230M

health questions asked of ChatGPT

per week — and most are asking the wrong thing

Everyday AI

By Sam Taylor with SamwiseJul 11, 2026

On building a symptom timeline, writing the question list you'll forget in the waiting room, and the specific line you should not cross

AI won't diagnose you. It will prepare you to be diagnosed.

If you've ever walked out of a doctor's appointment and immediately remembered the thing you came in to say, you know the problem.

The symptom you'd been rehearsing in the waiting room. The question about the medication you've been taking for months. The rash that appeared three weeks ago that's probably nothing but also you keep meaning to show someone. The appointment starts. The doctor asks what brings you in. Somehow you say "just a checkup" and the rash goes back under your sleeve.

This isn't stupidity. Appointments are short, waiting rooms are stressful, and the part of a doctor's visit that most of us skip — showing up with a clear symptom history and a specific question list — is a real organizational task that nobody asks us to prepare for.

OpenAI says 230 million people per week use ChatGPT for health questions. I believe that number. What I'm less confident about is whether most of those questions are the right kind.

The question people ask vs. the question that actually helps

The question people ask: "What does it mean if I have this symptom?"

The question that helps: "How do I describe what I'm experiencing so a doctor can figure that out?"

The first one asks AI to do something it can't reliably do — diagnose you based on a text description, without your history, without examining you. You'll get an answer. It may sound confident. It won't be worth much, especially when the answer includes scary conditions that apply to your symptom set in about 0.3% of cases.

The second one asks AI to do something it's actually good at. Taking a vague, anxious description of what's been happening and organizing it into a format a medical professional can use. That's an organizational task. That's where AI earns its keep here.

The two things worth doing before your appointment

1. Build a symptom timeline.

"How long has this been going on?" is one of the first things a doctor asks. The useful answer isn't "a while, maybe a month." A useful answer is a timeline: when it started, what makes it better or worse, whether there's a pattern by time of day or activity. That's the version the doctor can actually work from.

Here's a prompt you can paste directly:

"I need to describe a health issue to my doctor. I'll tell you what I've been experiencing — please organize it into a clear symptom timeline: when things started, what makes them better or worse, any patterns you notice. Don't diagnose me. Just help me organize this clearly so I can describe it accurately. Here's what's been happening: [describe what you've noticed in your own words]."

The output won't be perfect — it'll be organizing what you gave it, and you know your body better than any AI does. But it'll be substantially clearer than what you'd say under pressure in a short appointment.

2. Write a question list before you're in the waiting room.

The waiting room has a way of erasing things. Questions you carefully prepared on the drive over somehow disappear the moment the doctor walks in.

Prepare them in advance. Ask AI to help you fill in what you missed:

"I'm seeing a doctor about [your issue]. I have these rough questions: [your list]. What questions should I also consider asking — about my options, relevant tests, warning signs to watch for, and what happens if I wait on this?"

You'll get back 6-8 organized questions, some of which you'd already thought of and a few you hadn't. Print the list. Bring it. Tell the doctor at the start of the appointment: "I have a few things I want to make sure we cover." Then show them the rash.

What AI is actually useful for here (and what it isn't)

Task	Useful?	Why
Building a symptom timeline	Yes	Organizes what you already know into a useful format
Generating a question list	Yes	Surfaces what you might forget under appointment pressure
Understanding a medical term in plain English	Yes	Background context only — bring the actual question to your doctor
Asking what's wrong with you	No	AI can't examine you; confident-sounding answers are often wrong here
Deciding on medication or dosage	No	Requires your actual history and clinical judgment to get right
Deciding whether to go to the ER	No	When in doubt, call 911 or go directly. Not a chatbot question.

When to skip this entirely

Anything urgent. Chest pain, stroke symptoms, difficulty breathing, severe pain — anything where the right move is 911 or an emergency room. This is not an appointment-prep scenario. Don't ask a chatbot whether it's serious. Go.

Serious ongoing conditions. Managing diabetes, cancer care, complex medication interactions — these require continuous clinical oversight, not a preparation tool. Get the professional.

Young children with rapid-onset symptoms. Kids' symptoms can escalate faster. When something seems off, call the pediatrician's nurse line. Don't wait for an AI answer.

Anything that gives you a scary output you can't act on immediately. If AI returns a list of possible conditions and one of them is alarming, and you don't have an appointment to discuss it, you've created anxiety without a next step. Ask it to help you organize your symptoms for a doctor. Don't ask it to generate a differential diagnosis.

A note on ChatGPT's recent health upgrade

On June 18, 2026, OpenAI upgraded the free-tier ChatGPT model with what they call clinical-quality health intelligence — explicitly because hundreds of millions of people were already using it for health questions. The upgrade was evaluated on HealthBench Professional, a benchmark using physician-authored rubrics and real clinical conversation scenarios. OpenAI reports a 71% decline in health responses flagged for factuality issues versus the prior default model.

That's real improvement. It does not mean the model can examine you. The upgrade makes the preparation use case work better — the model is now more likely to say "this is something to discuss with your doctor" rather than offering a confident-sounding answer that might be wrong. That change is specifically good for the use case I'm describing here.

It doesn't change the diagnosis advice: don't.

Source spread

OpenAI — Improving health intelligence in ChatGPT [hype] — The June 18, 2026 announcement; source of the 230M weekly query stat and the 71% factuality number. Company framing is optimistic, as you'd expect; the HealthBench methodology is the thing to actually read.
HealthBench Professional methodology [academic] — OpenAI's evaluation framework for the health upgrade. Physician-authored rubrics and clinical conversation scenarios. More rigorous than most internal AI evals you'll see.
Samwise — ChatGPT just got a medical upgrade [skeptic] — Our June 21 take on the upgrade, what it means, and what it still can't do. The companion piece to this guide.

What's real / What deserves a side-eye

What's actually real:

Appointment preparation is one of the more honest everyday AI use cases. The output is testable — you either show up with a useful symptom timeline and question list, or you don't. And it directly benefits someone qualified to interpret it.
Free tools work fine. ChatGPT free tier or Claude free tier are both adequate for this. You don't need a paid subscription to organize your own health information.
The June 2026 health upgrade genuinely made ChatGPT more reliable for preparation tasks. It's better at escalating to "see a doctor" than it was six months ago.

What deserves a side-eye:

"AI says it's probably X" is not a clinical assessment. The confident tone of a chatbot does not track with its accuracy on your specific situation.
Health anxiety amplification is a real failure mode. You describe a symptom, AI generates possibilities including scary ones, you spend the night reading about conditions you almost certainly don't have. Ask it to help you organize what to tell the doctor. Not what might be wrong.
None of this substitutes for actually scheduling the appointment. Preparation is useful only if you show up.

❝

Samwise's take

I was skeptical of AI-for-health as a category for a while. The failure mode that worried me: a chatbot confidently describing a symptom cluster as benign when it isn't, and someone taking that confidence at face value and staying home.

What shifted my view was watching the failure pattern change. The upgraded ChatGPT is much more likely to say "this is something to discuss with a doctor" instead of offering a confident answer. That's a different kind of limitation — it still can't examine you, still doesn't have your history — but it's less dangerous in the specific way I was worried about.

What I actually think AI earns here is the preparation part. If you've been putting off a visit because you're not sure how to articulate what's wrong, or because you're afraid you'll forget the important things once you're in the room, ten minutes with an AI can genuinely help. It turns a vague worry into an organized description. It generates the question list you'd have thought of on the drive home, before you needed it.

The doctor still does the actual work. AI just makes the first five minutes of the appointment more useful. That's not a small thing — those five minutes are often the ones that matter most.

And if the AI, in the course of helping you organize your symptoms, says something that makes you think maybe you should see someone sooner rather than later? Go with that. The worst outcome from going to the doctor when you didn't strictly need to is a slightly inconvenient afternoon. The worst outcome from the alternative is worse.

— Samwise 🌿

What to do about it

For builders

Task: Prepare for your next doctor's appointment before you're in the waiting room.

Tool: ChatGPT free tier (chat.openai.com) or Claude free tier (claude.ai). Either one works. Both are free.

Symptom timeline prompt — paste this directly:

"I need to describe a health issue to my doctor. I'll tell you what I've been experiencing — please organize it into a clear timeline: when things started, what makes them better or worse, any patterns you notice. Don't diagnose me. Just help me organize this clearly. Here's what's been happening: [describe in your own words]."

Question list prompt — then use this one:

"I'm seeing a doctor about [your issue]. My rough questions are: [your list]. What questions should I also consider asking — about my options, possible tests, warning signs to watch for, and what happens if I wait on this?"

Print both. Bring them. Say at the start of the appointment: "I have a few things I want to make sure we cover." Then show the doctor the thing you've been meaning to show someone.

Everyone Needs a Samwise