Gains:
- Ability to distinguish midwifery jobs into green (use comfortably), yellow (support, carefully) and red (artificial intelligence does not decide) zones according to risk level
- Ability to recognize the risk of hallucination and secure each AI output with the Verify-Customize-Link discipline
- Ability to adopt the principles of privacy and responsibility and apply that the clinical decision always remains with the midwife and physician.
Midwifery is a profession that brings many different jobs together in the same day. A midwife measures a pregnant woman's blood pressure and evaluates the test results in the morning, calms down an anxious expectant mother in the afternoon, corrects the breastfeeding position in the afternoon, prepares a prenatal education brochure in the evening, and writes her notes at the end of the day. Some of this work involves organizing information, producing and remembering text; Some of them are clinical decisions with a narrow margin of error that directly concern the health of the mother and baby. This entire module teaches you to clearly separate these two groups and to use artificial intelligence only in the first group, within safe limits.
Artificial intelligence (AI for short; computer systems that have been trained on vast amounts of text and can write and read in human languages) is a powerful aid to midwifery practice when used correctly. Here we are specifically talking about large language model (LLM; a type of AI that produces text by corresponding with you — for example, a chat assistant) type of tools. These tools summarize a text, simplify a brochure, compile a story, write an appointment message. But it doesn't make a diagnosis, rate a risk, or recommend a medication. The main principle we will repeat throughout the module is:
Artificial intelligence is an assistant: it generates drafts, reminds, simplifies. Diagnosis, risk decision, medication and clinical guidance are the responsibility of the midwife and physician. An unverified AI output is like an unsigned patient note; It has no binding effect on its own.
What exactly does AI speed up in midwifery?
To understand the value of AI in midwifery, let's divide things into three areas based on risk level. This distinction is the skeleton for the rest of the module.
Green zone — Use AI comfortably (text and information work): Draft of an educational brochure, appointment and preparation message to the pregnant woman, translating a complex medical text into plain language, streamlining a handover note, first translation of a multilingual information text, road map of where to look for current literature on a topic. Here AI saves time and the error of the output is easily caught when the midwife reads it and corrects it.
Yellow zone — Use AI with caution, only as support: listing risk factors in a story (as a reminder, not a decision), making a red flag checklist, “bringing to mind” possible differential situations. Here AI can act as a checklist / forget shield; But the midwife filters each item on the list with her own knowledge and the decision remains with the midwife.
Red zone — AI does NOT make decisions: Making a diagnosis (e.g., preeclampsia diagnosis), rating the risk and making a referral decision, determining medication and dosage, managing labor, emergency intervention decision. In this region, the AI output can be at most one input; can never take responsibility.
Tip: Before outsourcing a task to AI, ask yourself one question: "If this output is wrong, who will suffer, and how quickly?" If the damage occurs quickly and directly to the mother/baby, you are in the red zone; Use AI only to edit text, not to judge it.
Hallucination: Why AI seems confident but is wrong
The most dangerous feature of big language models is hallucination: the model produces information that does not actually exist, in a fluent and confident language, just as if it were true. Instead of saying "I don't know," AI can fill the gap with a believable fabrication. Concrete examples of this in midwifery include: making up a guideline statement that does not exist, confidently prescribing an incorrect medication dose, giving an incorrect “normal range” for an assay value, citing a study that has not actually been published.
The hallucination is not a "malfunction" but a natural consequence of the way these tools work; does not disappear completely. Therefore, the solution is not to rely on AI, but to verify every numerical and clinical statement from independent source. Fluency of the text is not proof of accuracy.
Verification discipline: ANC rule
Take each AI output through three steps before releasing it. Let's call it ANC for short: Verify – Customize – Connect.
- Verify: Verify every number, dose, assay value, weekly limit and guideline statement in the text from the current official source (Ministry of Health guidelines, institution protocol, reliable professional guide). Delete the sentence you cannot verify.
- Customize: Adapt the general text to this pregnant woman, this institution, this cultural context. AI writes for the average patient; There is a real person in front of you.
- Link: Link the output to the clinical context and chain of responsibility: add note "This is a draft, midwife/physician approval required", consult physician if necessary.
three mini cases
Case 1 — Brochure draft (green area): A midwife will prepare a "pregnancy nutrition" brochure for a pregnancy school with 20 students. He asks AI for a simple draft of 350 words, receives it in 8 minutes, corrects 2 sentences according to his institution's iron-folic acid protocol, and distributes it. If he had written it by hand, it would have taken 45 minutes; AI reduced the time by about 80% and the risk of error was low because the midwife verified every number.
Case 2 — Fake dose (shift into the red zone): Another midwife asked YZ "what is the daily dose of iron in iron deficiency anemia in pregnancy?" he asks. AI gives a confident number. When the midwife is about to tell this directly to the pregnant woman, she stops and looks at the institutional protocol; He sees that the range given by the AI does not match the protocol. The dosage decision belongs to the physician; The midwife does not use the AI output at all. Rule worked: numerical/clinical output not used without verification.
Case 3 — History collection (yellow zone): The third midwife has AI summarize a messy 12-page pregnancy file (omitting identifying information) and asks for a list of “salient risk factors.” AI produces 6 items. The midwife reads the list, realizes that 1 item was misunderstood, removes it, and adds 1 missing item. The list is not a decision, but an input into the midwife's evaluation.
Copiable templates
You can adapt the four templates below to your own practice. What they all have in common: rewriting the role, limit and verification requirement into the prompt.
Role: You are a junior assistant writing for an experienced midwife.Task: Write a SIMPLE informational draft for pregnant women on the following topic.Topic: [topic]Rules:- DO NOT make a diagnosis, recommend medication/dose, risk stratify.- DO NOT WRITE any numbers, doses or guide sentences that you are not sure about; leave a label "[MIDWIFE VERIFIED]" instead. - Language: 6th grade level, short sentences, explain medical term at first occurrence. Length: 300 words or less.
Scan the following AI output with a "verification eye". Output: [paste text]Just give me a list of:1) Every NUMBER, DOSE, WEEK, and NORMAL RANGE statement in it (separately).2) Every sentence that references the guideline/protocol.3) Every claim with definitive ("always", "never", "definitely") language.These are the items I will verify by HAND; adding comments.
Role: You are the text simplification assistant. Task: Rewrite the medical text below in a way that a high school graduate expectant mother can understand. DO NOT CHANGE meaning, ADD or REMOVE information. If you cannot simplify a term, explain it briefly in parentheses. Text: [text]
This is a draft handover note; does not involve clinical judgment.Organize the following scattered notes by headings:[Story] [Findings today] [Actions] [Reminders for follow-up].No new information is FAILED; If there is a missing field, write "[BLANK]". Raw note: [note]
Weak prompt / Strong prompt
<w:tcPr><w:tcW w:type="dxa" w:w="2880"/></w:tcPr><w:p><w:r><w:rPr><w:b/></w:rPr><w:t>Weak prompt
Powerful prompt
text
"Give information about iron during pregnancy"
"Role: midwife assistant. Write an informative outline for pregnant women, at 6th grade level, of maximum 250 words. Recommend dosage/medication; mark the number you are not sure of as '[MIDWIFE VERIFY]'."
Conclusion
Unsourced, difficult-to-verify text containing doses and claims
A safely editable draft with clear boundaries and marked areas to be verified.
At the weak prompt, the AI is released; It easily enters the red zone. In the powerful prompt, the role, limit, verification tag and length are defined from the beginning; The output remains under the midwife's control.
Ethics and privacy: two rules from the start
Starting from the first unit, two ethical principles apply throughout the entire module. The first is privacy: never write identifying information such as the pregnant woman's name, TR ID number, phone number, file number, address to the AI (details in unit 9). Second, transparency and accountability: Do not present AI-generated text as written by a human; Every statement that enters a clinical document is responsible for the midwife who approved it.
Caution: AI writes according to the cultural "average"; therefore, it may overlook or misassume some social, religious or local sensitivities. The midwife always makes the cultural adaptation (details in unit 10).
Common mistakes
- Asking the AI for a "decision": "Should I refer this pregnant woman?" Questions like these are the red zone; You cannot ask AI.
- Using numbers without verifying them: Leaving each number, such as dose, week, analysis interval, in the text without confirming it is the most common and dangerous mistake.
- Writing identification information: Any information that makes the pregnant woman recognizable should not be included in the prompt.
- Confusing fluency with accuracy: Just because the text is written well does not mean that its content is correct.
- Not writing down the role and limit: Unlimited prompts produce unlimited (and risky) output.
- Relying on a single output: On critical texts, it is useful to re-run the AI as a second “verification eye” and scan its own output.
In summary
Artificial intelligence is a powerful assistant that accelerates text and information work in midwifery; but it does not replace decisions such as diagnosis, risk, medication, and clinical referral. Divide tasks into green (use comfortably), yellow (support, carefully) and red (decision-making) zones. Verify each numerical and clinical statement from independent sources due to the risk of hallucinations; pass each output through the Validate–Customize–Link discipline. The principles of privacy and responsibility apply throughout the module: the decision is always up to the person.
Application task
Choose 6 of your own weekly tasks and make a chart placing each in the green/yellow/red zone. Use the brochure template above to produce an outline for a business in the green zone; then scan your own printout with the "verification eye" template and mark the items that need verification. For a job you put in the red zone, write in two sentences why you won't use AI.
checklist
- [ ] I placed the job in the green/yellow/red zone; If it is red, I did not use AI for the decision.
- [ ] I wrote the role, limit (no diagnosis/dose) and verification tag in the prompt.
- [ ] I have independently verified each number, dose, week, and guideline statement in the printout.
- [ ] No identifying information was entered into the prompt.
- [ ] I went through the steps of Validate–Customize–Link the output.
- [ ] I confirm that I am responsible for the statement entered in the clinical document.