The government wants AI doctors. One study shows why you should slow down.

Washington is fast-tracking AI chatbots that can diagnose and prescribe. But when 1,300 people asked for a diagnosis, the chatbots got it right just 34.5% of the time, and one missing detail turned a brain bleed emergency into “lie down in a dark room.” Here’s how to use AI for your health the smart way.

⚡ TL;DR

  • Washington is fast-tracking AI chatbots that can diagnose and prescribe. A Utah pilot program lets them refill prescriptions.
  • A massive Oxford study of 1,300 people found chatbots identified their condition correctly only 34.5% of the time. People who googled scored higher.
  • The fix isn’t avoiding AI for your health. It’s knowing exactly how to use it.

📖 Read time: 2 minutes

ChatGPT/Kim Komando

I need your help: Add Komando.com as a preferred source on Google

Your next doctor might be a chatbot. The government is trying to make sure of it. Before you cancel your physical, you need to see what happened when 1,300 people tried this.

The Washington Post reports the administration is laying the groundwork for AI chatbots that can diagnose illness and prescribe medicine, including an FDA fast track for AI health tools. A pilot program in Utah already lets chatbots refill prescriptions instantly.

🩺 The 34% problem

Oxford researchers ran the largest study of its kind, published in Nature Medicine. 

When chatbots read medical scenarios on their own, they nailed the condition 94.9% of the time. Straight-A med student. Then nearly 1,300 real people described those same scenarios to the chatbots. Success crashed to 34.5%. The control group that googled scored 47%. And chatbot users picked the right next step, like ER versus rest at home, less than half the time.

The scariest example: Two people described nearly identical symptoms of a brain bleed. One mentioned the headache came on suddenly and was told to get emergency care. The other left out that detail and was told to lie down in a dark room. 

Same emergency. Opposite advice. One missing scrap of information.

The failure runs both ways. People leave things out, and the bots rarely ask follow-up questions the way a real doctor would. They hand you a list of possibilities and let you guess. 

🔑 Use AI the right way

AI is a fantastic medical sidekick and a terrible doctor. My rules:

1. Real emergency? Call 911. Chest pain, stroke signs, trouble breathing. No typing.

2. Give the full picture. Symptoms, when they started, how fast, severity 1 to 10, your medications. One missing detail flipped that brain bleed answer.

3. Make AI brief you, not diagnose you. Paste this: “I have these symptoms: [list]. Give me the 5 most important questions to ask my doctor, what to track before my appointment and which warning signs mean I need care sooner.”

4. Always ask: “Which symptoms mean I should seek emergency care right away?” That answer is where chatbots earn their keep.

5. Never start or stop a medication on a chatbot’s say-so. Your pharmacist answers questions for free.

The AI doctor is coming whether we like it or not. Walk in informed, not replaced.

What do you call an AI and a real doctor working together? A pair-a-docs.

📩 Send this to someone who asks ChatGPT about every ache and pain. It might save them a misdiagnosis.