How AI is used

Where it is used

• Reading a scanned or photographed report, when you choose to send one. The whole page image goes to the service, so redaction cannot run first — the other two paths send nothing at all, which is why they are offered first.
• Writing additional explanation beyond the reviewed content shipped with the site.

Where it is not needed at all

• A PDF your laboratory generated already contains its own text. It is read in your browser and no service sees it.
• Turning report text into a list of results is done with fixed rules, not a model, wherever the text can be read on your device.
• Deciding whether a result sits inside or outside your range, and whether it needs urgent attention. Both are arithmetic and fixed rules, and both work with no AI service at all.
• Explanations for supported markers come from reviewed content written and checked by people.

Where it is not used

Whether a result sits inside or outside the range on your report is worked out by comparing numbers, not by a model. So are the safety notices: they come from fixed, reviewed rules and from any critical flag your laboratory printed on the report. If explanation generation fails entirely, your values, ranges and safety notices are still shown in full.

It gets things wrong

Reading values off a report is the step most likely to introduce errors, which is why you are asked to check every value against your own copy before anything is explained. Rows the system was unsure about are marked, and you have to confirm those individually.

What the model is not allowed to say

Generated text is checked before you see it and is rejected if it names a condition you have, recommends treatment or a dose, estimates a risk or outcome, or tells you there is nothing to worry about. Rejected text is replaced with reviewed, written content rather than edited.

Your content is not used for training

Every service that processes report content is contractually required not to use it to train models, and receives only an opaque job identifier alongside the text. See the subprocessors page.