Tip: ask the tool what it is unsure about
A one-line prompt that turns a confident draft into a checkable one. It is not a reliability signal, but it does point you at the right paragraphs.
After the tool has given you a draft, add one line:
Which parts of this are you least confident about, and what did you have to assume?
You will usually get back a short list: a date it inferred, a name it could not make out, a gap it filled. That list is where you start checking.
Why it helps
AI output arrives in the same calm, fluent prose whether it is right or invented. CQC’s own risk list notes that AI “can hallucinate and present untrue information, authoritatively as fact”. There is no tonal cue to tell you which sentence to distrust.
Asking the question makes the tool commit to something specific about its own working, which is easier to check than the whole draft.
The limit, which matters
This is a way of narrowing where to look. It is not a reliability score.
A tool can be entirely confident and entirely wrong, and it will not list that error, because it does not know. Everything it says it is sure about still needs checking against what you know — especially names, medicines, doses and times.
Use it to prioritise your reading, never to shorten it. See how to spot the mistakes AI makes.