Somebody is paying nurses fifty to a hundred dollars an hour to read what a chatbot wrote and say whether it’s right.

I found the first posting in August while I was reading coach job postings for a different reason. Title: Registered Nurse.  Work: evaluate AI outputs for medical accuracy, across specialties, remote, contract, no patients anywhere in the role. I found a second one this week. It pays more. It wants someone who can catch what a trained clinician would catch across diagnosis, treatment, and patient safety, and in the same breath it says no degree is required.

Read that pairing twice. They want the judgment. They don’t want to ask what it’s attached to.

I’ve thought about those postings more than I expected to, and here’s why. Every nurse I talk to who’s building a coaching practice has the same quiet fear about AI. Either it’s going to replace what she knows, or it’s going to get her in trouble. And here is a company that looked at the whole problem and decided the one thing it couldn’t automate was a nurse’s judgment. They’re renting it by the hour.

So before we talk about what you should do with these tools, let’s be clear about what they can’t do. They can’t decide what’s inside your scope. They don’t have one.

Three documents landed between February and April, and I’ve read all three at the source, because that’s the rule around here.

The American Academy of Nursing approved a position statement on AI in health care in February that listed thirteen policy recommendations. The one that matters for you is number eleven: human-in-the-loop oversight in every AI governance policy, to preserve clinical judgment and accountability. The Academy’s own position sentence says AI should support, rather than supplant, nursing judgment. Their word, supplant.

In April, the American Nurses Association convened a think tank and published its consensus findings, identifying five risks the profession agrees are already here. Number two is unclear accountability and liability, and the report says the fear of license exposure came up again and again. The first principle everyone agreed on: AI must support, not replace, professional nursing judgment. The nurse stays the accountable decision-maker.

Closer to home, my own board’s chair wrote about AI in the Winter Bulletin. Her line: using an AI tool does not transfer clinical responsibility. The same page notes that the board doesn’t have a position statement on AI yet. So that’s a chair’s message, not a rule. I’m quoting it because it’s the clearest sentence any of the three produced, and because it’s the sentence I’d have written myself.

Even the US Food and Drug Administration, which loosened its guidance on clinical decision-support software in January, still draws its line partly on whether a clinician can independently review the reasoning behind what the software says. The regulators assume a person is reading the why, not just the answer.

Put them together and the profession has said one thing four ways. The tool can be in the workflow. The nurse stays in the judgment. Accountability doesn’t move.

Every one of those documents pictures a nurse inside a system. A hospital deploying a monitoring tool. A health system procuring a documentation product. A governance committee.

Not one of them pictures you at your kitchen table at ten at night, asking a chatbot to draft a client email, a guide, a caption, or a session summary. Nobody’s procuring anything. There’s no committee. There’s you, a tool, and your license.

That gap is where this blog lives. Because the principle doesn’t change when the room does. If accountability doesn’t transfer to a monitoring system in an ICU, it doesn’t transfer to a chatbot in your kitchen either. If a tool wrote it, you still signed it.

Here’s what I’ve learned, partly the hard way. The failure you’re braced for is the obvious one. Cure your thyroid. Stop your medication. That version is easy to catch, and most of us would never publish it.

The version that gets through looks careful.

Hedged language. A screening question. A line telling the reader to check with her provider. Those are the exact features your eye uses to decide a document is responsible. So when a tool produces them, your review passes. You’re checking for recklessness, and the page is performing responsibility.

The specific trap is what I've started calling the disclaimer finding. Here's what I mean by that. When a piece of wellness content needs a contraindication warning, that warning isn't protecting your reader. It's evidence about the content. It's telling you that whatever the page recommends can't safely go to a stranger until somebody screens her first, and screening isn't yours to do.

Take the kidney example. A wellness guide has no business asking anyone whether she has kidney disease. The moment the page asks, it's claimed the authority to screen, and then it owes her something in return: what to do if the answer is yes. It never delivers that. It just asks, and leaves her holding a question she can't answer on her own.

​The fix isn’t writing a better warning. It’s taking out the thing that required one.

If you choose to use them, I’m not going to tell you not to. Here’s the line I’d hold:

Those postings are the whole argument in one line. The market has already decided what your judgment is worth. It’s the one thing in the AI pipeline that needs a license.

So no, the tool isn’t going to replace you. But it will happily write something you’d never say, in a voice that sounds like yours, with a disclaimer that makes it look careful. And then it walks away, because it has no license to lose.

You do. Which is why you read every page like a nurse before it goes out under your name.​

The three questions I run on every draft, mine or a tool’s, are free.
Get the Three-Question Scope Check here