OpenAI is fundamentally changing how its AI models approach healthcare by training GPT-5 with a stronger emphasis on medical accuracy from the ground up. With more than 230 million people using ChatGPT each week for health and wellness advice, the company has made healthcare a core focus, led by Karan Singhal, a former Google researcher who joined OpenAI in mid-2024.

Why Is OpenAI Investing So Heavily in Healthcare AI?

The scale of demand is staggering. Hundreds of millions of people are already turning to ChatGPT for medical guidance, but the AI's responses have historically been less reliable than those from trained physicians. OpenAI recognized this gap and decided to make healthcare a priority rather than an afterthought. Singhal felt a personal responsibility to improve the system as more people began relying on it for health information.

Unlike previous models where healthcare performance was an add-on, GPT-5 represents the first set of models at OpenAI to be trained throughout development with a stronger emphasis on health-related performance. This means the models are learning medical accuracy from the earliest stages of training, not trying to patch it in later.

What Concrete Improvements Has OpenAI Achieved?

The results are measurable. According to OpenAI, its latest free model, GPT-5.5 Instant, performed better than both physician-written responses and GPT-4o in company evaluations. The company also recorded a 71% reduction in health-related responses flagged for inaccuracies over the past two months after analyzing billions of anonymized conversations.

To achieve these improvements, Singhal built a team of health researchers and established partnerships with more than 200 physicians. This collaborative approach, which he described as a strategy focused on "aggregating the wisdom of the crowd," led to the creation of HealthBench, a framework designed to evaluate how well AI systems perform on health-related tasks.

"Once you know how to evaluate it, it becomes a lot easier to improve it," said Karan Singhal.

Karan Singhal, Health Research Lead at OpenAI

How Is OpenAI Making ChatGPT More Like a Real Doctor?

One of the biggest limitations of AI healthcare tools is that they lack context about a patient's personal medical history. A doctor knows your age, medications, allergies, and past diagnoses. ChatGPT starts from scratch with every conversation. To address this, OpenAI introduced a health-focused ChatGPT feature earlier this year that allows users to connect health applications and upload medical records.

A key priority for OpenAI's healthcare team is making ChatGPT better at gathering information before offering guidance. Rather than immediately providing answers, the goal is for AI systems to ask follow-up questions similar to those a doctor would ask during a consultation. Singhal shared an example involving his own Apple Watch sleep data; after analyzing the information, the system suggested that his bedroom temperature could be affecting the quality of his deep sleep.

Steps to Improve AI Healthcare Interactions

• Provide Personal Context: Upload medical records and connect health apps so the AI understands your individual health profile, not just generic symptoms.
• Ask Follow-Up Questions: Encourage the AI to ask clarifying questions before providing advice, mimicking how a real doctor would approach a consultation.
• Evaluate Accuracy: Use frameworks like HealthBench to assess whether AI health responses are reliable before trusting them for important decisions.

The broader goal, Singhal explained, is to make AI healthcare tools useful not only for technology enthusiasts but for everyday users as well. People's adoption will only move at the speed of their readiness in practice, so the technology needs to guide users toward better understanding as capabilities improve.

"You definitely want the models to be ahead of everything else," said Karan Singhal.

Karan Singhal, Health Research Lead at OpenAI

As AI companies increasingly compete in healthcare, OpenAI sees the sector as a major area of opportunity. With hundreds of millions already using ChatGPT for health-related queries each week, the company is betting that conversational AI could become an increasingly important part of how people access medical information and support. The shift from GPT-4o to GPT-5 represents a fundamental commitment to making that vision work reliably.