Newsletters
More than 5% of all ChatGPT messages worldwide are already about health care, according to a January 2026 OpenAI report, part of what it describes as "billions of messages each week."
The report says that, among OpenAI's more than 800 million regular users, one in four asks a health care question weekly, and more than 40 million do so every day.
Patients view ChatGPT as an "ally" in their health care journey, based on an analysis of anonymized interactions with ChatGPT and a survey conducted by Knit, an AI-powered tool, among ChatGPT users.
Seven in 10 health care-related conversations on ChatGPT take place outside regular clinic hours.
When care is not available, asking ChatGPT can help patients decide whether they should wait for an appointment or seek urgent care.
In its report, OpenAI said: "Reliability improves when answers are grounded in the patient's correct context, such as insurance plan documents, clinical guidance, and health care portal data."
However, ChatGPT may provide incorrect and potentially dangerous advice, particularly in mental health-related conversations.
In OpenAI's accounting, the most common pain point looks less like diagnosis and more like navigation. The report says roughly 1.6 million to 1.9 million ChatGPT messages each week focus on health insurance, comparing plans, understanding prices, handling claims and billing, and parsing coverage and cost-sharing details.
It also says "three in five" U.S. adults have used AI tools for health or health care in the past three months, describing a pattern: People ask when they first feel unwell, prepare for clinician visits, translate medical terms, and then deal with the "administrative aftermath" of claims, billing, and denials.
The report cites American Medical Association survey findings that 66% of U.S. physicians used AI for at least one task in 2024, up from 38% in 2023, and that 46% of nurses use AI weekly. It also reports weekly AI use among other roles, including medical librarians (53%), administrators (43%), and pharmacists (41%).
It highlights a rural example in Miles City, Montana (population 8,400), where family physician Dr. Margie Albers uses Oracle Clinical Assist—described as relying on OpenAI models—as an AI scribe to draft visit notes and cut data entry, coding, and billing work, freeing time for triage and follow-up.
The report says she also uses OpenEvidence for “precise diagnosis and guideline checks,” returning concise answers with citations
The report urges “safe” AI expansion through secure links to public medical data, stronger validation via labs and trials, and job-training pathways.
It also calls for clarifying Food and Drug Administration pathways for AI medical devices and the regulatory scope for AI services that support physicians.
OpenAI is currently facing multiple lawsuits from people who claim their loved ones harmed themselves or died by suicide after using the technology.
Some U.S. states have also passed new laws focusing on the use of AI-powered chatbots, prohibiting apps or services from providing mental health services or making treatment decisions.
The company behind ChatGPT says it is working to improve how the tool responds in health care contexts.
It says it is continually evaluating its models to reduce harmful or misleading outputs and is working with physicians to identify risks and improve performance.
According to the company, GPT-5 models are more likely to ask users follow-up questions, search the internet for the latest research, use more cautious or softened language, and direct users to seek professional evaluation when needed.
Set as preferred source
2 min read
