Last updated: October 2025
AI in healthcare gets more hype than any other industry. Half the headlines promise AI will cure cancer. The other half warn it will misdiagnose you. The reality is more boring and more useful: AI handles specific, well-defined medical tasks better than humans, while remaining terrible at the holistic judgment that good medicine requires. (See also: AI for Recruiting: Source, Screen, Evaluate, and)
Here’s what’s actually working, for patients and for healthcare providers.
For Patients
These tools are available directly to consumers, no doctor’s office required.
Symptom Checking
Ada Health: Best Symptom Checker
Ada asks you questions about your symptoms, medical history, lifestyle factors, and risk factors, then provides a ranked list of possible conditions with recommended next steps. The AI is trained on medical literature and validated against clinical outcomes.
How it works: “I have a headache, fatigue, and mild fever for 3 days” → Ada asks follow-up questions (location of headache, severity, other symptoms, recent travel, medications) → provides 5-8 possible conditions ranked by likelihood → recommends “See a doctor within 48 hours” or “Seek emergency care” based on severity.
Accuracy: Studies show Ada’s top-3 suggestions include the correct diagnosis 70-80% of the time. Not perfect, but better than Googling symptoms (which always suggests cancer).
What it replaces: The 2 AM panic Google search. Ada provides structured, medically-informed guidance instead of random forum posts and worst-case scenarios.
What it doesn’t replace: A doctor. Ada explicitly states it’s not a diagnostic tool. It helps you understand your symptoms and decide whether/when to seek care.
Cost: Free
Chronic Condition Management
Livongo (for diabetes): AI-powered blood glucose monitoring that learns your patterns and provides personalized recommendations. “Your glucose tends to spike after lunch on workdays. Consider a 15-minute walk after eating.” The AI adapts to your specific metabolism, not generic guidelines.
Whoop / Oura (for general health): Wearables that track sleep, heart rate variability, and activity to provide daily health insights. “Your resting heart rate increased 8 BPM overnight, possible early sign of illness. Consider resting today.”
Cost: Livongo (covered by many insurance plans) | Whoop from $149/year | Oura $6/month + hardware
Mental Health
Wysa: AI Mental Health Companion
Wysa provides cognitive behavioral therapy (CBT) techniques through a chat interface. It’s not a replacement for a therapist, but it’s available at 3 AM when your therapist isn’t.
What it does well: Daily check-ins, mood tracking, CBT exercises, and guided meditation. For mild anxiety and depression, the structured CBT approach is clinically validated. Wysa has been evaluated in multiple peer-reviewed studies.
What it doesn’t do: Handle severe mental health crises, prescribe medication, or provide the human connection that therapy offers. Wysa directs users to crisis resources when needed.
Cost: Free (basic) | $100/year (Premium)
Headspace / Calm: AI-personalized meditation and mindfulness programs. The AI adapts session recommendations based on your mood, stress level, and usage patterns.
Cost: $13-15/month
Medical Record Management
Apple Health + AI: Aggregates medical records from multiple providers into one place. AI highlights trends: “Your blood pressure readings have increased over the past 6 months. Discuss with your doctor at your next visit.”
Cost: Free (iPhone required)
For Healthcare Providers
The provider side is where AI gets really interesting. These tools are changing how hospitals and clinics operate.
Medical Imaging
AI Radiology (Aidoc, Viz.ai):
AI reads medical images (X-rays, CT scans, MRIs, ultrasounds) and flags abnormalities for radiologist review. The AI doesn’t diagnose; it prioritizes. A CT scan showing a potential stroke gets flagged as urgent and moved to the top of the radiologist’s queue.
Impact: Time-to-treatment for stroke patients decreased 25-30% at hospitals using AI triage. The AI catches what a tired radiologist at 3 AM might miss on their 200th scan of the day.
Accuracy: AI matches or exceeds radiologist accuracy for specific conditions (lung nodules, brain hemorrhages, fractures). It’s worse than radiologists for rare conditions and complex cases.
Clinical Documentation
Ambient AI Scribes (Dragon Copilot / DAX Copilot, Abridge):
AI listens to doctor-patient conversations and generates clinical notes automatically. The doctor focuses on the patient instead of typing into a computer. After the visit, the AI-generated note is reviewed and signed. Dragon Copilot (formerly Nuance DAX) is now deployed across 650+ health systems and deeply integrated with Epic.
Impact: Doctors using AI scribes spend 40-60% less time on documentation. That’s 2-3 extra hours per day for patient care. Burnout decreases. Patient satisfaction increases (because the doctor is looking at them, not a screen).
Cost: Dragon Copilot ~$369/provider/month | Abridge and other alternatives from $200/provider/month
Drug Discovery
AI Drug Discovery (Insilico Medicine, Recursion):
AI identifies potential drug candidates by analyzing molecular structures, biological pathways, protein interactions, and clinical data. What used to take 4-5 years of lab work (identifying a promising compound) now takes 12-18 months.
Reality check: AI accelerates the discovery phase but doesn’t skip clinical trials. A drug still takes 8-12 years from discovery to approval. AI shaves 2-3 years off the total timeline — significant but not the revolution headlines suggest.
Predictive Analytics
Epic Sepsis Model and similar tools:
AI analyzes patient vital signs, lab results, and medical history to predict deterioration before it happens. “This patient has a 78% probability of developing sepsis in the next 6 hours. Consider early intervention.”
Impact: Early warning systems reduce mortality for conditions like sepsis, cardiac arrest, and respiratory failure by 15-25%. The AI catches subtle patterns that humans miss in the flood of patient data.
The Comparison
| Tool | For | Cost | What It Does |
|---|---|---|---|
| Ada Health | Patients | Free | Symptom checking |
| Wysa | Patients | Free/Premium | Mental health CBT |
| Livongo | Patients | Insurance | Diabetes management |
| Apple Health | Patients | Free | Record aggregation |
| Dragon Copilot | Doctors | ~$369/mo | Clinical documentation |
| Aidoc | Hospitals | Enterprise | Radiology AI |
| Epic Sepsis | Hospitals | Enterprise | Predictive analytics |
What AI Gets Wrong in Healthcare
Bias in training data. AI trained primarily on data from white patients performs worse for other demographics. Skin condition AI that was trained on light skin misses conditions on dark skin. This is a serious, ongoing problem.
Over-confidence. AI presents probabilities as certainties. “85% likelihood of condition X” sounds definitive but means 15% of the time it’s wrong. Patients and sometimes doctors treat AI output as more certain than it is.
Missing context. AI knows your symptoms and test results. It doesn’t know you’re going through a divorce, that your mother died of the same condition, or that you’re terrified of hospitals. Context matters in medicine, and AI doesn’t have it.
Regulatory lag. AI capabilities advance faster than FDA approval processes. Tools that could help patients today are stuck in regulatory review. This protects safety but delays access.
The Patient AI Stack
Here is the practical recommendation by budget.
Free Stack ($0)
- Ada Health: symptom checking
- Wysa (free tier): mental health support
- Apple Health: record management
- ChatGPT/Claude Free: health questions (with appropriate skepticism)
Tracking Stack ($6-15/month)
- Oura Ring ($6/mo + hardware): sleep and recovery tracking
- Wysa: mental health support
- Apple Health: record management
- Good for anyone who wants passive health monitoring without committing to a full wellness routine. The Oura data alone gives you a surprisingly clear picture of how your lifestyle choices affect your body.
Wellness Stack ($20-50/month)
- Oura Ring ($6/mo + hardware): sleep and recovery
- Headspace ($13/mo): meditation
- Claude Pro ($20/mo): detailed health questions
- Total: ~$39/month
Using AI for Health Questions
Claude and ChatGPT handle health questions well, with important caveats:
"This evaluation comes from a 35-year-old male with no major health conditions.
There have been persistent lower back pain for 2 weeks after
starting a new desk job. No numbness or tingling.
Pain is 4/10, worse in the morning.
What are the likely causes? What can I do at home?
When should I see a doctor?"AI provides structured, medically-informed guidance. But always:
- Verify with a doctor for anything serious or persistent
- Don’t self-diagnose based on AI alone
- Seek emergency care for severe symptoms regardless of what AI says
- Remember AI isn’t your doctor and doesn’t know your full medical history
The Bottom Line
AI in healthcare is most useful at the extremes: simple questions that don’t need a doctor (symptom checking, wellness tracking) and complex analysis that exceeds human capacity (medical imaging, predictive analytics). The middle ground — diagnosis, treatment decisions, patient relationships — remains firmly human.
For patients: use AI tools for information and monitoring, not for diagnosis or treatment decisions. For providers: use AI to handle documentation and flag urgent cases, freeing time for the human judgment that medicine requires.
You might also like: AI healthcare tools.
Related guide: AI healthcare tools.
Related guide: AI healthcare tools.