You do not understand a technical term in a report. Two documents give different details. You need to ask your doctor about a recommendation but do not know how. Begin with the precise point that is unclear. You do not have to label the entire report correct or incorrect because of it. The goal can be an ordinary explanation after which you know what to ask at the consultation.
Separate checking a document from deciding on care. In the supplied material, AI can point out differing dates, units, dosage stated in two places or lists of preparations. But comparing text alone is not enough to decide what you should actually take.
First an accurate transcription, then an explanation
If you are working with a photograph or scan, check transcribed names, numbers, units, decimal separators and negations against the original. ‘No signs’ must not disappear, and an older finding must not become today’s. Remove unnecessary identifiers and consider whether you are willing to entrust the document to this service. Uploading is not a requirement. You can ask a general question or transcribe only the necessary section.
Explain this document clearly. First separate what it literally states from your explanation. Preserve dates, units, doses and uncertainties; mark illegible parts. Then list specific contradictions or unclear points worth asking a healthcare professional about. For each, quote the relevant passage. Do not claim that a contradiction automatically means a treatment error, and do not change treatment.
The result may be a precise question: ‘The medication list gives a different detail from the final recommendation. Which applies?’ Different findings may come from another time, method or purpose of examination. Ask about the cause of the specific difference. Such a question helps more than a general verdict that the report is wrong.
When you need to check a professional claim
If the model can search, ask for a current professional source and the specific part supporting the claim. Open the link and compare the title and content with the response. A study existing is not enough: it must concern the question you are addressing. If the assistant cannot open the source, it should say what it did not find. You can ask about unfamiliar words at this stage too; accurate evidence serves you better when you understand it.
A useful prompt is:
Check this specific claim: [sentence]. Look for current professional guidance or a primary study. State what the source actually supports, whom it applies to and what cannot be inferred from it for this case. If you cannot open the source, do not claim you verified its content. Prepare a question for a doctor or pharmacist, not a new dosage.
In an experiment with 1,298 adults and ten health scenarios, models alone identified possible diagnoses more successfully than people using them. Participants with AI did not choose the correct next healthcare step better than the control group. Errors arose both in communicating information and interpreting it.[1] We therefore need to check the whole process, including what we supplied to the model and how we read its answer.
When you have already changed treatment based on AI
Contact the prescribing doctor as soon as possible; a pharmacist can also help you understand how to use a particular medicine. Give the medicine’s name, prescribed dosage, what you actually took or missed, when, and what difficulties you have. Have the packaging or patient leaflet ready. You can say openly that the change came from a chatbot’s advice.
Do not make another automatic change based on this book. Returning to the original dose or making up missed doses may not be the right approach for every medicine; the specific circumstances matter.[2] If you suspect an overdose, have serious symptoms or have intentionally put yourself at risk, call 155 or 112 in the Czech Republic, rather than waiting for an ordinary consultation.[3][4]
Do not start or stop medication or change its dosage on a chatbot’s recommendation. Treat automatic meal plans just as cautiously when you have eating difficulties, as well as instructions that can reinforce body checking, restricting food or self-punishment. Agree on a specific approach with a qualified professional.
Sources and notes for this chapter
Bean, A. M. et al. Reliability of LLMs as medical assistants for the general public: a randomized preregistered study. Nature Medicine, 9 February 2026. Article. Relevant primary text and summary reviewed 12 September 2026. Simulated health situations with 1,298 adults; outcomes of the model alone and human use must not be conflated. The publisher records a subsequent correction; detailed percentage results are not given here. ↩︎
NHS Specialist Pharmacy Service. Advising on missed or delayed doses of medicines. Professional material. Reviewed 11 September 2026. The approach depends on the medicine and circumstances. This guide does not adopt a universal dosing scheme and refers readers to a doctor or pharmacist. The wording after a treatment change requires clinical review before publication. ↩︎
NHS. Poisoning. Health information. Reviewed 11 September 2026. Supports seeking urgent help for possible poisoning and not waiting for symptoms. British emergency numbers are not carried into the Czech text; Czech contacts are given by the emergency-help source. ↩︎
Czech Ministry of the Interior. Important Contacts, 72 Hours project. Emergency numbers and their purposes: 72h.gov.cz/cs/dulezite-kontakty. Verified 10 September 2026. ↩︎