Lesson 1: What AI Can and Can't Do in Care Work

Welcome, and one promise up front

Hello, and welcome to the course! Before we start, here's one thing you'll hear from me again at the very end.

Everything in this course is general information, not clinical advice. Your local policy and your own clinical judgement always come first. If something here doesn't match what your trust, practice or care provider says, follow your workplace. Agreed? Lovely.

So here's our first question. What can an AI assistant actually do for someone in your job, and what should it never do?

What is an AI assistant, really?

You've probably heard of tools like ChatGPT, Gemini, Claude or Microsoft 365 Copilot. You type a request and they write back in fluent, natural language.

The easiest way to picture one is as a very well-read new starter who has never met a patient. They've read millions of pages, so they write quickly and sound sure of themselves. But they've never stood at a bedside or watched someone deteriorate, and they don't know your patient.

They also don't look facts up the way you'd check a formulary. They work by predicting which words are likely to come next. That's why the writing flows so well. It's also why it can be wrong.

Confident doesn't mean correct

This is the most important idea in this level, so let's take it slowly.

The assistant isn't trying to tell you what's true. It's producing text that sounds right. Most of the time the two match. Sometimes they don't, and the wrong answer comes out in the same calm, polished tone as the right one. People call these made-up answers "hallucinations".

Here's a real example. In April 2026 the Royal College of Surgeons of England reported a study in its Annals. Researchers asked chatbots typical surgical questions. With some models, between 25% and 34% of the references given were fabricated or couldn't be verified. Some fakes had believable titles and invented web links, and some were credited to well-known institutions.

Think about that for a second. Would you spot a fake reference that looks like it came from a respected organisation? Most of us wouldn't, not without checking. And that's the habit we'll build in this course: you check, every time.

Where AI can genuinely help

The good news is that a lot of your working day involves words and admin, and that's where these tools shine. Three everyday examples:

1. Drafting non-clinical writing, such as an email to the team about a training session, a rota reminder or a poster for the staff room.
2. Turning approved information into plain language, such as rewording a leaflet you've already been given so it's easier to read. You check it against the original afterwards.
3. Helping you learn, for example by summarising a guideline so you can revise it, quizzing you before a study day or helping you organise your thoughts for a reflection.

In each case the AI writes a first draft and you stay the expert who checks it.

Where AI must never make the decision

Some things must stay with a human, and in particular with you and your team. Three clear examples:

1. Clinical decisions, such as whether to escalate, how to respond to an observation or what a patient needs right now.
2. Medication decisions, such as doses, timings or whether something is safe to give.
3. Judgements about a person's safety or wellbeing, such as safeguarding concerns, capacity, or how someone is really doing today.

Could an AI offer an opinion on any of these? Probably. Should it decide? No. It can't see the patient and it doesn't know the context. It also can't be accountable, and you are. Your professional standards stay with you whatever tool you use.

There's one more rule we'll cover properly next time: never paste identifiable patient information into a tool your workplace hasn't approved. For now, just keep that in mind.

How is the profession actually using AI?

Let's look at some evidence, and pay attention to the dates.

• May 2026, global: Elsevier's Clinician of the Future 2026 report included its first Nurses Edition. It surveyed 2,757 clinicians across 118 countries. 41% of nurses said they use AI regularly, compared with 57% of physicians. Only 42% of nurses saw AI tools as trustworthy, but 61% thought AI would improve care within 5 to 10 years.
• Survey run July to October 2025, UK: The Health Foundation surveyed just over 2,000 NHS staff and 8,000 members of the public. 57% of staff thought AI would improve the quality of care, compared with 38% of the public. The share of staff who thought technology would make care worse rose from 6% in 2024 to 19% in 2025.
• July 2026, UK regulation: The NMC confirmed a consultation on a new draft Code, running from September to December 2026. It includes proposals on the safe and effective use of digital technology, including AI.

So why do the dates matter so much? This field moves fast. A survey from two years ago may describe tools that no longer exist, and guidance can change within months. So when you read any statistic about AI, ask yourself three things: when was it collected, who was asked, and is there something newer?

Your turn

Take two minutes with a notepad. Write down three tasks from your last shift where AI could have helped with the writing. Then write three where it must never have the final say. Keep the list, because we'll build on it.

Key takeaways

• This course is general information, not clinical advice. Always follow local policy and your clinical judgement.
• AI predicts likely words. It can sound confident and still be wrong, even inventing references.
• It's useful for admin, plain-language drafts and learning. You check everything it produces.
• Clinical decisions, medication decisions and judgements about someone's safety stay with humans.
• Always check the date and source of any AI statistic. Things change quickly.