Most discharge handouts are written at a 12th-grade reading level. The average American reads at a 7th-grade level. AI fixes that — if you control the prompt and verify the dosing.
Generate plain-English patient handouts at a 6th-grade reading level, including multilingual versions, with a verification protocol that catches dosing or medication errors before the handout reaches the patient.
The Agency for Healthcare Research and Quality (AHRQ) and the U.S. Department of Health and Human Services have for years recommended that patient-facing health materials be written at a 6th-grade reading level or lower. The actual published average across hospitals is closer to grade 12. The gap matters: the National Assessment of Adult Literacy estimates that more than a third of U.S. adults have basic or below-basic health-literacy skills, and in those populations, complex handouts measurably increase readmission rates and decrease medication adherence.
This is exactly the kind of writing AI does well. Models like Claude and ChatGPT can rewrite a dense, jargon-heavy paragraph into short, concrete sentences in seconds. The risk is that they will do this confidently even when the underlying clinical content is wrong — a hallucinated dose, a swapped drug name, a missed contraindication. The workflow below treats AI as a writing assistant only. Every clinical fact is verified against your facility's MAR, the patient's discharge orders, or a pharmacist before the handout goes home.
Notice that this prompt produces structured output (the same five sections every time), pins the reading level, and explicitly asks the model to flag uncertainty. Those three properties — structure, target grade, and uncertainty flagging — are what make AI-drafted handouts safe to use as a starting point.
You are a patient-education specialist working with a registered nurse.
Produce a one-page handout with these five sections, in this order:
1. WHAT IS [condition]?
- 2-4 short sentences, plain English, 6th-grade level.
- No medical jargon. If a term is unavoidable, define it inline.
2. YOUR MEDICATION
- Drug name (generic AND brand, both in plain text).
- Dose written as a sentence: "Take 1 pill (10 mg) every morning."
- With food / on empty stomach / time of day if relevant.
- Common side effects in patient-friendly language.
- The single most important thing to NOT do.
3. FOLLOW-UP
- When to follow up and with whom (PCP, specialty clinic).
- What to bring (pill bottles, BP log, food diary).
4. WARNING SIGNS — CALL THE DOCTOR IF...
- 4-6 specific, observable warning signs.
- Use patient-observable language: "your ankles get more swollen"
not "increased peripheral edema."
5. WHO TO CALL
- Daytime number, after-hours number, when to call 911.
Hard rules:
- Target Flesch-Kincaid grade level: 6.0 or below.
- Sentences under 15 words where possible.
- One idea per sentence.
- If you do not know a specific dose, frequency, or follow-up
interval, write [VERIFY WITH NURSE] in that spot. DO NOT GUESS.
- After the handout, output a "FACTS THAT NEED VERIFICATION"
list summarizing every clinical claim you made.
Topic: [condition + meds + follow-up plan, in your own words]
After you have a draft, paste it back with this follow-up prompt:
Compute the Flesch-Kincaid grade level of the handout below. If it is above 6.0, rewrite the highest-grade sentences using shorter words and simpler sentence structure, and report the new grade. Do not change clinical meaning. [paste handout]
The Civil Rights Act Title VI and Section 1557 of the Affordable Care Act both require meaningful access for patients with limited English proficiency in any program receiving federal funding. Most hospital systems contract with translation vendors for legally binding documents, but for routine teach-back handouts, AI translation is increasingly acceptable when reviewed by a fluent speaker.
Translate the patient handout below into [target language] at a reading level appropriate for an adult with basic literacy. Do not translate drug names — keep generic and brand names in their original form. Translate the rest. After the translation, list any words or phrases where the target language has multiple acceptable translations and you chose one. The nurse will need a fluent reviewer to confirm. [paste English handout]
Practical guidance. Spanish, Mandarin, Vietnamese, Tagalog, and Arabic are the most commonly requested languages in U.S. hospitals based on Census ACS data. Modern frontier models handle all of them well for patient-education content. They are weaker on idiomatic regional dialects (e.g., Caribbean Spanish vs. Mexican Spanish vs. Argentine Spanish). Always have a native speaker on staff, on the interpreter line, or in the family review the translation before it goes home.
This is the section that protects your license. AI hallucination on medication facts is rare but real, and the cost of a single error is severe. The protocol below is mandatory before any handout you generate goes to a patient.
Most facilities require patient handouts to include the facility logo, a teaching-note reference, and standardized disclaimer language. Ask your patient-education department or compliance officer for the boilerplate; if there is none, the version below is a reasonable template that you can adapt with their approval.
This handout is a reminder of what we talked about today. It is not a complete list of everything you need to know about your condition or medications. If you have any questions, call the number above or talk to your nurse, doctor, or pharmacist. Reviewed by: [Nurse name, RN] Date: [yyyy-mm-dd] [Hospital name] · [Department] · [Phone]
Disclaimer. AI-drafted handouts are starting points only. Every clinical fact must be verified against the patient's orders, the package insert, or a pharmacist before use. Always check with your facility's compliance officer, patient-education team, and state board of nursing before adopting any AI workflow in patient communication.