Nurses: Turn a Discharge Summary Into a Patient Handout in 5 Minutes (Safely)

15 min read By FindSkill Team

44% of nurses now use AI, and patient education is the top use. The 5-minute de-identified workflow — reading level, red flags, and the nurse sign-off.

Table of Contents

Nurse AI use has nearly tripled in a year. The 2026 State of Nursing report — a survey of 2,240 US nurses released July 7 — puts adoption at 44%, up from 15% last year. And tied for the #1 thing nurses use it for, alongside charting: creating patient education materials (37%).

That makes sense. Writing a discharge handout is exactly the kind of task that eats twenty minutes you don’t have, and exactly the kind of task AI drafts well. But the same report has a second number that should give you pause: 83% of nurses say AI output is rarely or never reviewed before clinical use.

So this post is the workflow for doing the popular thing properly: a discharge summary in, a patient-ready handout out, about five minutes — with the two safety habits that make it defensible. One protects your patient’s privacy. The other protects your patient.

Why the raw AI draft isn’t patient-ready

If you paste clinical text into ChatGPT and ask for “a patient handout,” you’ll get something that looks impressive and fails the person it’s for. The research is specific about how:

It reads at college level
Readability studies put unprompted AI patient answers around a 13th-grade reading level — against the 6th-grade standard. About 36% of US adults read at basic or below-basic health literacy.
It skips the red flags
Studies of AI discharge instructions found they often omit the warning signs and 'call 911 if' guidance — the content patients most need.
It can invent medications
In an npj Digital Medicine study, 18% of AI-drafted discharge outputs had safety issues — including 3% that introduced medications nobody prescribed.
readability what breaks in an unprompted AI handout safety

None of this makes AI useless for patient education — it makes the unedited draft unusable. Every one of those failures is fixable with a better prompt and a two-minute review. That’s the workflow.

The rule before the workflow: nothing identifiable goes in

This part is non-negotiable, and it comes before any prompt. Consumer chatbots — ChatGPT, Claude, Gemini — are not HIPAA-covered tools. Whatever you paste leaves your control. So the input is never the chart, never the actual discharge summary, never anything with a name, date, MRN, or address on it.

The practical version of HIPAA’s de-identification standard, for this task: describe the clinical situation, not the patient. “A 68-year-old going home after a heart failure admission, new to furosemide, low health literacy, lives alone” contains everything the AI needs to draft a handout — and nothing that identifies anyone. If you’re about to paste text you didn’t type fresh, stop and strip it: names, any date more specific than a year, locations smaller than a state, record and account numbers, contact details. When in doubt, leave it out — the handout doesn’t need it.

And check your employer’s AI policy first. Some systems provide an approved internal tool (in which case, use that one); some prohibit consumer chatbots for anything clinical. This workflow assumes you’re cleared to use one with fully de-identified input.

The 5-minute workflow

Discharge summary → patient handout
De-identify clinical situation only — no names, dates, numbers
Prompt 6th-grade level + red flags + teach-back
Verify every med and dose against the chart
Sign off your edit, your name, then it prints
The AI drafts; the nurse verifies and signs off — that order is the whole safety model

Step 1 — Describe the case, de-identified (one minute). Type it fresh rather than pasting from the chart: condition, key meds by name, the two or three behaviors that matter at home, and anything about the patient’s situation that should shape the writing (reading comfort, language, lives alone, caregiver involved).

Step 2 — Use a prompt that fixes the known failures (thirty seconds). This one targets the reading level, forces the red-flag section, and keeps the AI away from inventing clinical content:

Write a discharge handout for a patient going home after [condition].
Medications to cover: [list — names and purpose only, exactly as I give them].
Key home instructions: [2-3 behaviors, e.g., daily weight, low-salt meals].

Rules:
- 6th-grade reading level. Short sentences. Direct instructions
  ("Weigh yourself every morning"), not explanations of physiology.
- Use ONLY the medications and instructions I listed. Do not add
  medications, doses, or medical advice beyond what I gave you.
- Include a "Call your doctor today if..." section and a separate
  "Call 911 if..." section — I will fill in the specific warning
  signs, so leave clear placeholders for me to complete.
- End with 3 teach-back questions a nurse could ask to check
  understanding.
- Format: short headers, bullet points, room to write in appointment
  details by hand.

The “use ONLY what I listed” line matters most — it converts the AI from a medical reference (which it must never be here) into a rewriter of content you provided. The placeholders in the warning-signs section keep the highest-stakes content in your hands, where it belongs.

Step 3 — Verify like it’s your signature on it (two minutes) — because it is. Read the draft once as a clinician: every medication name against the chart, every instruction against the actual discharge orders, warning signs filled in from your unit’s standard or the provider’s note. The npj study’s 3%-invented-medications number is the reason this step is not optional. Then read it once as the patient: would your most tired, most overwhelmed patient understand every line? If a sentence makes you translate in your head, ask the AI to simplify it.

Step 4 — Sign off and use the teach-back. The handout is yours now, not the AI’s. The three teach-back questions at the end are the underrated payoff — “Can you show me how you’ll take this?” catches misunderstanding that no handout, human- or AI-written, catches on its own.

The bilingual bonus — with a back-translation check

Patient reads Spanish, Vietnamese, Tagalog? Ask for the same handout in that language — then add the verification loop, because you can’t check a language you don’t read: “Now translate that handout back into English, sentence by sentence.” If the back-translation matches what you approved, the translated version almost certainly says what you think it says. If a sentence comes back changed, fix it before it prints. Two extra minutes, and it turns “I hope this is right” into something you actually checked.

What this doesn’t replace

  • Your hospital’s approved materials. If your system has vetted handouts for the condition, those come first — this workflow shines for the gaps: the unusual combination, the literacy level your stock handout misses, the language you don’t have on file.
  • Clinical judgment on content. The AI formats and simplifies; it does not decide what a heart-failure patient needs to know. That was decided by the care team, and by you.
  • A HIPAA-compliant documentation tool. This is a writing workflow with de-identified input — not a place to process charts.

The report’s most quietly damning stat is that only 5% of nurses say they got AI training that actually prepared them. The gap between “44% are using it” and “5% were trained” is where handouts written at college level get handed to patients who read at a 6th-grade level — with nobody checking. Five minutes of workflow closes that gap for the most common use case in nursing.

If you want this exact workflow taught step by step — the de-identified case description, the three-protection prompt, the two-pass verification, teach-back, and the bilingual back-translation loop — our new course AI for Nurses: Patient Handouts in 5 Minutes starts free. For the other half of the survey’s #1 tie, AI for Bedside Nurses: Charting & Documentation covers charting, and AI for Nurses: The ANA-Aligned Routine covers the professional-standards side.

Sources

Build Real AI Skills

Step-by-step courses with quizzes and certificates for your resume