ChatGPT Prompts for Nursing Notes and Medical Charting

The short version: ChatGPT can speed up nursing notes and medical charting, but only if you strip out every identifying detail first and treat the AI as a drafting tool for structure and wording, never as somewhere you paste patient data. Used this way, a nurse can shave real minutes off every note by feeding ChatGPT a de-identified template and letting it handle the formatting, the SBAR structure, or the tidying up of shorthand into full sentences. Used the wrong way, with real names, NHS numbers, or dates of birth typed into the box, it's a data protection incident waiting to happen.

Why this topic makes people nervous, and why that's correct

Most guides to ChatGPT prompts for nursing notes skip straight to the fun part: here are twenty prompts, copy and paste, job done. They quietly assume you'll type in the patient's actual vitals, actual history, actual name. Nobody says that part out loud because it sounds obvious once you say it, but it's exactly what happens in practice. A nurse on a busy ward, trying to finish charting before handover, pastes in "Mrs Patterson, 78, admitted with UTI, confused, BP 98/60" because that's the fastest way to get a clean note out the other end. That single sentence, typed into a free consumer ChatGPT account, is a HIPAA problem in the US and a UK GDPR problem under the Data Protection Act. The US Department of Health and Human Services is clear that anything identifying a patient's health status counts as protected health information, and a public AI chatbot is not a covered entity you have a business associate agreement with.

So the uncomfortable bit, the part most "10 best prompts" posts leave out, is this: the prompt itself isn't the risky part. The risk is what you paste alongside it. If you want to use ChatGPT for charting safely, the entire skill is learning to anonymise first and prompt second. That's a different habit to build than just memorising clever phrasing, and it takes longer to explain, which is probably why most articles don't bother.

What ChatGPT is useful for in charting

Once you accept the anonymise-first rule, ChatGPT becomes handy for a narrow set of jobs:

• Turning rough shorthand ("pt c/o abd pain 6/10, no N/V, afebrile") into full, grammatically correct sentences for the permanent record.
• Restructuring loose notes into SBAR (Situation, Background, Assessment, Recommendation) format for handover.
• Drafting the skeleton of a care plan from a de-identified set of symptoms, which you then edit and confirm clinically.
• Writing patient-facing discharge summaries in plain language, again from de-identified input, that you personalise afterwards.
• Checking your own documentation for missing elements, like whether you've recorded pain score, intervention, and reassessment in a pain management note.

What it is not useful for, and should never be asked to do, is clinical judgement: deciding whether a presentation needs escalation, suggesting a diagnosis, or recommending a medication change. That's not a caution for legal cover, it's just true. ChatGPT has no access to the patient, no vital signs feed, no lab values beyond what you type, and it will confidently produce a plausible-sounding answer even when it's wrong. Treat it as a very fast typist with decent grammar, not a colleague.

The five-step anonymise-first method

This is the bit worth following, step by step, every single time:

1. Strip every direct identifier before you type anything: no name, no initials that could identify, no NHS number, no exact date of birth, no bed number, no ward name if the ward is small and identifiable.
2. Replace specifics with generic placeholders: "Patient A," "78-year-old," "admitted three days ago" instead of an actual admission date, "a long-term care facility" instead of naming the facility.
3. Round or generalise anything that could narrow the patient down to one person, such as an unusually rare diagnosis combined with a specific age and location. On a small rural ward, "the only patient on dialysis this week" is identifying even without a name.
4. Write your prompt around the placeholder version, get the draft back from ChatGPT.
5. Re-insert the real identifying details yourself, by hand, directly into your hospital's electronic health record system, never by copying ChatGPT's output straight into the chart without that manual re-identification step.

That fifth step matters more than people think. It forces a pause where you re-read the note before it goes into the permanent record, which catches AI phrasing errors too, like ChatGPT occasionally inventing a plausible-sounding but wrong abbreviation expansion.

A worked example: a medical-surgical ward at shift change

Say you run a 24-bed medical-surgical unit and your nurses are documenting handover notes at the end of a 12-hour shift, with maybe 6 patients each. One nurse has a patient recovering from a hip replacement who's been slow to mobilise and a bit confused overnight. Instead of typing "Mr Davies, 82, bay 4, hip replacement day 2, confused overnight, family worried," she types this into ChatGPT:

"Write an SBAR handover note for an 82-year-old patient, post-operative day 2 following hip replacement, who was more confused than baseline overnight, with family expressing concern at the bedside. Vitals stable, afebrile, pain controlled with current regimen. Keep it to four short sections, clinical tone, no speculation about cause."

ChatGPT returns a structured SBAR draft in seconds: Situation, Background, Assessment, Recommendation, each two or three lines, phrased the way a chart reads. She then copies that structure into the real EHR entry, adds the patient's actual name, bed number, and exact observations, and reviews every line before saving. That whole exchange, from opening the prompt to pasting the edited version into the chart, realistically takes under two minutes once you've done it a few times. Across six patients on a single shift, that's a plausible saving of ten to fifteen minutes of documentation time, not because ChatGPT is thinking for her, but because it removes the blank-page problem of turning clinical shorthand into formatted prose.

Ten prompts worth keeping on hand

Each of these assumes you've already anonymised the details, following the method above. Swap the bracketed placeholders for your de-identified version each time.

• "Turn this shorthand into a full SBAR note: [de-identified shorthand]."
• "Rewrite this nursing note in plain, formal clinical English without changing any clinical content: [de-identified note]."
• "Draft a discharge summary section for a patient with [de-identified condition and plan], written for a patient with limited health literacy."
• "Check this pain management note for completeness: does it include pain score, intervention given, and reassessment time? [de-identified note]."
• "Convert these vital signs and brief observations into a narrative nursing note: [de-identified data]."
• "Write a care plan template with headings for Problem, Goal, Intervention, Evaluation for a patient with [de-identified general condition, e.g. 'reduced mobility post-surgery']."
• "Shorten this note to fit a 100-word charting field without losing the key clinical facts: [de-identified note]."
• "List the standard elements my unit's falls-risk documentation should include, generically, not tied to any patient."
• "Rephrase this family communication note to sound neutral and factual, removing any emotional language: [de-identified note]."
• "Create a generic shift-handover template with sections for new admissions, changes in condition, and outstanding tasks, with no patient data filled in, that I can reuse and fill in by hand."

Notice that the last one produces a reusable template rather than a one-off note. That's the better long-term habit: build three or four templates once, in a document you keep for yourself, and reuse them by hand every shift rather than going back to ChatGPT for every single patient. It's faster and it removes the temptation to paste real data in under time pressure, which is when mistakes happen.

Where employer policy fits in

Plenty of hospitals and care providers, both NHS trusts and US health systems, have explicit policies banning the use of public AI tools for anything touching patient data, precisely because of the identification risk above. Before you build any of this into your routine, check your trust's or employer's IT and information governance policy. Some organisations are rolling out their own licensed, contracted versions of AI tools with real data agreements in place, which is a completely different risk profile to a personal ChatGPT account. If your employer has one of those, use it instead of the free public version, full stop. If they don't, and you're using a personal account, the anonymise-first method above is the minimum, not a nice-to-have.

This is also where the broader small-business AI advice applies even in a clinical setting: the businesses and teams who get real value from AI are the ones who build a repeatable, checked process rather than improvising prompts under pressure. The same thinking that applies when a dental practice sets up patient communication prompts without ever pasting patient data applies directly here. Different clinical setting, identical rule: templates and placeholders, never real identifiers, in the box.

Frequently asked questions

Is it legal to use ChatGPT for nursing notes?

It can be, as long as you never type in anything that identifies a real patient, such as a name, date of birth, NHS or medical record number, or a combination of details specific enough to point to one person. The moment identifiable health information goes into a public AI tool without a business associate agreement or equivalent data protection arrangement, it becomes a compliance issue under HIPAA in the US or UK GDPR and the Data Protection Act in the UK.

Will ChatGPT write a diagnosis or clinical recommendation for me?

It will try, because it's built to give a plausible answer rather than say "I don't know enough," but that answer has no clinical weight and shouldn't influence care decisions. Use it for formatting, phrasing, and structuring notes, and leave every clinical judgement to you and your team.

How much time does this save?

On a typical shift where a nurse writes 15 to 20 notes, even a modest saving of one to two minutes per note from skipping the blank-page problem adds up to roughly 20 to 30 minutes across a shift. The bigger saving usually comes from building a handful of reusable, de-identified templates once rather than prompting fresh every time.

What should I do if my employer bans AI tools for documentation?

Follow the policy. Many health systems are introducing their own contracted AI tools with real data agreements, and those are a different and generally safer option than a personal ChatGPT account. If no approved tool exists yet, it's worth raising it with your information governance team rather than working around the policy quietly.