question. It establishes that you know AI exists in the workflow, that you're not afraid of it, and that you'd like to understand its role.
A good answer is specific. "Yes — we use an AI tool that flags abnormal patterns on the EKG, and another one that drafts our visit notes." Or "Yes — the radiology group runs every chest CT through a pulmonary nodule detector before the radiologist reads it." Or even "Not in clinical decisions yet, but the after-visit summary I'll send you was drafted by an AI tool."
A less-good answer is "No, we don't use AI." That answer is almost certainly wrong in 2026 — the question is whether the doctor knows what their EHR vendor, lab, radiology partner, and dictation system are doing. If you hear it, a gentle follow-up: "What about the radiology group? Or the lab? Or the dictation system?"
The point isn't to catch the doctor. The point is to surface the conversation so the next questions have somewhere to land.
2. "Was this AI tested in patients like my parent?"
This one matters more than people realize.
AI tools in medicine are trained on data. The data is whatever was available — often academic medical center patients, often disproportionately younger, often white, often English-speaking, often without the multiple chronic conditions older adults have. A tool that works beautifully on a 45-year-old marathoner with one health concern can give startlingly wrong answers about an 82-year-old with four chronic conditions, two languages, and a long medication list.
You're asking the doctor to consider whether this particular tool was validated in patients who look like your parent. The honest answer is sometimes "I don't know — I'll check." That's a good answer. The not-good answer is annoyance at the question.
A useful follow-up if you have time: "Are there known patterns of who the AI tends to be wrong about?" Most well-built tools come with documentation about this. Most clinicians haven't read it. Asking creates the small social pressure that nudges the system toward reading it.
3. "Who reviews the AI's output before it affects my parent's care?"
This question gets at the deepest concern, which is agency — who is actually making the decision.
In a well-functioning AI-assisted clinical workflow, the AI is a co-pilot and the clinician is the pilot. The AI flags. The doctor decides. The AI drafts. The doctor edits. The AI suggests. The doctor accepts, modifies, or rejects.
In a less well-functioning workflow, the AI's output flows through with very little review. The radiologist clears 200 chest CTs an hour because the AI cleared them first. The dictation tool's note becomes the chart note nobody re-reads. The medication-recommendation engine's suggestion becomes the prescription that gets written.
You're asking the doctor to surface their actual workflow. Most will answer honestly. "I review everything the AI flags." "I read the AI-drafted note and edit before signing." "The radiologist looks at every scan even if the AI says it's clear." Those are good answers. "It happens automatically" is a worse answer; it tells you the doctor is not the decision-maker for that piece.
The asymmetric truth: you can't expect every clinician to have a thoughtful answer for every AI tool in their workflow. You can expect them to know which ones they personally use, and what their review process is for those. If the answer is fuzzy, that's information.
4. "Can I see what the AI saw — and what it recommended?"
This one is harder for the doctor to answer, but the conversation is worth having.
When AI is woven into the workflow, the output is often visible to the clinician but not surfaced to the patient. The drafted note becomes the chart note; you see only the chart note. The medication recommendation becomes the prescription; you see only the prescription. The image flag becomes the radiologist's read; you see only the read.
Most patients have a legal right to their chart — including the AI-generated parts of it. The 21st Century Cures Act in 2021 made notes broadly available to patients through the patient portal. AI-generated notes are notes; they should be there.
What's often missing is the AI's underlying recommendation — what the model flagged or suggested before the clinician edited it. Some EHRs now log this; many don't. Asking your doctor about it does two useful things. It surfaces whether such logging exists in your parent's chart. And it signals to the clinician that AI transparency matters to your family — which, multiplied across enough families, shifts the institutional defaults over time.
5. "What happens if the AI is wrong?"
Every AI tool is wrong sometimes. The good ones are wrong rarely; the bad ones are wrong often; all of them are wrong eventually. The question is what the system does about it.
A good answer to this question has a few elements. The clinician describes how they would catch an AI error: their own clinical judgment, a second look at the data, a peer review, a patient or family member raising a concern. The clinician describes the path to correct an error once it's caught: an addendum to the chart, a discussion with the patient or family, a re-evaluation of the recommendation. The clinician describes any formal reporting they would do: many AI tools have feedback loops that improve them; some clinicians are part of those loops; some are not.
A worse answer is "It doesn't get wrong" (it does), or "The vendor handles that" (the vendor doesn't know about your parent), or a shrug.
The follow-up worth saving for this one: "What's the role of the family in catching an AI error?" You're often the person most likely to notice when the recommendation doesn't match the person you've been caring for. "Mom doesn't react well to that class of medication; I think the AI may not have known her last response." That observation, made calmly and quickly, can rescue a treatment plan from a quiet mistake.
How to actually have the conversation
The conversation is easier than it sounds. Most clinicians genuinely don't mind being asked. Some are relieved — they've been thinking about it too. A few are defensive, but defensiveness usually fades once they realize you're not anti-AI, you're pro-thoughtful-AI.
The conversation lands better if you bring it in once, not five times. Pick one of the five questions for the next appointment. Hear the answer. If the answer is good, the relationship strengthens. If the answer is fuzzy, you have somewhere to go from there.
The conversation lands better still if you frame yourself as part of the team rather than as oversight. "I want to understand the tools so I can help mom understand them too" is a different posture than "I want to make sure you're using AI responsibly." Both are reasonable; the first lands better with most clinicians.
If the answers don't satisfy — at all, repeatedly, across multiple clinicians — that's a real signal. The next appointment can be with a different doctor, or in a different system, or in a different specialty group. The market for clinicians who can answer these questions thoughtfully is growing. You're allowed to choose one.
A note about Kintaria
We use AI in Kintaria. We disclose it specifically. Visit summaries are drafted by AI and labeled as such. The voice line at (888) 704-0999 is an AI agent that identifies itself as one. Smart upload is an AI extraction tool that surfaces what it found and asks you to confirm. The medication review uses AI to flag interaction risks for the clinician's review. We don't pretend any of this is something else.
The reason we list it here, at the bottom of an essay about asking your doctor: we'd want a thoughtful family caregiver to ask us the same five questions. The answers are public, the tools are documented, the limitations are written down. That's the version of AI in healthcare we want to see across the field. The conversation you have with your doctor is part of getting there.