How AI Is Used in Healthcare Today
Artificial intelligence is already part of healthcare, most visibly in medical imaging, but also in triage, documentation, scheduling and billing. The U.S. Food and Drug Administration maintains a list of AI-enabled medical devices that are authorized for marketing in the United States, intended to give providers and patients transparency about which devices use AI. The FDA notes that devices on the list have met its applicable premarket requirements, including a focused review of safety and effectiveness, and that the list is not comprehensive (FDA, List of Artificial Intelligence-Enabled Medical Devices, read October 9, 2026). This guide explains the main categories of healthcare AI, how to evaluate a tool, and the risks to manage.
Clinical Uses
- Imaging analysis: software that helps detect or prioritize findings on X-rays, CT, MRI and other images. Many FDA-authorized AI devices are in this area.
- Clinical decision support: tools that flag patients at risk of deterioration or suggest guideline-based next steps.
- Monitoring: analysis of data from wearables, heart monitors and remote patient monitoring devices.
- Pathology and dermatology: image-based analysis to assist specialists.
In all of these, the software supports clinicians; responsibility for diagnosis and treatment remains with licensed professionals.
Administrative Uses
- Documentation: ambient listening and speech recognition tools that draft clinical notes for clinician review.
- Scheduling and patient communication: chat tools that answer routine questions and manage appointments.
- Revenue cycle: coding assistance, claims checks and denial management.
These tools aim to reduce paperwork, but generated notes and codes still need human review for accuracy.
How to Evaluate a Healthcare AI Tool
- Regulatory status: if the tool functions as a medical device, check whether it appears in FDA databases and what its authorized intended use is.
- Evidence: ask for published studies or validation data, ideally in patient populations and settings similar to yours.
- Performance across groups: ask how the tool performs across ages, sexes, races and other groups, since performance can vary.
- Integration: confirm it works with your electronic health record and workflows.
- Privacy and security: review how data is stored and used, business associate agreements under HIPAA, and security certifications.
- Total cost: ask vendors for written quotes covering licensing, implementation, training and support; pricing varies widely by organization size and scope.
- Monitoring plan: decide how you'll track accuracy and problems after go-live.
Risks to Manage
- Errors and overreliance: AI outputs can be wrong; clinicians should verify rather than accept them automatically.
- Bias: tools trained on limited data may perform worse for some patient groups.
- Changes over time: performance can drift as practice patterns or patient populations change.
- Privacy: patient data used by AI tools must be protected under applicable law and contracts.
- Transparency: patients may want to know when AI is used in their care; the FDA list is one transparency resource.
Generative AI and Large Language Models
Large language models are being used to draft messages, summarize records and support documentation. The FDA has said it will explore ways to identify devices that incorporate foundation models, including large language models, to help providers and patients recognize when such functionality is present. For non-device uses such as drafting patient messages, organizations should set clear policies on review, privacy and acceptable use.
Getting Started in a Practice or Hospital
- Pick one clearly defined problem, such as documentation time or imaging backlog.
- Involve clinicians, IT, compliance and patient representatives early.
- Pilot with a small group, measure results against a baseline, and gather feedback.
- Scale only after the pilot shows benefits and acceptable risk.
Comparison: Categories of Healthcare AI
| Category | Example use | Key question |
|---|---|---|
| Imaging AI | Flagging findings for radiologist review | Is it FDA-authorized for this intended use? |
| Clinical decision support | Risk alerts | How accurate is it in our population? |
| Documentation AI | Drafting visit notes | How are errors caught and data protected? |
| Administrative AI | Scheduling, coding | Does it integrate with our systems? |
What Patients Can Ask
Patients don't need to understand algorithms to ask useful questions. It's reasonable to ask a provider whether software was used to help read an image or flag a result, whether a clinician reviewed the result, and how your information is protected. If a patient portal message or visit summary was drafted with AI assistance, you can ask who reviewed it. For consumer health apps that are not medical devices, read the privacy policy and remember that they are not a substitute for professional care.
Training and Change Management
New tools change workflows. Successful rollouts include training for every user group, clear guidance on when to rely on the tool and when to override it, a simple way to report errors, and regular review meetings during the first months. Measuring time saved, accuracy and staff satisfaction against a baseline shows whether the tool is worth keeping, expanding or replacing.
Governance
Many organizations set up a review group that includes clinicians, IT, compliance, privacy and patient representatives to approve new AI tools, set usage policies and monitor performance. A written inventory of AI tools in use, their intended purpose and who owns each one makes oversight much easier.
Frequently Asked Questions
Does AI replace doctors?
No. Current tools assist clinicians; licensed professionals remain responsible for care decisions.
How do I know if a device uses AI?
The FDA's AI-Enabled Medical Device List identifies many authorized devices that use AI, though the FDA notes the list is not comprehensive.
Is patient data safe with AI tools?
It depends on the vendor's practices and contracts. Review privacy terms, security measures and HIPAA agreements.
What does an AI tool cost?
Costs vary by vendor, scope and organization size; request written quotes that include implementation and support.
Questions to Ask Vendors
- What is the tool's intended use, and is it FDA-authorized for that use if it is a medical device?
- What data was it trained and tested on, and how does it perform in settings like ours?
- How are errors reported and fixed, and how are updates validated?
- Where is patient data stored, who can access it, and is it used to train other models?
- What support and training are included, and what are the total costs?
Key Takeaways
- AI is used in imaging, decision support, documentation and administration.
- Check FDA status, evidence and performance across patient groups.
- Keep clinicians in the loop and monitor tools after deployment.
- Protect patient data and be transparent about AI use.
Sources and References
- U.S. Food and Drug Administration, List of Artificial Intelligence-Enabled Medical Devices (read October 9, 2026).
Disclaimer
This article is general information, not medical advice. It cannot account for an individual's history, medications or risk factors. Anyone weighing a treatment, a test or a change in medication should speak with a qualified healthcare professional first, and treat urgent symptoms as an emergency rather than a research question.
Related articles
A note on sources: an earlier revision of this article carried figures attributed to named bodies that we could not open and check at the source. Rather than leave an unverifiable number — or leave the attribution standing without one — we state the point in plain terms. Checked 2026-10-07.