Evidence is more than an answer on a screen. It includes the source, the circumstances in which information was collected and the transformations applied afterwards. When those details disappear, a tidy record can look more certain than the underlying conversation.
The aim is not to give every reviewer more text. It is to make the relevant material easy to inspect, with the original context available when needed. NICE's medicines optimisation guidance makes clear that computerised decision support should not replace clinical judgement. A sensible workflow preserves that boundary in its everyday controls.
1. Keep a usable link to the source
For imported information, record where it came from and enough detail to identify the original record. For a conversation, retain an authorised source reference and the relevant context. If somebody links an excerpt to a question, keep that link visible rather than leaving only the resulting answer.
Source material and a reviewer's note serve different purposes. The source records what was supplied; the note records a considered interpretation or decision. Editing the note should not silently rewrite the source. Where corrections are needed, make the correction and its reason understandable to another authorised colleague.
2. Preserve uncertainty as information
Unknown, not asked, declined and unclear are different situations. They may need different next steps, so give them different labels. A person who declines a question has not supplied the same information as someone who gives a clear negative answer.
Do not use a completed conversation, a full transcript or a high technical confidence score as a substitute for complete evidence. A recording can be clear while the meaning is ambiguous. A structured answer can be present while its source is weak. The workflow should help the reviewer notice those distinctions.
Make gaps readable. An explicit missing response tells the reviewer something useful. Filling the gap with an assumption removes that information.
3. Show where interpretation enters
Turning a conversation into a structured answer involves interpretation. Summarising several answers involves another layer. Make those steps visible: what is a quotation, what is a summary and what has been confirmed by a professional? Avoid presenting them in a way that makes their status indistinguishable.
Keep the question and relevant surrounding context alongside the selected evidence. An isolated phrase can lose a qualification or correction. Reviewers should be able to reject an unsuitable link and record a better interpretation without making the earlier source disappear.
4. Design for checking, not just reading
A review screen should help a professional move from the summary to the evidence and back without losing their place. Show gaps, outstanding questions and the current revision before sign-off. Use clear labels for saved drafts and reviewed notes so that an unfinished interpretation cannot look final.
Professional review needs an explicit action and the appropriate authority. Record who made the decision, when it happened and which information it covered. If relevant evidence changes afterwards, make the changed state visible and use the locally agreed review process. A previous signature should not imply that later material was considered.
Access to evidence should follow the organisation's approved responsibilities and information governance arrangements. Being useful for review does not mean that every team member needs access to every source. The approved retention and access rules belong in the evaluation from the start.
5. Test the evidence chain with imperfect examples
Use fictional examples containing a correction, a declined question, an ambiguous phrase and a missing response. Ask a reviewer to explain the recorded answer from its source. Then change an interpretation and check that the reviewed note, history and next steps still tell a coherent story.
Hannah Health's local development workflow retains traceable evidence and requires explicit human review. It does not turn a transcript into automatic clinical sign-off. These controls are being evaluated; they are not evidence of patient-use approval, clinical effectiveness or accepted production integrations.
Read the approach to clinical oversight for the current boundary. For the wider process, explore a clearer medication review workflow; for what happens after review, see follow-up and filing.