How I prompt AI like a nurse
Nurses have used SBAR for decades to hand off the most important information in someone’s life without dropping any of it. It turns out AI needs almost exactly the same thing.
Jump to the templateWhat SBAR actually is
When a nurse needs to call a doctor at two in the morning, there is no room to ramble. SBAR is the format we use so nothing important gets lost: situation, background, assessment, recommendation. Four parts, in that order, every time.
It has been in hospitals for decades because it solves a specific problem. The person on the other end of that call has none of the context you have. They can’t see the patient, they weren’t in the room, and they need enough to make a decision in about ninety seconds.
That is very close to the situation you’re in every time you open a blank AI chat. It can’t see anything. It wasn’t in the room. Most people type one line and then feel let down by what comes back, and honestly, that’s a handoff problem more than an AI problem.
The four parts, translated
Situation: what you’re trying to do right now
One or two sentences, no history yet. In the hospital this is the reason for the call. In a prompt it’s the actual task, stated plainly.
Here’s the thing worth noticing: for most people, this is their entire prompt. That’s the whole message. Then the answer comes back generic and they decide AI isn’t that useful. It’s the first line of four.
Background: everything it can’t see
Who you are, who this is for, what the constraints are, and how much room you have. Length, deadline, format, budget, what tools you actually have in front of you.
The piece almost everyone leaves out is what you already tried and what was wrong with it. That single line saves you two rounds, because otherwise it hands you back the thing you already rejected.
Assessment: what you already think
This is the hard one, and I want to be honest about that rather than pretend it isn’t. In nursing, assessment is where you stop reporting facts and say what you believe is actually going on. New nurses report. Experienced nurses assess. It is genuinely the hardest letter to learn, and it’s the one that makes the whole thing work.
In a prompt, it’s your read. What you suspect the real problem is, what you think would probably work, where you think you’re stuck. It matters because it gives the AI something to agree with or argue against instead of starting from nothing.
If you genuinely don’t have a read, say that. Write “I don’t know what the right approach is here, that’s part of what I’m asking.” That’s still an assessment. It tells it where you are, which is the entire point of the letter. Don’t skip it and don’t make something up.
Recommendation: what you actually want
SBAR does something clever at the end. It forces the nurse to make an ask instead of just reporting and waiting. You don’t call to describe a problem, you call to say what you want to happen.
Same here. Not “help me with this email,” but “write me a short email to my landlord, four sentences, firm but not rude.” Name the thing, the format, and the length. Vague in, vague out.
The two nursing habits that matter more than the letters
Make it ask you questions before it starts. In the hospital we don’t just take an order and walk away, we read it back and confirm it before anything happens. You can ask AI for the same thing, and it’s the single most useful line I’ve added to how I prompt: tell it to ask you whatever it needs to know before it produces anything. It will surface the thing you forgot to mention, which is usually the thing that was going to make the answer wrong.
Real time feedback, in the same conversation. Nursing is not a set of instructions handed over once. You watch what happens, you adjust, you correct in the moment. Do that here instead of starting a fresh chat every time something comes out wrong. Tell it specifically what’s off and ask it to change only that. One fix at a time works far better than six at once, which usually gets you a different answer rather than a corrected one.
The template
SITUATION What I'm trying to do: [One or two sentences. The task itself, not the history.] BACKGROUND What you can't see from where you're sitting: - Who I am and who this is for: [ ] - Constraints that matter: [length, format, deadline, budget, tools I have] - What I've already tried, and what was wrong with it: [ ] ASSESSMENT What I already think: [Your read on it. What you suspect the real problem is, or what you think would work. If you honestly don't have a read yet, write exactly that: "I don't have a read on this yet, and that's part of what I'm asking."] RECOMMENDATION What I want from you: [The specific thing you want made. Include the format and roughly how long.] BEFORE YOU START Ask me anything you need to know before you begin. One question at a time, and wait for my answer before asking the next. When you have enough, tell me your plan in two or three lines and start only after I say go. AS WE GO When I tell you something is off, change that one thing and leave the rest alone. Don't start over from scratch unless I ask you to.
You don’t have to use the headings. The order is what does the work.
Three honest things
SBAR isn’t magic. It works because it makes you say out loud what your situation is and what you actually want, and that’s the step nearly everyone skips. If you don’t know what you want, no framework is going to save you. The framework just makes it obvious sooner.
Don’t use it for everything. A quick question doesn’t need a handoff. I use this when the answer matters, or when I’ve already asked once and gotten something useless back. Most of my messages are still one line.
The impressive answer and the useful answer are not always the same answer. The first time I tried this I asked for a tool to help me get better on camera, and what came back was genuinely clever and completely not what I wanted. The only reason I could tell the difference was that I knew what I had asked for. That’s really what all of this is for.
Guides go out to the newsletter before they go anywhere else. No hype, no daily emails, just the things that held up when I actually used them.