OpenAI says GPT-5.5 Instant, the default exemplary for free ChatGPT users, now performs comparably to its frontier Thinking models connected wellness questions. The claim is based connected the company’s ain wellness evaluations.
Health is 1 of the categories drafting the astir scrutiny complete AI-generated answers. For example, a Guardian investigation reported that immoderate Google AI Overviews provided inaccurate aesculapian guidance, and Google later removed AI Overviews for certain medical queries. OpenAI’s update lands successful that aforesaid high-risk category, but pinch a declare of betterment alternatively than a retreat.
For publishers and SEOs successful health, that intends a large, free assemblage tin get aesculapian answers successful ChatGPT alternatively of clicking done to a source.
What OpenAI Reported
OpenAI points to gains connected HealthBench and HealthBench Professional, the objective version. It says GPT-5.5 Instant scores higher than GPT-5.3 Instant, the exemplary it replaced.
The institution besides reported a driblet successful factuality problems connected unrecorded traffic. It says the complaint of wellness responses flagged for astatine slightest 1 imaginable factuality rumor fell 71% complete 2 months. That fig comes from monitors OpenAI runs connected accumulation traffic.
OpenAI ran a 3rd comparison against physicians. It asked doctors to constitute responses to typical wellness conversations, past had a abstracted sheet of physicians comparison those pinch exemplary responses. In that comparison, the sheet rated GPT-5.5 Instant’s responses higher than the physician-written ones connected criteria including accuracy, communication, and completeness, crossed 3,500 reviewed responses.
OpenAI says the exemplary showed less nonaccomplishment modes than some older models and the physicians. It pointed to less cases of missing a reddish emblem aliases failing to inquire the personification for much context.
How OpenAI Measured It
HealthBench is simply a benchmark the institution built pinch its expert network, utilizing doctor-written rubrics alternatively than exam-style questions.
OpenAI says it useful pinch much than 260 physicians crossed 60 countries and that doctors person reviewed much than 700,000 illustration responses to date. The institution has cited the 260-physician fig since it launched ChatGPT Health successful January. None of the results person been published for extracurricular review.
Health Is Already One Of ChatGPT’s Biggest Use Cases
OpenAI has said much than 230 cardinal group inquire ChatGPT wellness and wellness questions each week, 1 of the astir communal reasons group usage the chatbot.
Health besides sits successful a protected class successful OpenAI’s policies. When the institution began testing ads successful ChatGPT, it said it would not tally them successful conversations astir health, intelligence health, aliases politics.
Why This Matters
Medical queries already tie dense AI-answer exposure, pinch the highest complaint of immoderate class successful a caller Ahrefs analysis of Google’s AI Overviews. More of that request moving into ChatGPT’s free tier could summation the zero-click unit connected publishers.
The accuracy claims are harder to enactment on. OpenAI ran the tests in-house, truthful you look the aforesaid measurement spread arsenic pinch different AI answers successful health. The institution says its wellness responses improved, but the claims aren’t verified by an independent third-party.
Looking Ahead
The station doesn’t specify really changes effect citations. If much platforms displacement wellness answers to free tiers, verifying answers and handling postulation nonaccomplishment go the practitioners’ responsibility.
English (US) · 
Indonesian (ID) ·