OpenAI will let ChatGPT Health conversations feed model training from November 19

OpenAI has told users that conversations in ChatGPT Health will be eligible for use in model training starting November 19, according to a report carried through Google News. The change affects one of the most sensitive categories of data that passes through the company's products.

The policy applies to ChatGPT Health, the dedicated health-facing experience rather than general ChatGPT chats. What the notice establishes is the date: from November 19, those conversations can feed into training. The report centers on that effective date rather than on retention windows, deletion mechanics or whether users have a say, which are the details most people will want first.

Health conversations are unusual data. They are intimate and unstructured, and they often contain things people would not type into a general chatbot: symptoms described in their own words, medication names, test results, family history. Training on that material raises questions that ordinary chat logs do not.

Timing matters here. Health is one of the fastest-moving product categories in generative AI, with companies racing to turn triage, symptom checking and record summarization into default entry points. Whoever holds real health question-and-answer data has a better shot at differentiating a model's performance in clinical language.

For OpenAI the appeal is straightforward. Real patient questions and clinical language are hard to source at scale: hospital data is bound by privacy rules, and public corpora rarely contain the colloquial way people describe being unwell. Conversations inside the product are the most direct supply.

The cost is trust. People tell a health assistant things they would not tell a search engine, and that candor rests on an assumption about what happens to the transcript. A policy change can be lawful and still alter how users phrase their symptoms, which in turn changes the quality of the data being collected.

Three things are worth watching. How OpenAI communicates the change inside the product; whether the notice clearly separates training use from inference; and whether health-data regulators in stricter markets respond. For users, the practical question is not the November 19 date but how long a conversation about their own body is kept, and who gets to see it.