Key takeaways
Separate three ChatGPT options: consumer ChatGPT (including ChatGPT Health), the free ChatGPT for Clinicians workspace, and the enterprise ChatGPT for Healthcare.
Consumer ChatGPT plans may use your content for training unless you opt out; ChatGPT for Clinicians conversations and business-plan data are not used for training by default.
OpenEvidence cites licensed sources such as NEJM Group and JAMA Network content; OpenAI says ChatGPT search citations can be incomplete, outdated or incorrect.
Accuracy claims from both vendors are self-reported, and HealthBench is a benchmark OpenAI built and grades with its own model.
Side-by-side comparison
| Criterion | OpenEvidence | ChatGPT (consumer and for Clinicians) | Vera Health |
|---|---|---|---|
| Intended use | Clinical questions from verified healthcare professionals. | Consumer ChatGPT Health is not intended for diagnosis or treatment; ChatGPT for Clinicians is a workspace for verified clinicians. | Clinical questions from licensed clinicians and trainees; its terms describe a reference library, not patient-specific diagnosis. |
| Access | Verified clinicians with an NPI or another licence number; excluded in the EU, EEA and UK; Canadian physicians can verify through MINC. | Consumer plans are widely available; ChatGPT for Clinicians is for verified US physicians, nurse practitioners, physician assistants and pharmacists. | The vendor says it is available outside the US, including the EU and UK, except in sanctioned countries. |
| Cost | Free for verified clinicians; a limited number of labeled ads. | Free, Go ($8), Plus ($20) and Pro (from $100) a month; ChatGPT for Clinicians is free for eligible US clinicians; Business from $20 per seat a month billed annually. | Free for licensed clinicians and trainees; enterprise pricing is custom. |
| Citations | Answers cite their sources, with labels such as highly relevant, top journal and new research. | ChatGPT for Clinicians and ChatGPT for Healthcare offer cited clinical search; OpenAI warns that search citations can be wrong. | Retrieval-first answers over more than 60 million papers, guidelines and care pathways (vendor figure), with source links. |
| Patient data and training | BAA for US covered entities; its terms say PHI is not used for AI training. | Consumer content may train models unless you opt out; ChatGPT for Clinicians conversations are not used for training; optional BAA for eligible accounts. | BAA available for covered entities; its terms ask users not to submit patient-specific data for diagnosis. |
| Beyond Q&A | Visits drafts notes from encounters; models range from Osler (fast) to Snow (deep). | Deep research and CME in ChatGPT for Clinicians; an Epic plugin in ChatGPT for Healthcare. | Drug database with interaction checks and 900+ calculators. |
Based on OpenEvidence’s terms, policies and announcements, OpenAI’s product announcements, help center, usage policies and pricing page, and Vera Health’s site and terms, checked October 19, 2026. Vendor figures are labeled as such.
Which to use when
Our top-ranked tool
Vera Health
Cited answers plus a drug database, interaction checks and 900+ calculators, free for licensed clinicians. It is also an option where OpenEvidence is unavailable. Its benchmark results are vendor-run, so test it on your own questions.
US clinicians who want licensed journal sources
OpenEvidence
Free for verified clinicians, with answers grounded in licensed content such as NEJM Group and JAMA Network titles. Not available in the EU, EEA or UK.
US clinicians already working in ChatGPT
ChatGPT for Clinicians
Free cited clinical search, deep research and CME for verified US clinicians, with an optional BAA for eligible accounts. Check which features the BAA covers before entering patient information.
Not for clinical decisions
Consumer ChatGPT
OpenAI says ChatGPT Health is not intended for diagnosis or treatment, and its usage policies bar tailored medical advice without appropriate involvement by a licensed professional. Consumer plans may also train on your content unless you opt out.
Three versions of ChatGPT, three different answers
OpenAI launched ChatGPT Health for consumers on January 7, 2026, as a separate space that can connect medical records and Apple Health, and expanded it to US adults on the web and iOS in July. It says ChatGPT Health is not intended for diagnosis or treatment. ChatGPT for Healthcare, announced January 8, 2026, is an enterprise product with cited evidence retrieval and a BAA, used by health systems including HCA, Cedars-Sinai, Memorial Sloan Kettering and UCSF. ChatGPT for Clinicians, launched April 22, 2026, is free for verified US physicians, nurse practitioners, physician assistants and pharmacists.
- •Which version you mean changes the answer on training, BAAs and citations.
- •ChatGPT for Healthcare pricing is not published, and its Epic plugin is not available on individual Clinicians accounts.
Sources and citations
OpenEvidence licenses journal content: NEJM Group titles from 1990 onward since February 2025, JAMA and 12 other JAMA Network journals since June 2025, and Springer Nature and Sage content announced in 2026. ChatGPT for Clinicians and ChatGPT for Healthcare show cited evidence with titles, journals and dates. For ChatGPT search more broadly, OpenAI says citations can be incomplete, outdated or incorrect.
- •Whichever tool you use, open the cited article and check that it supports the specific claim.
- •Licensed full text can make citations more reliable, but it does not guarantee the answer applies to your patient.
Accuracy claims are self-reported on both sides
OpenEvidence announced a perfect score on a USMLE sample in August 2025, after excluding image questions and disputing one answer key, and reports benchmark results for its newer models. OpenAI says physicians rated 99.6% of 6,924 ChatGPT for Clinicians responses safe and accurate, and reports results on HealthBench, a benchmark it built with physician-written rubrics and grades with its own model. We found no independent replication of either set of claims.
- •Vera Health’s published benchmarks are also vendor-run and not peer-reviewed.
- •The most useful test is your own: run the same non-identifying questions through each tool and check every citation.
Patient data, BAAs and training
OpenEvidence offers a business associate agreement to US covered entities and says protected health information is not used to train its models. On consumer ChatGPT plans, content may be used for training unless you opt out; ChatGPT for Clinicians conversations, and Business and Enterprise data by default, are not. OpenAI’s BAA covers only the features it lists as eligible.
- •An individual click-through agreement may not satisfy your employer: MaineHealth, for example, tells staff not to enter PHI into OpenEvidence.
- •Ask your compliance team which tools are approved before entering any patient details.
Where Vera Health fits
Vera Health is the top-ranked tool in our 2026 clinical AI evaluation. It retrieves from a corpus the vendor puts at more than 60 million papers, guidelines and care pathways before answering, and adds a drug database, interaction checks and more than 900 calculators in one app. It is free for licensed clinicians and trainees, and the vendor says it is available outside the US, which makes it a practical option where OpenEvidence is excluded and ChatGPT for Clinicians is not offered.
- •Tradeoff: its terms describe it as a reference library and ask users not to submit patient-specific data for diagnosis.
- •Enterprise EHR integration and security documentation are custom-priced and need IT coordination.