Your labs, medicines and history — read together.

A single value rarely tells the story. A low ferritin next to a low haemoglobin and a history of heavy periods points to iron deficiency. The same low ferritin next to a high CRP points to inflammation. A creatinine that looks stable next to a new medicine that affects kidney function is a different problem from a creatinine that has been stable for years.

Whole-record reasoning means reading the findings together and against the context that changes their meaning: age, sex, pregnancy status, medicines, prior results and symptoms. It is the part of clinical reasoning that a ten-minute appointment often has to compress, and it is the part a machine can help with when the material is available.

It also means knowing what is missing. A good answer says which finding needs a repeat, which needs a clinician's examination, and which is probably nothing. That is more useful than a confident paragraph that treats every value as an isolated fact.

A general chatbot is optimized to produce a fluent answer, not to verify one. It has no structured intake, so it does not systematically ask about onset, severity, red flags or medicines. It has no access to your report unless you paste it, and even then it may retype a reference range from memory instead of reading the one on the page.

The failure is quiet: the answer sounds confident because the language is confident. That is why evidence grounding matters. Premedice keeps the source visible beside the claim, states when the answer is uncertain, and escalates red flags instead of explaining them away. No AI is hallucination-free, which is why the output is for review, not automatic action.

The practical test is simple: ask for the source, and check whether it exists. If the tool cannot show its basis, treat the answer as a draft, not a finding.

Start with the timeline, not the diagnosis. Write down when the problem started, what changed, what has been tried and what the results were. Premedice can assemble that from documents and your description, which is usually the part that gets lost in a short visit.

Then list the findings that matter. A panel of thirty values is not thirty questions. Ask the tool which three or four findings deserve the clinician's attention, and why. Bring the list, not the whole report, to the start of the conversation.

Then write the questions. What could explain this pattern? What should be repeated? What would change the plan? What should I watch for before the next visit? A prepared patient gets more out of ten minutes than an unprepared one, and the clinician gets a clearer starting point.

Privacy depends on the product, not the category. The questions to ask are what is stored, where it is stored, who can access it, whether it is used for training and how you delete it. Premedice encrypts data in transit and at rest, does not use user data for model training, does not sell health data and lets you delete your record.

You can reduce the footprint before you upload. Redact your name, date of birth, medical record number and address from a lab report; keep the values, units, ranges and dates. Those fields are enough for interpretation. The privacy policy is the canonical source for retention, residency and subprocessors.