AI for Health & Wellness: Where It Helps & When It Crosses the Line

Whether it’s an AI Overview at the top of a search results page or a chatbot app pinned to your home screen, AI-generated information is everywhere these days.

Dr. Olivia Tomasco, primary care physician at Houston Methodist, sees this firsthand in the clinic:

“I’ve definitely had people bring printouts of their ChatGPT conversations," she says. "People come in wanting to feel better — or to stay well. And they’ve asked ChatGPT what they need to have their doctor do for them."

That impulse to take control of your health is inherently good. The catch, though, is that AI can sound confident even when it’s wrong — and when it comes to your health, following guidance that's “sort of right” can be risky.

First: What is an LLM (and why does it matter for health info)?

Most of today’s popular chat tools — ChatGPT, Google’s conversational results, Perplexity and many others — are powered by LLMs, or large language models.

Here's a simplified explanation of how they work:

• LLMs are trained on massive amounts of text — books, articles, websites and more.

• They learn patterns in language, predicting which words tend to come next based on your prompt.

• The result can sound fluent and persuasive — without the model truly “knowing” whether the information is correct, especially if it isn’t grounded in verified medical sources.

This is why AI can be great at organizing ideas, yet still risky for anything requiring critical thinking and expert judgment.

How trustworthy is AI-generated health information?

LLMs often do well on structured medical questions (the kind with clear prompts and limited answer options). But real health care isn’t a multiple-choice test. It’s messy. Symptoms are vague. People forget details. Context changes everything.

A recent Nature Medicine study found that while LLMs can perform well on medical licensing-style exams, their accuracy drops significantly when asked to reason through nuanced, back-and-forth medical conversations with users.

In other words: LLMs can look “smart” in controlled settings, yet struggle when asked to perform like a clinician in conversation with a patient.

The biggest trap: “It sounded evidence-based”

Dr. Tomasco sees a specific pattern: AI answers often blend credible ideas with shaky ones, and people can’t always tell which is which.

“These AI tools might give some evidence-based information, but then also give some less evidence-based information," she explains. "This gives the person a false impression that the entire conversation is evidence-based, when it’s really not.”

That "mixing bowl" effect can send people down expensive or distracting rabbit holes:

“Someone may come in wanting this hormone test that just doesn’t make sense for their personal health history and what’s going on with them currently," she adds. "This can be a really tough mindset for a physician to turn around. It's hard for someone to hear that my recommendation doesn't match what they were expecting."

She adds that sometimes what looks like “wellness optimization” may actually signal something deeper going on — anxiety, burnout, loss of control or fear.

Dr. Tomasco poses the question: “Is the actual seeking of the wellness trend a symptom of something deeper that we need to kind of peel back the layers? For instance, why do you feel the need to do a cleanse if you’re not having symptoms?”

Where AI can help: Low-risk wellness wins

Here’s the sweet spot: AI is useful when the downside of being wrong is low.

For instance, Dr. Tomasco says it's great for helping you organize healthy habits, especially if you live a hectic, busy lifestyle.

Meal planning

Dr. Tomasco actually uses AI herself for this. “I think that it’s really great for meal planning," she says. "I’m busy, my husband works, I have two kids. I’ll say, ‘Hey, give me an easy, healthy, kid-friendly recipe to cook tonight,’ and it’ll just give me something. I don’t even have to think much about it.”

Try AI for:

• High-fiber meals, plant-based protein options or other hard-to-meet nutrition goals

• Quick grocery lists

• "Make it easier" recipes (ever heard of crockpot lasagna?)

• Ingredient substitutions

"Be careful if you have a chronic condition that impacts your diet," warns Dr. Tomasco. "For instance, diabetes, fatty liver, kidney disease — it's best to consult with a trained dietitian for personalized nutrition advice.”

(Related: Want to Eat Healthier? Try These 7 Easy Tips)

Building an exercise routine

For many people, having a plan — even a simple one — can make the difference between intending to exercise and actually doing it.

“If you have a hard time starting a workout regimen and ChatGPT can take that stress off for you, then that’s a really great way to support a healthy habit like getting active," she adds.

Try AI for:

• Beginner-friendly workout plans

• Organizing workouts on a busy schedule

• Accountability nudges

Keep in mind: It's always best to talk to your doctor before starting an exercise program, especially if you have a known health condition.

(Related: How to Start Exercising After a Break of Any Length)

Stress management

“If you saw online something about how deep breathing can help with stress and you just want to use it to plan a guided meditation, that’s totally fine," says Dr. Tomasco.

Try AI for:

• Scripts for breathing exercises or meditations

• Journaling prompts

• Bedtime wind-down routines

Dr. Tomasco does, however, draw a firm line when anxiety is acute or a person is experiencing depression:

“If you're having a panic attack or feeling depressed, you need to talk to a human being about what’s going on," Dr. Tomasco urges. "That is not something AI can safely help you through."

(Related: How to Tell the Difference Between Stress & Anxiety — Plus, Tips for Coping)

When AI crosses the line: “No, AI isn’t ready to be your doctor”

This is the core boundary: anything requiring medical decision-making belongs with a licensed professional who can examine you, ask the right follow-up questions and recognize red flags.

Dr. Tomasco emphasizes that clinical training isn’t just “knowing facts.” It’s knowing what to worry about, what to rule out and what questions to ask next. Your doctor is also trained to pick up on facial expressions — looking for concerns that may lie just under the surface, or things you might be downplaying or brushing off.

And going to AI for medication advice is a hard no:

“Anything that requires medical management — medication changes, side effects, medication interactions — do not turn to AI for that sort of thing," Dr. Tomasco stresses.

How primary care physicians think about wellness

Dr. Tomasco says what she focuses on in appointments — especially when someone says they 'just don’t feel well’ — is typically the foundational factors people often underestimate.

She asks:

• What are your stressors in life?

• How's your diet?

• What's your sleep like?

• How often do you exercise?

• Do you smoke or use recreational drugs?

• How many drinks do you have per week?

"In addition to discussing whatever issue brought someone in, I touch on these things because people may not feel like these are things to bring up or that they're not that important, but they are," says Dr. Tomasco. "And if your doctor is asking you about it, that’s basically my way of saying this is important.”

She ends by reminding us of the importance of face-to-face human interaction:

“You need to have a person who cares about you and cares about your health," says Dr. Tomasco. "That’s what matters most.”