HarnessHealthStart a pilot

How do I check if what an AI told me about my health is accurate?

Six checks you can do yourself, in about the time it takes to read the answer twice. Then two tools here that do part of the work in your browser, with nothing sent.

In danger now: call 911. Crisis: call or text 988.

The checks, by hand

  1. 1.

    Warning signs come first.

    Chest pain, trouble breathing, weakness on one side, thoughts of suicide, or someone who will not wake up: call 911, or call or text 988. Do not stop to check an AI answer.

  2. 2.

    Ask where it came from.

    Ask the AI to name the source for each claim: a medicine label, a government health page, a medical society guideline, or your own record. If it cannot name one you can open, treat the claim as unchecked.

  3. 3.

    Read the source yourself.

    Open it and find the sentence the answer relies on. Check that it says the same thing, applies to someone like you (age, pregnancy, other conditions), and is current. A source that does not contain the claim does not support it.

  4. 4.

    Compare with your own record.

    Numbers, doses, dates, test results and medicine names should match your record exactly: your patient portal, the after-visit summary, or the label on the bottle. A summary that changes a number or drops a word like "not" is wrong.

  5. 5.

    Look for what was left out.

    Allergies, other medicines, a recent hospital stay, kidney or liver problems. An answer can be accurate sentence by sentence and still miss the one thing that matters for you.

  6. 6.

    Bring it to the person who knows you.

    Show your clinician or pharmacist the AI answer and the source together, and ask: is this right for me? They decide; the AI does not.

Tools that do part of it

These tools use fixed rules, not a model, and run in your browser; nothing you paste is sent or stored. They check what they check and nothing else. Not medical advice.

How do I check if what an AI told me about my health is accurate? | Harness Health