Should you use AI to diagnose your symptoms or to research treatment?
Be careful, says Dartmouth Health Chief Health Information Officer Randa Perkins, MD, MBA, FAMIA. “In the IT world we often remind people that ‘If something's free, then you're the product.’”
That is not to say that Perkins discourages doctors and patients from using AI tools. In fact, she is leading efforts at Dartmouth Health to make better use of AI technology, such as AI-enabled ambient scribes, to improve care, outcomes, and efficiency.
But Perkins, who also practices medicine at the Dartmouth Hitchcock Clinics Heater Road Primary Care, is aware of the risks and pitfalls of AI, and of the importance of privacy safeguards.
“We all have to be very aware,” Perkins says, stressing that your clinician is central to any medical diagnosis or treatment.
Here are five caveats to using AI technology to better understand your symptoms and possible treatment.
1. That study might not be real
It is well-documented that AI chatbots sometimes produce fake citations. According to this 2026 article in the medical journal The Lancet, an audit of millions of biomedical papers found they contained more than 4,000 citations to bogus studies.
And it’s not just fake studies that AI draws from. “AI tools can easily facilitate misinformation at scale,” says Perkins, explaining how AI has the ability to pick up false information and quickly communicate that message without checking for accuracy.
“What's to stop an AI tool from amplifying false information that could actually harm you?” she asks.
The takeaway: If you use AI to help do your research, always check the source.
2. Your shared personal data could get hacked
You need to be careful about the information you share online. It is no coincidence that when you search or even talk about a condition, you often get pushed related information.
“Again, if a product is free, then consider how that product is being provided – it’s likely using you to generate content, or at least activity data, that it’s monetizing later,” says Perkins. “Someone out there is likely using your information in some way.”
And while getting pushed an ad might seem relatively innocuous, there is little to stop commercial organizations from leveraging your personal information for their own gain.
If you disclose your data, AI chatbots aren’t typically governed by the Health Insurance Portability and Accountability Act (HIPAA).
Once more, when you upload your medical record to tools like ChatGPT or Claude, that data might include information that even a clinician would never share with a consulting colleague, reports this July 2026 article in JAMA.
And while OpenAI, which owns ChatGPT, vows to keep users’ data secure, once you turn over your records, you could lose control over what happens to them, warned one of the study’s authors, Glenn Cohen, in a July 2026 article in Harvard Law Today.
“This is not the equivalent of asking your doctor something, or of getting a second opinion on a health problem you have, from a clinician,” Cohen said.
That sharing of data could harm you in unanticipated ways.
“We know patients are Googling their symptoms left and right, and that people have moved from Googling symptoms to checking ChatGPT or checking Claude,” says Perkins. “I was shocked when I happened to pull up ChatGPT to do some research, and I got the same pop-up that I know many others did. It said, ‘Would you like to integrate with your Apple Health record?’ I nearly dropped my phone to avoid accidentally enrolling, which would have been three simple clicks. It’s all too easy to do, but the consequences are real and hard to reverse.”
The takeaway: Be careful what you share and never share personal health information, especially your health records, with untrusted resources.
3. AI doesn’t know your whole story
No matter what information you give, an AI chatbot is unlikely to understand your full medical history. An AI chatbot is a machine. It can’t see you, hear you, or ask vital follow-up questions. It only has the information you give it.
That’s not to say AI cannot sometimes provide helpful insights. In her practice, even Perkins sometimes uses computer-assisted clinical decision support (CDS) tools, also called Decision Support Interventions (DSIs).
She gives the example of a patient she saw in clinic whose asthma was acting up. As she discussed the situation with the patient, she was able to use one such CDS tool in Dartmouth Health's HIPAA-compliant, secure, confidential data system to review recommendations. The system provided guidance that was aligned with and able to cite national guidelines on asthma management.
“I reviewed what it presented, confirming the source was accurate, and agreed with the indicated next steps. I'm not sure I wouldn't have done the same thing otherwise, but certainly AI-enabled tools can expedite and support the diagnostic and treatment process,” she says.
At present, tools like these are not available to patients. Until some form of equivalent chatbot is developed to help you, as patient, navigate and review your health records and medical education materials in a HIPAA-compliant, secure, confidential system, you need to remain very careful about how you make your AI queries, ensuring you exclude any identifying information.
The takeaway: AI can help put a piece in a larger puzzle for you. Just know that help is limited. Keep out all identifying information.
4. You might not be represented in the data AI is using, and it could be flat-out wrong
Flawed and skewed studies can lead to misleading health advice, unproven treatments, and unequal care, warns Perkins.
That’s because people who conducted the research might have had a vested interest in a particular result, made mistakes, or drawn results from a population that was insufficient in some way. The studied population’s age, ethnicity, race, gender or other important characteristics also might not match your own, so the reported results might bear little relation to you.
Unfortunately, that same study might be the one the AI chatbot drew its conclusions from.
The takeaway: Continue to use trusted resources, including your clinical team, to evaluate clinical situations. Consider whether the information used to train the AI tools was biased, as that can distort the results and conclusions in ways that aren’t easy to detect once they are provided.
5. Keeping humans in the loop means accountability
Perkins welcomes when her patients come to appointments informed, having researched their symptoms or condition and prepared with questions beforehand.
“AI can help patients craft their messaging so that they are communicating in a way that's more thorough and precise,” she says.
But she also stresses that no matter how sophisticated AI becomes, it is essential that clinicians stay central in conversations about their patient’s health.
Not only can clinicians draw from their own expertise, experience, and knowledge of you as a patient, but they are accountable for the diagnosis or treatment they prescribe according to their best judgment. They can then monitor your health and modify your treatment based on what they see and what you tell them.
AI chatbots lack that personalized context, accountability, and ability to monitor, observe, follow up, or modify treatment.
The takeaway: “AI tools are just that, tools to support the clinical care experience,” says Perkins. “Your clinical team will always be your best resource to partner with you in your healthcare journey, providing the compassionate care and wisdom that no technology can replace.“
