Units
1. Introduction to Artificial Intelligence in Midwifery: Roles, Boundaries, Validation and Ethics 2. Pregnancy Follow-up and Antenatal Monitoring: History, Appointment and Reminder 3. Risk Early Warning: Preeclampsia, Gestational Diabetes and Danger Signs 4. Mother-Baby Educational Material Production: Brochure, Video Script and Visual Instruction 5. Prenatal Counseling and Empathic Communication: Difficult Conversations 6. Breastfeeding and Postpartum Consultancy 7. Clinical Documentation: Midwife Note, Partograph Summary and Handover 8. Evidence Scanning and Guide Reading: Accessing and Verifying Current Information 9. Privacy, KVKK and Data Security: Protecting Pregnant Data 10. Adapting for Cultural Sensitivity, Multilingualism and Low Literacy 11. End-to-End Workflow, Quality Control and Personal AI Protocol
Unit 8 / 11

Evidence Scanning and Guide Reading: Accessing and Verifying Current Information

Gains:

  • Ability to use AI not as a source of evidence, but as an assistant for guide simplification and search roadmapping
  • Knowing that artificial intelligence can fabricate non-existent articles and sources and being able to verify every reference and claim from the original source
  • Ability to minimize the risk of hallucinations with an approach of 'process this text I found' instead of 'give information'

Midwifery is an ever-changing field of knowledge. Yesterday's truth may have been updated in today's guide; Evidence of a practice strengthens or is abandoned over time. A good midwife is one who can see the difference between "this is how it's always been done" and "current evidence says so." Evidence-based practice (basing a care decision on current, reliable research findings and guidelines) is the quality assurance of midwifery. AI is very helpful here for two things: simplifying and understanding a complex guide or article, and providing a roadmap of where to look when researching a topic. But here is the most critical limit: AI is not a source of evidence; It is an assistant on the path to the source.

In this unit we will learn to use AI like a "literature assistant", but never to directly rely on the "fact" claims it produces. AI can summarize a study to you; But you verify from the original source whether that study exists, what it says, whether it is up-to-date or not.

Beware: Large language models can spoof non-existent articles, authors, journals, or even fake "DOIs" (a publication's digital identification number). This is the most insidious form of hallucination in evidence screening; because a made-up bibliography looks real. Do not use any reference given by an AI without first finding the original source.

What does AI do well and what can it never do?

What they do well (leads):

  • Translating a guide/article you have into plain language and extracting its main points.
  • To suggest with which keywords and in which source types (guide, systematic review, professional organization publication) you can search on a subject.
  • Marking parts of a text that are relevant to your question.
  • Explaining a complex statistic (e.g. "relative risk", "confidence interval") in everyday language.

What they can never do (not a substitute for resources):

  • To produce current and accurate evidence with confidence "from one's own memory". The data on which the AI ​​is trained is frozen at a specific date; later and updates may not be there.
  • To show correctly the evidence on which a claim is based (can make up a reference).
  • To guarantee the most up-to-date version of the guides.
  • Knowing whether an application complies with the applicable protocol in your country/institution.

Step by step: secure evidence scanning

  1. Clarify the question: "Which practice is recommended in the current guideline in this case?" Pose a concrete question like:
  2. Use AI for a roadmap: Ask which sources to look at, with which keywords.
  3. Go to the original source: Ministry of Health guidelines, recognized professional organization publications, reliable databases — open the source itself.
  4. Have AI summarize the source you have: Paste the text you found and say "summarize based on this text only". So the AI ​​doesn't make things up, it just processes the document in front of you.
  5. Currentness and context control: Self-check the date of the resource and its compliance with your institution's protocol.
  6. Decision: The application decision still belongs to the midwife/physician; Evidence becomes an input into the decision, it does not replace the decision.

The fourth step — “summarize your own text to the AI” — is the golden rule of this unit. When you use AI as a processor of the text you have, rather than as a source of information, the risk of hallucinations is much reduced.

three mini cases

Case 1 — Simplifying the guide: A midwife cannot find time to read an up-to-date 60-page antenatal care guide. He gives the PDF of the guide (public, official document) to YZ and says, "Based solely on this document, summarize the main topics and practical changes that concern the midwife." A 20-minute job turns into 5 minutes. Then he confirms the three critical points in the summary one by one from the relevant pages of the guide. AI reduced the reading load; The midwife guaranteed accuracy.

Case 2 — Fake bibliography capture: A second midwife tells the AI ​​for a presentation, “give me 5 recent studies on this topic.” YZ lists 5 studies by title-author-year-journal; It's all very convincing. When the midwife tries to search for each one from its source, she finds that two of them never existed. He throws away the list and instead searches for it himself in a reliable database. Lesson: The citation given by AI is considered "non-existent" until its source is found.

Case 3 — Statistics translation: In the third example, the midwife cannot make sense of the statement "relative risk 1.4 (95% confidence interval 1.1–1.8)" in an abstract of an actual study. He pastes the expression to the AI ​​and says, "Explain this to me in everyday language, do not add new information." “The probability of the event occurring in this group is approximately 1.4 times, and this increase appears to be statistically significant,” explains YZ. The midwife understands the concept but evaluates the translation into practice in a clinical context.

Copiable templates

Role: You are the literature roadmap assistant (not the SOURCE of information).Question: [concrete clinical/practice question].Task: To research this question, (1) what TYPE of sources should I look for (guide, professional organization, systematic review, etc.), (2) with what keywords can I search, (3) what should I pay attention to (currentness, country)?Rule: CONCRETE study/MAKING citation; Just tell me where to look.

Role: You are the source summarizing assistant. Task: Summary based ONLY on the text I have pasted below. DO NOT ADD any information, numbers or conclusions that are NOT in the text. If you are unsure on a point, write "[NOT CLEAR IN TEXT]". Make a separate list of practical implications that are relevant to the midwife. Text: [paste document]

Role: Statistics/terminology translator.Task: Explain the following sentence in EVERYDAY language that a midwife would understand.ADDING new data/comments; just simplify the meaning. If it's unclear, say it's unclear. Sentence: [statistic/term]

Role: You are the verification check assistant. Task: List each EVIDENCE CLAIM and each REFERENCE in the text below; Write "[TO BE VERIFIED FROM THE ORIGINAL SOURCE]" next to each one. Do not comment; only extract items that need to be verified.Text: [text]

Weak prompt / Strong prompt

Weak: "What is the current evidence about X administration in pregnancy? Give sources."

Result: AI produces outdated "evidence" from its own (frozen and incomplete) memory, possibly with fabricated attributions.

Strong:

Role: literature roadmap assistant. What types of sources and keywords should I look for for X application during pregnancy, and what should I pay attention to in terms of up-to-dateness? CONCRETE work/citation fabrication; Just tell me where to look.

Difference: Strong prompt places AI in the role of roadmap rather than resource; It prevents you from producing fake citations and sends you to the real source.

AI's place in the chain of evidence

business

AI

midwife

Suggest where/how to call

Yes

makes the call

Summarizing the document I have

Yes (just that text)

truths

Statistics/term translation

Yes

Put it in context

Export current evidence from memory

no

Finds it from the original source

Citation/source generation

No (can make up)

He finds the source himself

Implementation decision

no

midwife/physician

Tip: When asking the AI for evidence, always switch to "process my own found text" mode. The question should not be "find me evidence" but rather "help me understand this evidence I found." This single mental change eliminates most of the risk of hallucinations.

Common mistakes

  • Mistaking AI for a source: AI is not a source of information; His memory is frozen and incomplete.
  • Using citations without verification: Fabricated bibliography is very convincing; Every reference must be from the original source.
  • Assuming up-to-dateness: AI may present an old version as "up-to-date"; You control the date.
  • Skipping country/institution context: Generic evidence may not fit your protocol exactly.
  • Not giving the actual text for the summary: If you don't give the text, the AI ​​will make it up; always say "just rely on this text".
  • Substituting evidence for decision: Evidence is input; The application decision belongs to the midwife/physician.

In summary

In evidence screening, AI is a powerful roadmap and text processor: it simplifies a guide, suggests where to look, translates statistics into everyday language. But he is not a source of evidence — his memory is frozen, he can be incomplete, and he can make up attributions. Golden rule: use AI in "process this text I found" mode, not "give information"; Verify each claim and attribution from the original source; Check yourself for up-to-dateness and compliance with the institution's protocol. The application decision always belongs to the midwife and the physician.

Application task

Find an up-to-date, official and publicly available midwifery/antenatal care guide. Summarize the source with the summary template, saying "just based on this text", then confirm the three critical points in the summary from the relevant sections of the guide. Then ask the AI ​​for a "working list" on the same topic and try to look up each citation from its source; Note how many actually exist. This experiment embodies why you shouldn't trust AI as a resource.

checklist

  • [ ] I used AI as a roadmap / text processor, not a source.
  • [ ] I gave the original document for the summary and said "just rely on this text".
  • [ ] I have verified each attribution and evidence claim from the original source.
  • [ ] I checked the up-to-dateness of the source and its compliance with the institutional protocol.
  • [ ] I consciously tested for made-up attribution.
  • [ ] I made the application decision as a result of clinical judgment, not evidence.
Previous Next