THE WOMEN’S HEALTH TECH REPORT: How to Talk to AI — Part 1 of 3
You Are Getting Bad Answers Because You Are Asking Bad Questions
The quality of what AI gives you is almost entirely determined by how you ask. In women’s health, that gap between a good prompt and a bad one is not an inconvenience -- it is a clinical risk.*
The Technology
Prompt engineering is the practice of structuring questions and instructions to AI systems in ways that produce more accurate, more useful, and more appropriate responses. It is not a programming skill. It does not require technical training. It is, at its core, the art of communicating precisely with a system that interprets language literally and has no access to the context you have not provided.
Every clinician and every patient who has typed a question into ChatGPT, Claude, Gemini, or any other AI chatbot has been doing prompt engineering -- almost always without knowing it, and almost always doing it poorly. The difference between a well-constructed prompt and a poorly constructed one is not subtle. It routinely determines whether the AI gives you a useful clinical framework or a confident-sounding answer that is wrong in ways that are not immediately obvious.
The Clinical Application
The appeal is straightforward. AI systems contain an enormous amount of medical knowledge. They can synthesize literature, explain mechanisms, draft patient education materials, summarize guidelines, and help clinicians think through differential diagnoses. For a field like obstetrics and gynecology, where the evidence base is vast, the guidelines are frequently contested, and the clinical questions are often complex and time-pressured, a well-functioning AI assistant could be genuinely valuable.
The problem is that most clinicians and patients interact with AI the way they use a search engine -- typing a short, vague question and expecting the system to infer what they actually need. Search engines are designed to handle that kind of query. AI language models are not. They respond to exactly what you ask, filling gaps with plausible-sounding content that may or may not reflect your actual clinical situation.
A clinician who types “what is the management of preeclampsia” will get a generic, textbook-level answer that may not reflect current evidence, may not account for gestational age, and will not ask whether the patient has comorbidities, what her blood pressure trend looks like, or whether she is already on magnesium. A clinician who types “I have a 32-week patient with new-onset blood pressure of 158/102, proteinuria of 2+ on dipstick, and a headache that started this morning. She has no prior hypertension. What does current ACOG guidance say about inpatient versus outpatient management, and what are the criteria for delivery at this gestational age?” will get a response that is specific, actionable, and worth reading critically.
The same question. Completely different outputs. The difference is the prompt.
This is not a trivial distinction in women’s health. Prenatal care, labor management, gynecologic oncology, and reproductive medicine all involve clinical questions where the wrong answer -- confidently delivered -- can delay appropriate intervention, falsely reassure a patient, or lead a clinician down a management path the evidence does not support. The stakes of bad prompting in a clinical context are not the same as the stakes of bad prompting when asking AI to plan a vacation.
The Women's Health Tech Report: Safety analysis, the evidence critique, and the verdict are below -- for subscribers who want the full picture.
