What research shows about AI health advice
People are using it. A Pew Research Centre survey of 3,488 US adults (June 2026) found:
- 34% had used an AI chatbot for at least one health reason.
- 25% had used one to work out what was causing their symptoms.
- 15% had used one to help decide whether to see a doctor.
These figures are from the US and may not reflect Australian usage, but they highlight a useful question for local triage practice: has the caller already sought advice from AI?
Accurate answers don’t always lead to accurate decisions.
A UK study published in Nature Medicine (February 2026) gave 1,298 people realistic medical scenarios to work through. The researchers then compared two ways of using the chatbots:
| How the chatbot was used | Got the right condition |
|---|---|
| Researchers entered the full case details themselves | About 95% of the time |
| Members of the public described the case in their own words | About 1 in 3 times |
People using the chatbots also chose the right next step (such as GP, ED or self-care) in fewer than half of cases. They did no better than people who just searched the internet.
Why the gap? People often left out key details, gave inaccurate information or misunderstood the chatbot’s reply. The chatbot could only work with what it was told.
The lesson for us: a good outcome depends on a good history, and that is exactly what a skilled triage nurse provides. The study used hypothetical scenarios, not real-world triage calls.
Five practical tips for telehealth nurses when callers use AI
These points are practical implications for our work, not findings measured in the studies.
| Situation | What to do | What you might say |
|---|---|---|
| 1. Caller leads with a diagnosis (“AI says it’s food poisoning”) | Acknowledge it, then go back to symptoms: onset, severity, progression, history. Don’t let the label narrow your assessment. | “Thanks. Can you tell me in your own words what’s been happening?” |
| 2. AI has reassured or alarmed them | Base the outcome on your assessment and the approved decision-support pathway, never on the chatbot. Explain your reasoning. | “Based on what you’ve told me today, this is the care I recommend, and here’s why.” |
| 3. They may have already acted on AI advice | Ask about self-treatment, delays or medications. Keep it curious, not critical. | “What did it suggest, and have you done anything because of it?” |
| 4. A polished, confident summary | Separate what they experienced, what others observed and what the chatbot said. Fluent medical language isn’t the same as an accurate history. | “Let’s go back a step. What did you actually notice first?” |
| 5. Your advice differs from the AI’s | Safety-net clearly: what to do, when, and what changes need urgent help. Confirm understanding. Follow procedure if they decline. | “Just so I know I’ve explained it well, can you tell me what you’ll do next?” |
Documentation: record any AI advice as information reported by the caller, alongside your own assessment, advice and any escalation.
Using AI responsibly in nursing practice
A general-purpose chatbot is not our approved clinical decision-support system.
- Use only tools authorised by the service.
- Never enter caller or client information into a public AI tool.
- Follow established assessment and escalation processes.
Ahpra is clear that your professional responsibilities still apply when you use AI and when you provide care by telehealth.
Why clinical judgement remains essential in telehealth triage
AI may change what a caller brings to the conversation. It doesn’t change our role: listen, assess independently, recognise risk and guide people safely to the right care.
