Gains:
- Ability to use AI to organize scattered notes into structured headings (e.g. SBAR) and mark missing areas
- Ability to recognize the risk of artificial intelligence fabricating an unobserved finding and verify each sentence line by line with the 'spread, fabricate, leave blank' rule
- Ability to apply clinical evaluation, signature and responsibility always remains with the midwife
Clinical Documentation: Midwife Note, Partograph Summary and Safe Handover
The most time-consuming but least visible task of midwifery is documentation — writing down everything that is done. Midwife note, labor follow-up, handover (patient transfer from one midwife to another), referral letter, discharge summary... These documents are not just records; It determines the quality of the next care, safe communication between the team and legal assurance. A poorly kept note puts a real patient at risk. Artificial intelligence is a powerful assistant in this field: it organizes scattered notes into an organized structure, marks missing fields, puts a handover into a standard framework — but the midwife is always responsible for the accuracy and content of the document.
In this unit, we cover how to both speed up and secure documentation with AI. Critical principle: AI shapes, marks missing; The midwife enters and approves the content. AI cannot “make up” a finding that has never been observed—which, in documentation, is the most dangerous form of hallucination.
Attention: A midwife's note should not include any findings that the AI added but that you did not observe. The AI might add a sentence like “fetal heart sound is normal” because it sees it as “normal”; If you did not write this, that sentence is a lie and should be deleted. The note is a record of what was done, not what was expected to happen.
Why structured documentation is safer
Free text—writing as it comes to mind—seems fast but is dangerous: an area will be missed, a finding will remain unclear, someone else will misunderstand the note. Structured documentation (standard headings, fixed fields, common abbreviations) reduces these risks because each document sits on the same framework and missing fields are immediately visible. This is the most powerful contribution of AI: it distributes a scattered observation into standard headings and warns that "this heading is empty".
For example, a common framework for a handover is SBAR: Situation (situation — who is the patient, what is happening now), Background (background — history, relevant history), Assessment (assessment — midwife's clinical assessment), Recommendation (recommendation/request — what should the transferee do, what should he pay attention to). AI fits your notes into these four headings; but the clinical evaluation in the "Assessment" section is yours.
Step by step: producing a safe midwife note
- Anonymous raw note: Write down your observations as they occur; Do not enter any identification information.
- Structure it: Ask the AI to break down the memo into standard headings (e.g. Story / Findings / Actions / Plan).
- Prohibition of fabrication: In the prompt, write "DO NOT ADD any findings that are not in the note; leave the blank space [BLANK]."
- Missing check: Have the AI list areas that are expected to be in a standard note but are not in your note.
- Reading and approval: Read the note line by line; Verify that each sentence is something you actually observed.
- Signature and time: Add midwife name, date-time; The document is official only when it is approved.
The heart of these six steps are the third and fifth items: the prohibition of fabrication and reading line by line. Even minor errors in documentation, acceptable in all other areas, are dangerous here because a document becomes the basis for decisions by other healthcare professionals over time. A "normal" statement not written by a midwife but added by artificial intelligence may cause someone else not to want that examination again months later. So it's much safer to force the AI to always "leave incomplete" in documentation than to allow "fill in"; An empty space appears and is filled consciously, while a made-up sentence silently persists.
three mini cases
Case 1 — Handover safety: During a shift change in a busy ward, 8 patients are transferred and sometimes a detail is missed in the verbal transfer. The midwife has the AI transcribe the anonymous notes for each patient into SBAR format, reads and corrects the output. In three months, "missed information" incidents decrease significantly; Each cycle is done completely, with the same four titles. The time also decreases from ~5 minutes to ~2 minutes per patient.
Case 2 — Fake finding: A second midwife had AI edit her raw note and saw “urinalysis normal” on the printout — but she didn't even order the test. The AI assumed "usual flow" and made it up. The midwife deletes the sentence. This is a typical example of why the fabrication ban prompt and line-by-line reading are essential: without verification, a false finding would become a permanent record.
Case 3 — Partograph summary: In the third example, the midwife will create a transfer summary from the partograph records of a long labor. He gives the time-value notes to the AI and says "just write a timeline summary from the values I gave, don't add comments." AI produces a regular summary; The midwife adds the clinical comment (evaluation of the course of action) herself. AI arranged the data, the midwife made the decision.
Copiable templates
Role: You are the clinical documentation formatting assistant.Task: Distribute the following anonymous raw midwife note under the following headings:[Story] [Today's findings] [Actions] [Plan/Follow-up].STRAIGHT RULE: DO NOT ADD any findings, analysis, values that are NOT in the note.Write "[BLANK]" in the blank heading. Adding comments/decisions.Raw note: [note]
Role: You are the SBAR handover assistant.Task: Translate the following anonymous note into SBAR format:[S-Situation] [B-Background] [A-Evaluation: MIDWIFE WILL FILL IN][R-Suggestion/Request].Rules: Leave "A" heading BLANK; I will write the clinical evaluation. Do not make up new information. Note: [note]
Role: You are the missing-area examiner.Task: Compare the midwife note below with the fields EXPECTED to be in a standard antenatal/puerperal note and list only the MISSING fields.Rules: Interpret the content, do not say true/false; Just count the missing items as "there is no such title". Note: [note]
Role: You are the partographer/track data summarizer.Task: Write a straight TIMELINE summary from the following time-value records (time -> measured values). ONLY use the values I give. Rules: DO NOT FIT values, DO NOT ADD trend/decision COMMENT. I will add the clinical comment. Records: [records]
Weak prompt / Strong prompt
Weak: "Correct and complete this note."
The word "complete" is dangerous: AI can fill in the gaps it sees as missing. The result is an unreliable document containing unobserved findings.
Strong:
Role: documentation formatting assistant. DISTRIBUTE the following anonymous note under the [Story/Findings/Achievements/Plan] headings. DO NOT ADD any information that is not in the note; write the blank title [BLANK]. Adding comments/decisions.
Difference: The powerful prompt says "deal" instead of "complete", explicitly prohibits fitting, and marks the empty space. The document remains reliable.
Document type and role of AI
document
AI's job
midwife's job
midwife note
Distribution to titles, incomplete marking
Content, accuracy, approval
SBAR handover
cast into form
Rating (A), approval
Partograph summary
Writing data regularly
Clinical interpretation, decision
referral letter
draft format
Clinical justification, signature
Discharge summary
Simplification, order
Accuracy, instruction confirmation
Tip: In the documentation always force the AI output to "leave incomplete", not "fill in". Seeing an empty [EMPTY] space is a thousand times safer than a made-up sentence; Because you see the gap and fill it consciously.
Common mistakes
- “Make it complete” means: invites the AI to make up; Say "deal" and "leave blank".
- Confirming without reading line by line: A fabricated finding can only be detected by careful reading.
- Giving grades with identity information: De-identification is also a must in documentation.
- Having the clinical evaluation written by AI: The "A" of SBAR and partograph interpretation belongs to the midwife.
- Forgetting to add signature/time: An unapproved document is not official.
- Substituting the AI summary for the original: The summary is helpful; The original record is preserved.
In summary
In clinical documentation, AI formats and flags gaps; The midwife enters and approves the content. Structured frameworks (headings, SBAR) make documents more secure, and AI quickly fits scattered notes into these frameworks. The biggest danger is that AI will fabricate an unobserved finding; So ban fabrication at the prompt, mark the blank space, and verify each sentence line by line. Clinical evaluation, signature and responsibility always belong to the midwife.
Application task
Take a handover (anonymous) from your own shift and format it with the SBAR template; Fill in the "A-Assessment" heading yourself. Then lay out a raw midwife note with the “scatter, make up, leave [BLANK]” pattern and check line by line to see if the AI has made up. If you find a fabrication/omission, take note — this is a learning example to share with your team.
checklist
- [ ] I wrote the "make up, leave blank, distribute" rules in the prompt.
- [ ] I read the output line by line and verified that each sentence was actually observed.
- [ ] I wrote the clinical evaluation (SBAR-A, partograph interpretation) myself.
- [ ] I used a raw note with no identifying information.
- [ ] I added the midwife's name, date-time and confirmation to the document.
- [ ] I did not replace the original recording with the AI summary; I kept both.