AI can be genuinely useful for learning about health — and it can also sound confident while being wrong, incomplete, or simply not qualified to make the decision you're asking of it. Paste the question you're about to ask an AI, and in one click see how much caution the answer deserves, plus a safer way to ask.
Runs entirely in your browser. Your question is never sent to a server or stored.
No tool required. Run any AI health answer through these seven questions before you trust it.
This tool is educational and is not medical advice. It uses simple pattern-matching to gauge how much caution an AI answer deserves — it cannot diagnose you or judge whether any specific AI answer is medically correct. If you think you may be having a medical emergency, call your local emergency number (911 in the US) now. Always confirm health decisions with a qualified professional.
Short, evidence-based answers from Dr. Gily Ionescu, MS MD — an internal medicine physician and nutritionist.
It varies widely by task. Large language models often handle textbook-style knowledge and general explanations well, and studies have shown they can pass medical licensing-style exams — but passing an exam is not the same as safely managing a real person. Accuracy falls when a question depends on your individual history, when the topic is niche or fast-moving, or when the model must reason across several steps. Treat AI as a well-read but unverified study partner, not a clinician.
Yes — and often while sounding completely confident. It can miss a red-flag symptom, suggest a dose that is wrong for your age or kidney function, overlook an interaction with a medication you already take, or reassure you when you actually need to be seen. Confidence in the wording tells you nothing about whether the content is correct.
Run it through four questions: does it cite what kind of evidence it is based on; does it admit uncertainty rather than sounding certain about your body; would a clinician need information the AI does not have, such as your history and medications; and are you about to act on it? Reading to learn is low-stakes, but changing a medication, delaying care, or self-treating is where checking matters most.
Yes. Hallucination is the well-documented tendency of language models to produce fluent, plausible statements that are simply untrue — including invented studies, fake citations, wrong drug doses, and guidelines that do not exist. Because the fabricated parts read exactly like the accurate parts, you cannot spot them from tone; you have to check the underlying source.
Ask it to cite the type of source behind each claim — a major clinical guideline, a systematic review, a specific study — then follow those citations to confirm they exist and actually say what the AI claims. A phrase like “studies show” with no source, or a citation you cannot find, is a signal to distrust that claim. Evidence-based answers also separate what is well-established from what is uncertain or debated.
For general education it is a reasonable starting point, as long as you verify anything you act on. For anything personal — interpreting your symptoms, dosing a medication, deciding whether to seek care, or questions involving pregnancy or a child — it should not be your decision-maker. The safest use is to understand the topic and prepare better questions for a professional who can examine you and knows your history.
Trace each important claim back to a primary source: a named guideline body such as the USPSTF or a specialty society, a peer-reviewed study, or an authoritative reference — and confirm the source really supports that specific claim. Cross-check anything surprising against a second reputable source, and when a claim would change what you do, confirm it with a pharmacist, nurse line, or physician before acting.