When your patient arrives with an AI self-diagnosis

For years, GPs have managed patients arriving with information and self-diagnoses from Dr Google. Now, for many patients, Dr AI has joined the conversation. 

Unlike search engines, AI platforms such as ChatGPT and Claude don't simply return a list of websites. They generate detailed explanations based on the symptoms, medical history and other information, such as uploaded test results, that patients provide.

How AI self-diagnosis is changing GP consultations

"For a long time, patients were researching their health with Dr Google," says Dr David Fraile-Navarro, a trained GP and Research Fellow in Generative AI at the Centre for Health Informatics, Australian Institute of Health Innovation, Macquarie University. 

"AI is adding a layer of complexity because it doesn't just provide links, it gives a full explanation. An explanation that can either be right, or wrong."

AI is persuasive because answers feel highly personalised.

"People tell AI, 'ten years ago I had this,' or 'I've already been tested for this, this and this.' It can build a very rich picture,” Dr Fraile-Navarro says. 

“If done properly, the advice can be interesting to contrast with a clinician, but it can also go off track."

Patients may arrive at appointments not just with questions, but with diagnoses they already believe, investigations they think they need, and confidence in AI-generated advice.

A/Prof Emily Kirkpatrick, senior clinical lecturer in artificial intelligence at the Australian Institute for Machine Learning at Adelaide University, says advice generated by an AI doctor isn't inherently good or bad.

"I see it as a really positive that patients are coming into the consultation with information," A/Prof Kirkpatrick says.

“It may not always be accurate or tailored to the individual, and AI can't examine or test a patient. But people are more actively engaged in their healthcare."

The challenge for GPs is in responding to AI-generated information in a way that addresses patients' concerns without validating inaccurate conclusions.

Ask patients if they've used AI

Before you can respond to an AI self-diagnosis, you first need to know whether the patient has used AI.

"I can only remember one patient who voluntarily disclosed using AI," says A/Prof Kirkpatrick. "They'd uploaded photos of a skin lesion to ChatGPT, which correctly advised them to see their GP and seek a dermatologist referral for a biopsy."

Most patients are unlikely to mention it unless they're asked, meaning GPs may miss concerns or misconceptions shaped by AI. 

"In my own practice, I've found that people are quite upfront about saying, 'yes, I've used AI,' if they're asked the question," says A/Prof Kirkpatrick. "But what worries me is that if I don't ask, I might overlook concerns that have been generated by AI."

Knowing whether AI has influenced a patient's thinking provides valuable context for the rest of the consultation.

Respond to AI-self diagnosis with curiosity, not correction

Once you know AI has shaped the consultation, resist the urge to dismiss it.

"My first recommendation would be not to dismiss or ridicule people for using these tools," says Dr Fraile-Navarro. "Patients are using AI because they're worried, they want to understand what's happening, and it's not always easy to access a clinician."

Treat AI like any other information a patient brings. Reviewing the AI chat together can reveal what the patient entered and where it may have gone astray. 

"Bring me the conversation,” Fraile-Navarro says. “Let's look at it together." 

A/Prof Kirkpatrick agrees, saying patients arriving with AI-generated information should be seen as engaged in their healthcare, not challenging their doctor's expertise.

"Say, ‘AI can be a really useful starting point’,” Kirkpatrick explains. “I'm really glad you've brought this to my attention.” 

Framing AI as a starting point, not an answer, acknowledges the patient's effort without endorsing the diagnosis.

Explain what AI can't do

The next step is explaining what AI can't do.

"Patients are worried, so don't dismiss it," says Dr Fraile-Navarro. "But explain that there are things an LLM can't do. It can't examine you. We may need to run some tests or look for alarm signs first. Let's turn this into a diagnosis plan that makes sense."

A/Prof Kirkpatrick also recommends explaining that AI recommendations may not reflect Australian practice because many models are trained on US data. 

"Let's work through what the Australian guidelines recommend and what that means for you," is something she often tells patients. 

A practical framework for managing AI self-diagnosis in general practice

A/Prof  Kirkpatrick recommends a simple framework to keep discussions about AI collaborative rather than confrontational. 

It starts with the simple question: Have you used AI to help support you in this?

From there, she follows four steps.

1. Acknowledge: Recognise the work the patient has already done
2. Explore: Ask what information they entered into AI and why they used it
3. Assess: Compare the AI's conclusions with the history, examination and Australian guidance
4. Plan: Develop a shared management plan that explains what happens next

Common pitfalls when discussing AI with patients

The biggest mistakes are dismissing AI-generated information simply because it came from AI, or accepting it uncritically.

"The answer is to engage," says Dr Fraile-Navarro. "Not to dismiss it, but also not to trust it blindly."

A/Prof Kirkpatrick says an instinctively dismissive response can undermine rapport before the clinical discussion has even begun.

"If you're dismissive upfront, it's very hard to build rapport and engage a patient about their use of AI," she says.

Clinicians also shouldn't assume AI has reached the right conclusion. It may overlook medical history, rely on overseas guidelines or produce recommendations that don't fit the patient. 

The GP's role is to critically evaluate AI-generated information against the patient's history, examination findings and Australian clinical guidance.

Key takeaways for GPs

"Patients use AI because they're looking for answers," concludes Dr Fraile-Navarro. "Our job is to interpret that information in the context of the individual patient. We’re used to doing that with test results and other complex situations. AI advice is now part of that."

• Ask patients whether they've used AI before assuming where their concerns come from.
• Treat AI as the starting point for a conversation, not the final diagnosis.
• Review the AI chat together to understand the patient's thinking.
• Explain what AI can't do, including examining the patient or recognising important clinical signs.
• Apply clinical judgement and Australian guidelines before accepting or rejecting AI-generated advice.
• Build a shared management plan that acknowledges the patient's concerns while keeping the consultation collaborative.

"Don't see AI as a threat to your clinical credibility,” concludes A/Prof Kirkpatrick. “See it as an opportunity to build stronger relationships with your patients, while also challenging your own thinking and exploring possibilities you may not have otherwise considered."

FAQs about AI self-diagnosis

Are AI doctors any good?

An AI doctor can provide useful health information, but can't examine patients, recognise physical signs or apply clinical judgement. AI self-diagnosis should support, not replace, a GP consultation.

Why are patients using AI for medical advice?

A lot of patients use AI because they want answers quickly, want to better understand their symptoms or want help preparing for a GP appointment. 

How should GPs respond to an AI self-diagnosis?

Don't dismiss AI-generated information or accept it uncritically. Instead, ask how AI shaped the patient's thinking, explain what AI can't do, and use clinical judgement to guide next steps.