AI in Healthcare: What's Working in 2026

AI in healthcare has moved past the pilot stage. Most healthcare organizations now use it, and the projects paying off are narrow ones: clinical notes, imaging, scheduling and paperwork.

Results depend on how the tools are chosen and set up. Here's what the latest research shows, where AI is delivering results, and what a clinic or practice should get right before adopting it.

Most healthcare organizations now use AI

NVIDIA's State of AI in Healthcare and Life Sciences: 2026 Trends report surveyed more than 600 industry professionals in late 2025. Of those respondents, 70% said their organization actively uses AI, up from 63% a year earlier. Use of generative AI, the kind that writes and summarizes text, rose from 54% to 69%.

The biggest jump came from hospitals, doctors' offices and insurers, the part of the industry that has historically been slowest to adopt new technology. Their active use rose from 43% to 56%.

Individual doctors are following the same path. In the American Medical Association's 2026 physician survey, 81% of physicians said they use AI in their practice, more than double the 38% recorded in 2023.

The results show up in the numbers, too. In NVIDIA's survey:

• 85% of managers said AI has helped increase annual revenue.
• 80% said it has helped reduce annual costs.
• 85% expect their AI budget to grow in 2026.

Where AI is paying off

Clinical notes

AI scribes listen to a visit and draft the clinical note for the clinician to review. Canada now has strong evidence on how they perform.

Canada Health Infoway's national AI scribe program enrolled 12,019 primary care clinicians between June 2025 and February 2026. An independent evaluation by the Centre for Digital Health Evaluation at Women's College Hospital, released in June 2026, found that:

• Nearly 70% of responding clinicians reported less administrative burden.
• Clinicians saved an average of 1.11 hours a week during work hours and nearly an hour a week after hours.
• Same-day note completion rose from 55.9% to nearly 75% over six months.
• Over 80% felt more engaged with patients during visits.

Front desk and administration

Paperwork is where many providers start. In NVIDIA's survey, hospitals, doctors' offices and insurers named administrative tasks and workflow as their top AI use, at 52%.

Patient-facing tools are close behind. Among digital health companies, virtual assistants and chatbots that answer questions and book appointments were the top AI use, also at 52%, and the use where they reported the strongest return on investment.

Medical imaging

Imaging is where medical technology companies see the clearest return. Among medical technology companies in NVIDIA's survey, 61% named imaging their top AI use, and 57% said it has already delivered a return on investment.

AI agents are next

An AI agent goes a step further than a chatbot. Instead of answering a question, it completes a task: it books the appointment, updates the record or pulls together research.

According to NVIDIA, 47% of organizations are using or assessing agents, and 22% have already deployed one. The most common job is finding and organizing information (46%). NVIDIA expects agents to become a regular part of care workflows by 2027.

Most problems start with the rollout

When AI goes wrong in healthcare, the cause is usually how a tool was chosen, set up or supervised.

Take the 20 AI scribe systems approved for clinicians in Ontario, Canada. When the province tested them with simulated patient visits, evaluators recorded errors in the notes from most of them, according to a 2026 report by the province's Auditor General.

The invented details included referrals and blood tests that were never discussed. The mental-health gaps were found in at least one of two tests.

The audit's criticism was aimed at the selection process. Eleven of the 20 approved vendors hadn't submitted any third-party security audit reports or certifications, and five hadn't provided the required privacy and risk assessments. They were approved regardless.

The pattern is similar in the U.S., where the misuse of AI chatbots tops ECRI's list of health technology hazards for 2026. Its concern is general-purpose chatbots, which aren't designed or regulated for healthcare, being relied on for answers that affect care. A responsibly implemented chatbot is different: it has guardrails that keep it to the tasks it was built for, and it's thoroughly tested for safety and security before patients or staff rely on it. The same list warns about technology set up without enough attention to how care is actually delivered, which pushes staff into unsafe workarounds.

The fixes are straightforward: choose tools built and tested for clinical use, check vendors before you sign, and have a clinician read and correct every AI-drafted note before it goes into the record.

Doctors want these safeguards too. In the AMA survey, 88% said AI tools need thorough safety and effectiveness testing, 86% stressed the importance of data privacy, and 85% want a say in which AI tools their organization adopts.

The NVIDIA survey points to practical obstacles as well. Smaller organizations struggle most with budget (40%, twice the rate of large ones). Larger ones worry about data privacy and where data is stored (39%) and about regulation and ethics (37%).

Privacy law sets the rules for any tool that touches patient records. In Ontario, patient information is protected under the Personal Health Information Protection Act (PHIPA).

In the United States, the main federal law is the Health Insurance Portability and Accountability Act (HIPAA). It requires a clinic to have a written contract, called a business associate agreement, with any vendor that handles patient information on its behalf, such as an AI scribe provider. The contract sets out what the vendor may do with that information and requires the vendor to protect it.

On either side of the border, any AI tool that handles patient information needs clear answers on where data is stored and who can access it.

How to start in a clinic or practice

NVIDIA's clearest finding is that focused AI pays off. 47% of organizations now rank improving the AI they already use among their top spending priorities, up from 34% a year earlier. Finding new uses fell from 47% to 37%.

For a clinic, that suggests a simple approach:

• 1.Pick one task. Choose something frequent and time-consuming, such as visit notes, appointment booking or patient follow-ups.
• 2.Measure it first. Record how long it takes now, so you can judge the results.
• 3.Keep a clinician in charge. AI drafts and flags. A person reviews anything that affects care.
• 4.Check the vendor. Ask where patient data is stored, who can access it and what independent security testing the product has passed. In the U.S., get a signed business associate agreement first.
• 5.Train the team. Show staff what the tool does, where it falls short and how to correct it.

How Lumenta Digital helps

Our AI automation work starts with how your team spends its day, to find the task where AI saves the most time first. We build to a defined scope and agree on what success looks like before we begin.

A person approves anything with consequences. We work inside accounts you control, never feed your data to public AI models and build to respect PIPEDA and your confidentiality obligations.

Book a free consultation, and we'll show you where AI fits in your practice and where it doesn't yet.