Mayo Clinic Launches AI Chatbot for Doctors

By Crystal Lindell

If you’ve ever shared with a doctor medical information or advice you found online, they may have been annoyed or defensive. After years spent in medical school and/or residency, many doctors dislike anything that questions their expertise – especially if it came from the Internet.

Which makes a new artificial intelligence (AI) program launched by the Mayo Clinic all the more interesting.

Starting this month, the world renowned medical center is launching a new chatbot called the Clinical Decision Support (CDS) solution. The AI-driven chatbot is not for patients, but for the Mayo Clinic’s own doctors and healthcare providers. It will cover more than 35 medical specialties and have access to patient health records, providing “industry-standard medical information” that can “help clinicians reason through complex cases.” 

"Clinicians have access to a vast amount of medical knowledge, but finding relevant information for individualized patient care takes valuable time," says Maneesh Goyal, chief operating officer of Mayo Clinic Platform. 

"With this new solution, we are bringing trusted Mayo Clinic knowledge directly into the clinical workflow so clinicians can spend less time searching for information across multiple resources and more time focused on patient care." 

In other words, Mayo Clinic doctors will be doing the same thing patients do – searching for answers to their health questions.

The Mayo Clinic clarified that its doctors will still be responsible for all medical decisions, regardless of their use of AI. 

"The CDS solution is designed to assist healthcare professionals as they make patient care decisions; it does not replace clinical judgment," said Gelareh Zadeh, MD, chair of the Department of Neurosurgery at Mayo Clinic in Rochester. "Mayo Clinic physicians have provided feedback on the different tools to ensure the solution supports clinicians and reflects Mayo Clinic's approach to patient care." 

A press release says the new chatbot will be evaluated for “safety, effectiveness, privacy, security and clinical value, with appropriate human oversight and ongoing monitoring.”  

As a patient, I’m hopeful that Mayo Clinic would only use an AI chatbot if they felt comfortable with its accuracy and security. But as an AI user, I’m skeptical. 

Interestingly, the Mayo Clinic itself shared an article in April that asked, "Can you trust AI for health advice?" The answer they gave is essentially no, with five reasons cited: 

1. Diagnosing and treating illness is too complex for a machine

2. AI can be wrong, even when it sounds confident

3. AI-created information may be biased

4. AI spreads misinformation

5. People and AI don't communicate well together

Mayo Clinic providers would be wise to keep those five cautions in mind as they use their own chatbot. 

PNN has published multiple stories about the risks of using AI in healthcare. One recent article was about a woman who used ChatGBT to seek advice on how to taper off her seizure medication. The advice she received may have led to her death. 

We also published a warning to older adults about how they need to be cautious about taking medical advice from chatbots. 

"Using AI for health information can be particularly fraught because chatbots can be so helpful in explaining complex medical information, but also risky if users automatically trust the responses they receive,” warned Robin Brewer, PhD, an Associate Professor at the University of Michigan.

The hope is that the Mayo Clinic’s chatbot will be more accurate than ChatGPT, but it will take time and usage to reveal if that’s true. 

My other concern is that providers will start to overly rely on the AI system, and not use their own clinical knowledge for context or to fact check the information. 

A 2025 study by researchers at the Massachusetts Institute of Technology (MIT) found that people who relied on ChatGPT to write essays had lower brain activity than those who used their own intelligence. Those who used AI alone became dependent on it and struggled when asked to perform tasks on their own.

Time will tell how Mayo Clinic’s AI chatbot will perform and how accurate it will be. Let’s just hope no patients get hurt while they figure it out.

Older Adults Should Be Cautious About Health Advice from Chatbots

By Robin Brewer

It’s easy to stereotype older adults as tech-averse or novice tech users. Yet a growing number of adults age 60 and older are embracing artificial intelligence (AI), using it for entertainment, planning trips and even accessing health information.

I study how people interact with computers and technology – and specifically, how older adults use voice assistants and AI-powered chatbots like Amazon Alexa and ChatGPT to engage with information. 

Using AI for health information can be particularly fraught because chatbots can be so helpful in explaining complex medical information, but also risky if users automatically trust the responses they receive. That’s apparently what happened to a young woman in Seattle, who died after ChatGPT encouraged her to stop taking her epilepsy medication.

My research suggests that the most productive way of engaging with health information generated by an AI chatbot is from a position of informed skepticism – that is, remaining open-minded, yet using built-in features to evaluate the credibility and quality of chatbot responses.

The AI Appeal

About 1 in 5 adults turn to AI chatbots to ask questions about health, according to a 2025 survey of 300 adults. Questions might include asking about symptoms or seeking advice about diet and fitness. The chatbots might be mainstream services such as Claude, Gemini, ChatGPT or systems designed for a specific health provider.

People turn to these tools for many reasons. It can be easier to talk to a chatbot about sensitive topics, and chatbots reply almost instantaneously – much quicker than making an appointment to see your doctor weeks or months down the line. Chatbots can also explain complex topics in new ways and condense an overwhelming amount of information into a simple summary.

Although older people may be less likely than others to use AI, I’ve found in my own work that older people often use voice assistants like Siri and Alexa to ask health questions. Similarly, a poll I conducted with colleagues in 2025 shows that 1 in 2 older adults report using AI chatbots or voice assistants, and about 14% of people in this group are accessing health information.

Older adults often experience more health needs than younger adults due to long-term chronic health conditions or new age-related diagnoses. More health needs often means more health-related questions.

AI is particularly appealing for health questions because people get answers faster than searching across multiple websites when using a search engine – or having to wait for a healthcare visit. 

Moreover, being able to use chatbots and voice assistants at home can make people feel more in control of their health, compared with clinical settings where older people’s autonomy can be overlooked or where they may experience being excluded in lieu of conversations with caregivers.

What Are the Risks?

Although older people may use chatbots to supplement or replace medical professionals, AI use has risks.

For example, it can provide incorrect advice in medical emergencies, misclassify health conditions and be overly agreeable, responding with what it deduces the user wants to hear, even if the response is not factually accurate. AI experts call this last tendency “sycophancy.”

But these inaccuracies can be hard for a user to detect. AI-powered chatbots draw on vast amounts of data and use pattern-matching to predict words most likely to come next in a sentence. That means the output can appear correct but contain inaccurate information.

Most mainstream chatbots include a warning under the dialogue box noting that their output might contain mistakes, but it is often in small or gray text, making it easy to overlook. The warning is meant to encourage cross-checking information with published sources, but users rarely do so. In the 2025 survey mentioned above of 300 adults who used a health chatbot, 4 in 5 admitted they never verified its output.

In my own research, I found that older people searching with a voice assistant preferred to begin and complete their search process on the same device, suggesting they are unlikely to verify an AI chatbot’s output if it takes them to a different format. In contrast, younger adults expected AI to be a step in the search process, guiding them to seeking further information with the help of another option, such as a search engine. This difference makes it critical to provide strategies to older people as they use AI.

Another risk is privacy, since AI companies that make public chatbots often keep user prompts for training future models. Researchers generally recommend that users not include personally identifiable information when asking questions to chatbots.

Strategies for Safer Searching

Awareness that older adults may use AI in ways distinct from other demographic groups – and may encounter specific challenges – is growing. A bipartisan bill, the Aging with AI Act, introduced in the U.S. Senate in 2026, for example, underscores the need for more research on both its benefits and its risks.

My research points to a few guidelines to start with:

Learn how to opt out of AI. Although mainstream chatbots like ChatGPT are popular, the most common way people access AI is through a search engine because AI features are often built directly into search results. Some search engines, like Microsoft’s Bing, ask people to opt into using AI. Others, like Google Search, rely on users to opt out of AI.

Google shares an AI-generated summary at the top of each search result. To opt out when searching for information using Google, adding “-ai” to the end of the search query will remove AI content. Opting out of AI might be useful when searching for particularly sensitive health information that you wish to keep private.

Use the information quality features. Most AI chatbots have features that allow users to confirm the source of its output. For example, Claude, Gemini and ChatGPT include links to sources they used to predict the response. Make sure you explore these sources to confirm they are legitimate and to get more details that can help you find the information you need.

Further, you can change the model the chatbot is using, which can change how much depth the chatbot goes into in its response. For example, Gemini has a drop-down menu next to the search box that includes “flash-lite,” “flash” and “pro” models. Chatbots’ more advanced versions provide more thorough responses that can help users better determine information quality.

And users can report inaccurate, toxic or harmful content by clicking the thumbs down icon shown at the bottom of the response.

Ask better questions. Ask for objective information that can be verified, rather than subjective advice, particularly in high-risk health contexts. For example, instead of asking “which high blood pressure medicine should I take?” ask “what are the side effects of” a particular high blood pressure medicine.

And to prevent sycophancy, phrase questions neutrally. For example, instead of asking “Is it a good idea to get a flu shot?” a searcher can instead type “Tell me about the pros and cons of getting flu shots.”

Don’t use AI as a replacement for medical professionals. Health-related chatbot output is best used collaboratively with a medical professional. This might include saving the chatbot output to discuss during your next healthcare appointment, using chatbots to understand complex diagnoses from your doctor, or getting a chatbot to suggest questions to ask your doctor during your next visit.

Most importantly, confirm health-related answers that result in making changes to your health routine with your healthcare provider. A knowledgeable, licensed healthcare provider is best qualified to determine how general information about illness or treatments would apply to your specific situation.

Robin Brewer, PhD, is an Associate Professor at the University of Michigan in the School of Information. Her research covers old age, disability, and technology use. 

This article originally appeared in The Conversation and is republished with permission. 

Chronic Pain Shouldn’t Be Diagnosed With Brain Scans

By Jan Vollert

In 2006, Carl Koch sued his employer for damages after burn injuries during a workplace accident left him with chronic pain. The employer accused him of malingering, so the judge admitted a neuroscientist as an expert witness, who testified that he could see Koch’s pain on a brain scan. The case was settled for more than ten times the amount the employer initially offered.

This case is not unusual, as chronic pain can be severely debilitating. It can make it impossible to work, exercise and have a social life. It can make you depressed and anxious.

Having pain recognised by insurers, employers and government agencies can mean the difference between receiving benefits and being left without support. So it’s understandable that a judge wants to see objective evidence, such as a brain scan, showing chronic pain.

The problem is that it doesn’t work. However, as Koch’s case shows, it is used anyway. Good for Mr. Koch, in this case, but what if the neuroscientist had told the judge that he is faking his pain?

Who Is Believed?

Chronic pain is invisible, and everyone living with it has stories to tell about the time someone did not believe them: employers, friends, family, partners and doctors. Is it really that bad? Are you sure it’s not just in your head? Have you tried to toughen up a bit?

Who we believe and who faces more scepticism is not random either: women, people of colour, poor people and the less educated are far less likely to be believed that their pain is real. The cruelty is doubled as these are the groups most likely to experience chronic pain in the first place.

With modern brain scanners and AI, it seems tempting to develop what is effectively a pain-lie detector test. The pictures brain scanners show have been too complex to fully understand before, but AI can often solve such problems of complexity. Two studies published in Nature Neuroscience and Nature Medicine seemed to promise to be a first step towards a pain lie detector.

In the US, AI tools are already used to decide who should be prescribed opioids for their pain and who is probably an addict pretending to be in pain. An article in the technology magazine Wired uncovered how flawed the system was, and that it had a tendency to distrust all the groups mentioned above.

This is a problem with AI: even when it seems to work, it does not tell you how. This makes it dangerous to put too much confidence in it.

There can be no such thing as a pain lie detector for the same reasons that regular lie detectors do not measure if you lie; they measure if you are nervous – but you might be nervous telling the truth in a high-stakes situation.

It is similar for measuring pain. Pain causes distress, so any brain pattern of pain will also be a pattern of stress – and often anxiety. We cannot factor that into our pain lie detector: when we develop it, it will be with curious participants volunteering for a scientific study (not very stressful).

When we use it, it will be for the patient who has been disbelieved for years and knows if the scan does not “prove” their pain, no one will ever believe them, and they will not get compensation or benefits (as stressful as it gets). It will always be a catch-22, as explained in The Conversation in 2025.

A Private Experience

Pain is a personal and very private experience. It depends on your history with it, and on the context, meaning everyone’s pain is different, and your pain today can be different from your pain tomorrow. Some pain can even be pleasurable – spicy food being an obvious everyday example. This means that your brain scan of pain may look very different from mine. Until we have measured both, we cannot know.

It is for these reasons that the International Association for the Study of Pain has stated that pain lie detectors are useless. In a new article in Nature Neuroscience, my colleagues and I discuss technical and fundamental problems with the idea and ask the scientific community to abandon the idea.

So what does this mean for you? If someone tells you about their chronic pain, put all scepticism aside. Remember that most forms of chronic pain are invisible. Back pain, for example, is the biggest cause of disability worldwide, according to the World Health Organization, and a physical cause can rarely be found.

Keep in mind that machines make mistakes as well, and that they have the same stereotypes as we do. That is especially true for AI.

A pain scan will not tell us who is suffering and who is lying. But it may tell us something else: how easily we mistake technological confidence for truth. In medicine, insurance and law, that mistake could make already invisible pain even harder to prove – and easier to dismiss.

Jan Vollert, PhD, is chronic pain researcher and lecturer at the University of Exeter.

This article originally appeared in The Conversation and is republished with permission.  

Can AI Outperform Human Doctors?

By Crystal Lindell

It may not be long before a trip to the emergency room means telling your symptoms to an AI robot, potentially before you even talk to a human doctor. 

New research published in Science seems to highlight the potential for artificial intelligence to create such a future in healthcare.

The study -- which was conducted by both Harvard and Stanford researchers – tested OpenAI’s experimental “o1 preview” models against human physicians. OpenAI makes ChatGPT.

They asked the o1 models to do a patient diagnosis and create a diagnostic testing plan, then compared its skill in clinical reasoning to experts and generalist physicians.

They also assessed AI on 76 real-life emergency room patients at a Boston hospital in three stages: the initial triage at first arrival; first contact with a physician; and upon admission to the hospital. 

The results showed that the new AI model outperformed human physicians and showed improvement from earlier generations of AI. 

“Our findings suggest the urgent need for prospective trials to evaluate these technologies in real-world patient care settings and for health care systems to prepare for investments for computing infrastructure and design for clinician-AI interaction that can facilitate the safe integration of AI tools into patient-care workflows,” wrote lead authors Arjun Manrai, PhD, Assistant Professor of Biomedical Informatics at Harvard University and Adam Rodman, MD, Director of AI Programs at Beth Israel Deaconess Medical Center. 

In the emergency department cases, the o1 model was diagnostically correct 67.1% of the time during the initial triage, outperforming two expert attending physicians (55.3% and 50.0%).

Physicians who reviewed the diagnostic results – without knowing if they were made by AI or human doctors – were unable to distinguish between the two. 

“AI models are evolving from static question-and-answer tools into agents that can, for example, analyze patient records, monitor clinical encounters through ambient listening, and interact in real time with predictive models built on patient data," Ashley Hopkins, PhD, and Erik Cornelisse, PhD candidate, at the College of Medicine and Public Health at Flinders University in Australia, wrote in an op/ed on the study.

“This advance sets a new evaluation benchmark — testing AI against physician performance, and ideally alongside physicians, on authentic clinical tasks.”

Interestingly, Hopkins and Cornelisse pushed back on the idea that the ideal method for evaluating patients is physicians collaborating with AI. They think AI may perform better on its own. 

“That collaborative configuration itself must be tested,” they write. “It has been argued that for certain well-defined tasks across health care, AI may operate more effectively independently.”

They also wrote that since many doctors are already using AI in their practices, sometimes without institutional oversight, further studies are urgently needed to determine when AI improves patient care and when it does not.

In an article about the AI study published in Harvard Magazine, Arjun Manrai, the senior co-author of the study, said the results do not show that “AI replaces doctors, despite what some (AI) companies are likely to say.”

“I think it does mean that we’re witnessing a really profound change in technology that will reshape medicine,” Manrai said. “We need to evaluate this technology now and rigorously conduct prospective clinical trials.”

Manrai also makes an important point. The AI study was based entirely on text-based inputs, while practicing physicians evaluate many other forms of information and communication, such as listening to a patient, observing how a patient behaves, examining images and x-rays, and evaluating other test results. 

AI can’t do all those things – at least not yet.

Manrai’s co-author, Adam Rodman, also thinks it’s premature for AI to replace doctors in clinical settings. AI might prove useful in providing second opinions and finding diagnostic mistakes, but Rodman doesn’t want to see “AI doctor companies” replacing human physicians.

“I do not think that these results support that,” Rodman said. “What these results support is a robust and ambitious research agenda to try to figure out how we can use these technologies to make patients’ lives better.”

Patients Are Becoming Less Open to AI in Healthcare

By Crystal Lindell

Many of us are using artificial intelligence (AI) in our everyday lives, such as learning more about our medical conditions and symptoms. But when it comes to actually using AI in their own healthcare, patients are less open to it.

That’s according to a new poll by The Ohio State University Wexner Medical Center, which surveyed 1,007 adults across the country about their opinions about AI in healthcare.

They found that just 42% of adults are open to AI being used as part of their healthcare in 2026. That’s down from the 52% who supported it in 2024. The belief that AI can make healthcare more efficient also fell, from 64% to 55%.

However, the survey found that over half of the adults (51%) still used AI to help them make important health decisions, without consulting with a medical professional.

Participants said they use AI for a variety reasons:

• 62% use AI to help understand symptoms

• 44% use AI to help explain test results or a medical diagnosis

• 25% use AI to compare treatments and help make a treatment decision

• 20% use AI to prepare for an upcoming medical appointment

The drop in patient trust with AI is on par with the natural “hype cycle” of any new technology, according to Ravi Tripathi, MD, chief health informatics officer at Ohio State Wexner Medical Center.

“When we first see something new and shiny, we think it's going to fix the world and replace health care and solve all of our medical problems,” Tripathi said. “People are learning that there are pros and cons of artificial intelligence, where it has actual use and where it really doesn't have a place.”

Tripathi predicts that over the next two to five years, trust in AI will increase, as people become more familiar with artificial intelligence and it becomes more common in healthcare technology.

But he warned patients against relying too heavily on AI for their own medical research.

“We know that 2% of the time AI is going to be inaccurate or it will potentially hallucinate,” Tripathi said. “Physicians are not using AI 100%. We're not trusting it 100%. I would be really concerned about a patient who is following AI. The artificial intelligence doesn't understand your story.” 

Tripathi suggests using AI in partnership with your doctor. AI can help patients compile their health data, explain test results and diagnoses, and help identify questions to ask providers.

“There's a strong value for using artificial intelligence as augmented intelligence,” Tripathi said. “Patients should have oversight of what the technology is doing but consult with their health care team for the final plan.”

While patients have mixed feelings about AI, doctors appear to be more open to it. 

According to a recent survey by the American Medical Association, 81% of physicians use AI to stay current on medical research and to help them with record keeping. That’s about double the rate in 2023, when the AMA first polled doctors on their AI use.

While 76% of physicians say AI technology can help with patient care, about 40% said they are both excited and worried about it – citing concerns about patient privacy and the integrity of the patient-physician relationship.

The global AI healthcare market is projected to reach $868 billion by 2030, with its influence on the overall healthcare market more than doubling from roughly 15% today to over 30% by 2030.

AI in Healthcare: Designed for Progress or Profit?

By Crystal Lindell

As a pain patient, I take a controlled substance medication, which means every single time I need a refill I have to contact my doctor. 

It doesn’t matter that this refill comes every 28 days and that I have been getting it refilled every 28 days for years. It doesn’t matter that my condition has no cure, and that I will most likely need this medication refilled every 28 days for the foreseeable future.

No. I have to make sure to contact my doctor and specifically ask for it, every single time.  

There are ways to automate this process. They could give me a set number of automatic refills and have them sent to the pharmacy every 28 days. If we were even more practical, they could just give me 60 to 90 days worth of pills at a time, and save me from multiple trips to the pharmacy. 

But because of insurance rules, hospital policies and opioid-phobia legislation, all of those options are impossible. In fact, they actively work to make a process that could be automated into something that has to be done manually. 

Which is why I’m so skeptical of Artificial Intelligence (AI) in healthcare. 

The promise of AI is that it can automate away the mundane tasks so many of us hate doing. Many health related tasks could easily be automated. They just purposefully are not. 

The hospital I go to for my medical care, University of Wisconsin-Madison, recently released a report filled with recommendations for how AI should be integrated into healthcare. It was based on a recent roundtable discussion that included healthcare professionals from across the country. 

But while the participant list included doctors, IT staff, policy experts, and academics, there was one very glaring absence – the list of participants included exactly zero patients. 

UW Health was one of the organizers for the panel, along with Epic, a healthcare software developer. Their report includes some seemingly good recommendations. 

They ask that AI be used to supplement the work that doctors, nurses and other healthcare staff perform, as opposed to replacing the staff altogether. They say AI could be a great tool to help reduce staff burnout. 

They also recommend that the technology be set up in such a way that it also helps those living in rural areas, in addition to those in more metropolitan ones. The report also emphasizes that healthcare systems should prioritize “weaving the technology into existing systems rather than using it as a standalone tool.”

Additionally, the report stressed the need for federal regulations to “balance space for innovation with safeguarding patient data and ensuring robust cybersecurity measures.”

I don’t disagree with any of that. But it’s a little frustrating to see those recommendations, when some of those problems could already be solved if we wanted them to be. 

And while the panel’s report is new, UW Health’s use of AI is not. 

In April, UW Health announced that they were participating in a new partnership program with Microsoft and Epic to develop and integrate AI into healthcare. 

At the time they said the innovation would be focused on “delivering a comprehensive array of generative AI- powered solutions… to increase productivity, enhance patient care and improve financial integrity of health systems globally.”

That’s the real motivation to bring AI into healthcare: make more money by improving “financial integrity.” Something tells me that AI won’t be used to lower patient’s bills though. 

UW Health also recently shared that its nurses were using AI to generate responses to patients. Over 75 nurses were using generative AI, which assisted them in creating more than 3,000 messages across more than 30 departments.

“This has been a fascinating process, and one I’ve been glad to be part of,” said Amanda Weber, registered nurse clinic supervisor, UW Health. “I have found having a draft to start from helpful, and I’m glad I could provide feedback on improvements and features to ensure this can be a good tool for nurses and have a positive impact on our patients.”

Before I even knew about this program, I had a feeling that AI was involved. 

Recently, when I messaged my doctor about my upcoming refill, I received an overly-formal, odd response that felt very much like generative AI writing to me. Which is fine. I honestly don’t mind if my doctor saves time by using AI to respond to patient emails. Heck, I myself have used AI to write first drafts of some emails. 

But my doctor and his staff wouldn’t even need to reply to my emails if he was allowed to set up automatic refills of my long-time medication instead. 

There are many ways to improve health care, and tools like generative AI are likely among them. But AI can’t solve problems that exist on purpose. 

Unless patients are at the forefront of the conversations about these tools, I fear they’ll only be used to solve the sole problem hospital administrators actually care about: how to make more money.