OpenAI claims that GPT-5.5 Instant, an automated model for users of the free ChatGPT, is now more efficient than its own boundary-pushing models for health questions. The claim is based on the company’s own health assessment.
Healthcare is one of the sectors that is most closely watching the answers generated by AI. For example, a Guardian investigation reported that some Google AI Overviews provided inaccurate medical guidance, and Google later removed AI Overviews for certain medical questions. The OpenAI update remains in that high-risk category, but with the improvement claim there is a backlash.
For health publishers and SEOs, that means a large, free audience can find medical answers on ChatGPT instead of clicking on a resource.
Reported by OpenAI
OpenAI points to benefits in HealthBench and HealthBench Professional, the clinical version. It says the GPT-5.5 Instant scores higher than the GPT-5.3 Instant, the model it replaced.
The company also reported a decrease in authenticity issues in live traffic. It says the rate of health responses flagged with at least one true story dropped by 71% in two months. That number comes from monitors OpenAI runs on production traffic.
OpenAI ran a third comparison against doctors. It asked doctors to write answers to independent health interviews, and then a separate panel of doctors compared those with model answers. In that comparison, the panel rated the GPT-5.5 Instant answers higher than the prescriptions on criteria including accuracy, communication, and completeness, out of all 3,500 reviewed answers.
OpenAI claims the model showed fewer failure modes than both older models and doctors. It pointed to several instances of missing a red flag or failing to ask the user for more context.
How OpenAI scales
HealthBench is a benchmarking company that builds on a network of physicians, using physician-written rubrics rather than exam-style questions.
OpenAI says it works with more than 260 doctors in 60 countries and that doctors have reviewed more than 700,000 sample responses to date. The company has spoken with 260 doctors since launching ChatGPT Health in January. No results have been published for external review.
Life is already one of the biggest use cases for ChatGPT
OpenAI said more than 230 million people ask ChatGPT health and wellness questions each week, one of the most common reasons people use the chatbot.
Health also sits in a protected category in OpenAI policies. When the company first tested ads on ChatGPT, it said it would not run them in conversations about health, mental health, or politics.
Why This Matters
Medical questions are already drawing the heaviest AI exposure, with the highest rate of any category in Ahrefs’ latest analysis of Google’s AI Overviews. More of that demand to move to the free ChatGPT category may increase click-through pressure on publishers.
Accuracy claims are difficult to make. OpenAI has conducted experiments in-house, to address the measurement gap like other AI responses in health. The company says its health responses are improving, but the claims have not been verified by an independent third party.
Looking Forward
The post does not specify how the changes affect quotes. When many platforms switch health answers to free topics, verifying answers and handling the loss of traffic becomes a staff responsibility.
