Something genuinely interesting is happening inside the NHS right now, and it’s not getting nearly enough attention. Across England, a growing number of trusts are quietly plugging AI-powered symptom checkers and triage tools into GP surgeries and 111 call centres, watching what happens, and trying very hard not to make any bold promises about it. I’ve been following this patch for a while, and the early picture is genuinely fascinating, part hopeful, part messy, and very human.
The core idea behind NHS AI triage is simple enough. You describe your symptoms to a chatbot or an automated phone assistant, it processes what you’ve said, and it routes you to the right level of care, book a GP appointment, go to A&E, wait and see, or call 999 right now. In theory, it takes pressure off overstretched receptionists and 111 handlers, and gets patients to the right place faster. In practice, the story is considerably more complicated.

Which trusts are actually running these pilots?
Several NHS trusts have been running structured pilots since late 2024, with the bulk of activity expanding through 2025 and into this year. NHS England has been working with suppliers including Babylon Health (in various reincarnations), Kry, Doctorlink, and Huma, among others. The NHS England digital tools programme has published frameworks for evaluating these systems, though the rollout timelines have shifted repeatedly.
North East London Integrated Care Board ran one of the more closely watched 111 AI trials, where an automated symptom-assessment layer was added before patients reached a human call handler. In parts of the West Midlands, GP practices have been testing front-door triage apps that patients fill in before their appointment is confirmed. Bristol, North Somerset and South Gloucestershire ICB trialled a symptom checker integrated directly into the NHS App. The results, where published, are modest but not nothing: one reported a roughly 14% reduction in unnecessary A&E attendances in the trial cohort over a six-month window. That’s not transformative, but it’s movement.
What patients are actually saying
I’ve read through a fair few patient feedback threads, NHS patient panel notes, and a couple of published qualitative studies. The split is real. Younger patients, particularly those comfortable with apps, tend to like the speed. No sitting on hold for 40 minutes. No being made to feel like you’re wasting someone’s time. You type in your symptoms at 11pm and the system tells you whether it’s worth going in tomorrow or whether you should be calling an ambulance now. That’s genuinely useful.
Older patients and those with complex, long-term conditions are often much less enthusiastic. One recurring complaint in patient panel notes is that the chatbots struggle with multimorbidity, the very common situation where someone has three or four overlapping conditions and the symptom they’re describing only makes sense in context of everything else going on. A 72-year-old with heart failure, type 2 diabetes, and mild cognitive impairment describing breathlessness to a symptom checker is not a simple triage case. The AI tools currently deployed are, by most accounts, not reliably good at this.

There’s also a legitimate concern about digital exclusion. The accessibility problems that come with tech-first services crop up everywhere, and healthcare is no different. Patients without smartphones, those with poor literacy, and those whose first language isn’t English all face real barriers with current AI triage tools. The NHS has a legal duty under the Equality Act 2010 to ensure services are accessible, and some campaigners argue that rolling out AI-first triage without robust human fallbacks puts trusts in a difficult position legally, quite aside from the ethics.
The clinicians aren’t entirely sold either
GPs I’ve seen quoted in BMJ coverage and GP magazine articles are, at best, cautiously curious. The British Medical Association has been vocal about the risks of AI tools being used to reduce clinical capacity rather than support it, essentially, a concern that trusts will use AI triage to justify employing fewer human staff, rather than freeing up existing staff to do more complex work.
There’s also a diagnostic risk question that doesn’t go away however many efficiency figures you throw at it. AI symptom checkers are trained on datasets. Those datasets have gaps, biases, and historical blind spots, the same problems that have been documented in medical AI research for years. A tool that was primarily trained on data from one demographic group may perform less accurately across others. The NHS hasn’t been particularly transparent about the training data behind the tools it’s deploying, and some clinical informaticists think that needs to change before wider rollout.
My reading of the evidence so far is that these tools probably do some useful work at the lower-acuity end of triage, the “should I go to A&E or can I wait for a GP appointment” type of question, for a generally healthy adult describing a straightforward symptom. The problems mount quickly once you move away from that narrow use case.
Where does NHS AI triage go from here?
The government has been keen to push AI adoption across public services, and the NHS is clearly a priority. The risk of AI over-reach in public sector systems is something we’ve covered here before in a different context, and some of the same patterns show up. Efficiency pressure, a technology that works reasonably well in controlled conditions, and a rollout that moves faster than the oversight structures around it.
NHS England has committed to publishing outcome data from the main trials by the end of 2026, which should give a clearer picture. In the meantime, the trusts running pilots are supposed to be submitting regular safety reports to their integrated care boards, though how rigorously that’s happening varies by region.
One thing worth knowing if you’re a patient encountering one of these systems: you always have the right to ask to speak to a human, both in 111 and in GP reception. The AI layer is not a final decision-maker, or shouldn’t be. And if you ever feel a triage outcome doesn’t match your gut feeling about your own health, push back. Clinicians will tell you the same thing.
It’s a bit like getting an automated valuation on a property when you really need a proper surveyor, or accepting an online estimate of an EPC certificate without having anyone actually assess the building. The tool gives you a starting point, not the whole picture. NHS AI triage is exactly that right now: a starting point, with a lot of human work still needed around it.
I find myself genuinely curious about where this lands in three years. The potential is real. So are the risks. Watching this particular experiment play out is going to be one of the more important health tech stories in Britain for a while yet.

Leave a Reply