Speech-to-text is not the AI revolution in health and social care

Simply writing 'The wound looks better' in the electronic medical record is no longer good enough. If AI is to provide real decision support, the data must be structured from the first note.

By Trine Sønstelien, Domain Advisor at Systematic and former nurse in both Norway and Denmark. The opinion piece is published in Sykepleien.no

I've worked as a nurse in municipal health and social care. I've written 'the wound looks better' in a loose record note while rushing on to other tasks. Not because I didn't care about the documentation, but because that was what was possible in the moment.

Now I work with digital solutions for the health sector, and I still think about that sentence often. I can see more clearly now what it actually cost: not in time saved at the moment, but in everything it couldn't be used for afterwards.

There's a lot of talk about AI in health and social care right now. Decision support, early warning, predictive tools. But all of that technology relies on one thing: data. And not just any data: structured, consistent, usable data.

The real AI decision has already been made

There's a lot of talk about which AI solution municipalities should choose. But I think the most important decision has already been made, long before anyone mentioned artificial intelligence: which record system are we already using

The record system determines whether the underlying data exists at all. No AI can build decision support on data that isn't structured.

Speech-to-text is a good example. It's brilliant: I can speak instead of type, and it saves me time in the moment. But you can get speech-to-text almost anywhere today. It's no longer what separates a good tool from a mediocre one. What separates them is what happens to the text afterwards.

'The wound looks better' is no longer enough

I've written that sentence myself. Many times. It wasn't a lack of care. It was what fitted between two urgent tasks.

The problem is that the sentence can't be used for anything. It can't alert the next nurse that something has changed. It can't remind me to follow up in seven days. It disappears into a running record that the next shift may not have time to read.

This is where I think the real revolution lies: not in AI writing the text for us, but in us no longer documenting in isolated fragments to begin with.

A wound, from discovery to follow-up

Let me show you what I mean, rather than simply claim it.

I discover a pressure sore on a patient. I record it, not in a free-text field, but as structured data: wound edge, exudate, pain level. Then something happens that I wouldn't have believed possible a few years ago: the system takes it from there.

The next nurse on shift is alerted and sees the note, without having to search through a running record for every patient to spot what's changed. This may be the underrated benefit of structured data: you're shown what's relevant, instead of relying on someone having had time to read everything.

The system guides me on what to do next: what should be assessed now? From there, I'm offered suggested goals and measures, drawn from national guidance plans. Not because the system should override my professional judgement, but because I don't have to spend 20 minutes writing a care plan from scratch. The plan takes shape as I go, as a natural result of the recording I'm doing anyway.

And once the wound is documented, the system reminds me to set a follow-up date. I don't need to remember it myself. As the date approaches, I get a reminder.

This is how I think decision support should work: not as an AI chatbot I have to remember to ask, "are there any newly discovered wounds on the ward today?" and then work out the answer myself, but as an invisible structure that follows me through the nursing process, already knows the answer, and presents it to me without my having to look for it.

Why this is more than efficiency

Speech-to-text in free text saves time in the moment. Structured recording saves time in the long run, because the data can be used again, by me and by my colleagues.

But it's about more than time:

It reduces cognitive load. I don't have to remember everything myself, the system remembers for me. That's not a small thing. It's one of the reasons nurses I know have considered leaving the profession: not the workload alone, but the feeling of never being able to let go, because everything depends on remembering it yourself.

It improves patient safety. Nothing gets lost between shifts or buried in a long running record.

It makes our documentation more consistent, whether it's an experienced nurse or a newly hired care worker on the evening shift.

And it genuinely affects recruitment and wellbeing at work. A tool that supports you through a busy shift, rather than adding another layer of documentation on top, makes the profession more sustainable to stay in over time.

Not AI for AI's sake

I think we should set higher expectations for what AI actually does for us in health and social care. Speech-to-text is brilliant. But the real value doesn't come from the text being written faster. It comes from what happens to it afterwards: whether it becomes structured, usable, and part of a connected process, or just another isolated note in a long record.

Much of today's conversation about AI also starts at the wrong end. We're building chatbots you can ask, but that assumes someone remembers to ask the question. A record system with structured data doesn't need to be asked. It can present the answer before anyone thought to look for it.

Structured data beats speech-to-text every time. Not because speech-to-text is bad, but because it only solves the smallest part of the problem.