Mission
Our tools connect everyday people to clinical trial findings.
No personal health data stored here
We don’t store anything you paste. Submissions run through the evidence layer in the moment and aren’t persisted against any identity, or sent anywhere a third party can re-identify you. If you opt into our concierge follow-up, the email you provide is used only to send your audit back to you — nothing else.
No medical advice
We don’t tell you what to do. We surface what clinical trials actually said about people in situations like yours — and where the evidence runs out — so you and your clinician can decide. This is not a substitute for clinical care.
Think Carfax for used cars — but for the medical decisions you’re about to make.
Where NoBSmed fits in AI-assisted healthcare
NoBSmed is a consumer-facing evidence-verification layer for AI-assisted healthcare.
Clinicians are getting increasingly powerful AI evidence tools. Consumers are still left to navigate medical claims across doctors, AI chatbots, care portals, and online health content — with no clear way to verify what is actually supported by clinical evidence.
The new reality — the AI-human-hybrid doctor: AI lets clinicians review medical literature roughly 100x faster, but it also hallucinates. Doctors don’t always have time to verify each AI-assisted claim, and patients usually don’t know which advice was AI-generated.
A confidence gap follows: both the doctor and the AI deliver advice with certainty — but the patient hasn’t gotten the same 100x upgrade in research depth. You can’t easily push back on confident-sounding advice without evidence of your own. Asking the right question — for someone like you — is half the battle, and that’s where most patients get stuck.
For used cars, Carfax gives you red flags. For buying a home, an appraiser gives you red flags. For the medical advice you’ve been given — by ChatGPT, by Claude, by your doctor, by a discharge summary, by a care-plan note — who gives you the red flags?
That’s the gap NoBSmed fills. We are not a chatbot doctor. We are a claim checker: give us a medical claim or piece of advice plus your personal context, and we check whether the clinical evidence actually supports it for someone like you — flagged with the same four red flags we used in our HealthBench audit: hallucinate, overgeneralize, overlook, misweighted.
NoBSmed fills that gap by helping patients turn medical claims into evidence-backed questions they can bring to their doctors.
Under the hood: a medical evidence knowledge graph that AI agents can reason over with structured, deterministic queries — combining knowledge-graph construction with LLM-driven semantic parsing of clinical literature, canonically resolved to the SNOMED CT ontology.
Two kinds of audit — and why both are broken
There are really two ways to audit medical AI, and each breaks in its own way.
Either way, the principle is the one the AI-safety field keeps landing on: don’t trust an AI’s output — have an independent checker verify it against the evidence. That’s the role No B.S. Med plays for medicine, built as a transparent, open audit layer. (See where we fit in the landscape.)
| The patient one-off | The systematic audit | |
|---|---|---|
| Question | “Is this advice or denial right for me?” | “Are the AI tools and benchmarks physicians rely on actually good?” |
| Tool today | ChatGPT — reachable, fast, contextual | Rare — mostly vendors grading themselves to their own quality bar |
| The break | Ungrounded — fluent, but paraphrase and sometimes fabricated citations | Gated & self-interested — the systems are closed, and the benchmarks themselves have errors |
NoBSmed works on both sides. For patients, we ground the one-off check — tying an AI answer back to cited clinical studies, matched to your context. For the field, we run independent audits of the benchmarks and tools medicine is starting to trust — like our HealthBench audit, which flagged 29 possible patient-harm issues across 1,298 claims.
Why not use ChatGPT directly?
OpenAI grades every flagship release against its own HealthBench medical benchmark — criteria written by 262 physicians across roughly 48,000 rubric points. As of 2026, the picture across their published numbers:
| Model | HealthBench variant | Score |
|---|---|---|
| GPT-5 | HealthBench Hard | 46.2% |
| GPT-5.5 | HealthBench Hard | 31.5% |
| GPT-5.5 | HealthBench Consensus | 95.6% |
| GPT-5.5 | HealthBench Professional | 51.8% |
| ChatGPT for Clinicians (GPT-5.4) | HealthBench Professional | 59.0% |
| Physicians (baseline) | HealthBench Professional | 43.7% |
Original May 2025 paper used a single rubric-satisfaction rate: o3 = 60%, GPT-4.1 = 48%, o1 = 42%, GPT-4o = 32%, GPT-3.5 = 16%. HealthBench has since been split into the Hard / Consensus / Professional variants above.
OpenAI’s headline medical claim in 2026 is that ChatGPT for Clinicians beats physicians on HealthBench Professional (59.0 vs 43.7). That’s graded against the same answer-key methodology we audit — so the marquee medical-AI claim of the year rides on a scoring rubric whose reliability is itself measurable.
Think of it as a clinical-evidence red team of the benchmark itself: instead of testing whether a model can answer HealthBench, we test whether HealthBench’s gold answers, citations, and rubrics survive structured evidence review.
In our audit of HealthBench’s reference (“gold”) answers and rubric criteria, we audited both public variants — the May-2025-paper dataset (1,200 cited + high-stakes claims) and the newer HealthBench Professional (98 cited + high-stakes claims). Combined: 29 decision-changing findings (45 HIGH-confidence verdicts; 111 claims with evidence gaps). Not only can the AI players be wrong — the benchmark’s own reference answers and grading rubrics can be wrong, and those errors propagate into every model graded against it.
“How does NoBSmed itself score on HealthBench?”
Different task. HealthBench grades model outputs as medical advice; NoBSmed audits the evidence base those outputs cite. Scoring NoBSmed on HealthBench would mean cheating at a problem we don’t claim to solve. Full answer →
Sources: HealthBench paper (May 2025) · HealthBench Professional paper · OpenAI: Introducing GPT-5 · OpenAI: Introducing GPT-5.5 · GPT-5.5 System Card · TechRepublic: GPT-5 medical benchmarks · Vellum: GPT-5 benchmarks · BenchLM: GPT-5.5 benchmarks 2026
How is our approach different?
- We don’t give advice. We focus on filling in the gaps of what doctors and medical AI seem to miss — the fine print around clinical findings.
- Structured queries, not paraphrasing. ChatGPT probabilistically summarizes what it remembers. We extract the trial fine print — eligibility, subgroups, dose, comparator, outcomes — into typed data the system queries directly per question.
- We save you the multi-hour ChatGPT rabbit hole. Producing a defensible pre-visit summary by hand means dozens of follow-up prompts, cross-checking citations, and reading abstracts. Our system runs that audit for you in minutes, not hours.
- We extract from full-text trials, not just abstracts. ChatGPT can only paraphrase what its training corpus contained — mostly abstracts and freely-crawlable summaries. Most of the fine print that determines whether a study applies to you lives in the methods, eligibility tables, and supplementary material of full-text articles, which we extract directly under the PMC Open Access carve-out.
What’s this all about?
What
Doctors and patients make decisions using medical AI claims based on broad summaries of clinical-trial studies.
Why
Broad medical AI summaries run the risk of overgeneralizing and overlooking granular details of clinical-trial findings that are pertinent to your unique personal context.
How
We run deterministic queries to cross-check clinical-trial claims against your personal context, composed via AI elicitation.
Industry signals
Why this matters now — two recent peer-reviewed studies on how clinicians and patients are already using medical AI:
Clinician-side: OpenEvidence accounted for 98.7% of searches across leading AI-enabled clinical reference tools, with traffic rising to ~1.59 million visits/month by June 2025.1
Patient-side: In a study of 617,827 Microsoft Copilot conversations, roughly 1 in 5 involved personal symptom assessment or condition discussion. Microsoft explicitly notes that benchmark performance does not predict real-world reliability for high-stakes health questions.2
References
1 Patel VR, Liu M, Jena AB. Public Interest in an AI-Enabled Clinical Decision Support Tool. JAMA Network Open, Nov 20, 2025.
2 Costa-Gomes B, Tolmachev P, et al. (Microsoft AI). Public use of a generalist LLM chatbot for health queries. Nature Health, April 16, 2026.
Does this service exist anywhere else?
Not directly. Adjacent services handle pieces of the workflow — clinician-side evidence Q&A, patient-side AI triage, trial matching for enrollment — but none combine patient-supplied health data, a structured clinical-trial corpus, and a personalized applicability audit.
| Service | What they do | What they’re missing |
|---|---|---|
| OpenEvidence, UpToDate AI, AMBOSS AI, ReachRx | Clinician-side evidence Q&A | No patient-data ingestion, no applicability layer |
| Glass Health | Clinician CDS with FHIR context | Clinician-only — not patient-side audit |
| Hippocratic AI, Ada, K Health | Patient-facing AI agents (triage, intake, care management) | No structured trial backend |
| Deep 6 AI, TrialFit, TrialMatchAI | Trial matching for enrollment | Opposite direction — get into trials, not audit existing care |
| ChatGPT, Claude, Perplexity, Consensus, Elicit | General AI medical Q&A / paper search | No structured trial extraction, no patient-data ingestion |
| Cleveland Clinic Express Care, Mayo Clinic 2nd Opinion | Human clinician second opinions | Human-mediated, expensive, not data-driven |
Closest in shape: Glass Health (clinician-side). Closest in audience: Hippocratic AI (B2B health-system patient agents). What’s empty: the patient-side evidence-audit lane.
Read more on the Blog
Long-form essays on the problem we’re working on, the medical AI landscape, and the open-source tooling we ship for builders:
- The Evidence-to-Person Fit Problem — why matching clinical-trial findings to a specific person is hard, and what an auditable evidence layer looks like.
- The Medical AI Landscape — tools for everyday people vs. tools for medical pros.
- The After-Visit Summary Audit Landscape — three categories of "audit your after-visit summary" services and where No B.S. Med fits.
- Medical AI Developer Tooling — other MCP players and the benchmarks worth knowing.
Open-source: evidence-to-person-eval on GitHub — eval-driven design of FHIR Evidence representations, aligned with EBMonFHIR. v0.0.3, design phase.
Team
Jessica Johnson
Microbiology (MS), Science Journalism (MS)Science writer and breast cancer survivor. Keeps us creative and grounded.
LinkedIn
No B.S. Med retrieves and structures clinical-study evidence. It does not diagnose, prescribe, or replace professional medical judgment. Users should consult a qualified healthcare professional before making medical decisions.