Doctors Are Using AI More Than Ever. So Why Is Pajama Time Still the Same?
Physician AI adoption more than doubled in three years. It went from 38 percent in 2023 to 81 percent in 2026, according to the American Medical Association's 2026 Physician Survey on Augmented Intelligence. That is one of the fastest technology adoption curves ever recorded in medicine. But the problem AI was supposed to fix has barely moved. Physicians still chart from home after clinical hours. AMA data shows 20.9 percent of physicians spend more than eight hours a week on after-hours EHR work. That is the same figure recorded in 2022, before most of this adoption happened.
What Is Actually Driving Physician Burnout?
Documentation, more than any other single factor. For every hour of direct patient care, physicians spend roughly two more hours on electronic health record data entry. This finding comes from research cited in Doximity's 2026 State of AI in Medicine Report. Over a full clinical week, that adds up to 15 to 20 hours of admin work. Much of it happens on personal time, not clinical time. The AMA's survey identifies excessive administrative tasks, led by documentation, as the top driver of burnout. It beats every other factor measured by a wide margin.
Does AI Actually Reduce Documentation Time, or Just Move It Around?
The honest answer depends on how AI gets deployed. A 2025 study in JAMA Network Open followed clinicians using an ambient AI scribe. This is a tool that listens during a patient visit and drafts the clinical note on its own. After 30 days, measured burnout dropped from 51.9 percent to 38.8 percent. After-hours documentation time also improved. This is not a vendor claim. It is a controlled clinical study. And it points to something specific: general AI adoption does not reduce burnout on its own. Ambient documentation, used correctly and checked by the physician, does.
This distinction matters because most current adoption is not ambient scribing. Doximity's data shows physicians mainly use AI for literature search, at 35 percent of use cases, and voice-based documentation at 29 percent. Both are useful. But neither one, alone, is the exact intervention the JAMA study measured. That gap between broad AI use and this specific, studied tool likely explains why pajama time has not moved, even as adoption doubled.
Why Do 71 Percent of Physicians Still Not Trust AI Accuracy?
Because the concern is well founded. It is not just resistance to new technology. Doximity's 2026 report found that 71 percent of physicians cite accuracy and reliability as their top concern about AI tools. That skepticism is a feature, not a flaw, in a field where a wrong suggestion has real clinical consequences. The physicians using AI most successfully are not the ones who trust it blindly. They are the ones who use it for narrow tasks, like documentation and literature summaries, where a human checks the output before it enters a medical record.
What Should a Practice or Health System Actually Prioritize?
Start with the specific tool the data supports. Do not roll out general-purpose AI and hope it helps. Ambient scribing tools, paired with a mandatory physician review step, have the clearest evidence behind them for cutting after-hours work and burnout. Literature search and summary tools are a lower-risk starting point for building trust, since errors there are easier to catch early. What the data argues against is deploying one general AI tool system-wide and calling it a burnout fix. Adoption and burnout reduction are not the same outcome. The last three years show they do not move together on their own.
For a health system leader, the test is simple. Track after-hours EHR time before and after any AI documentation rollout. Track that number specifically, not general satisfaction scores. If it does not move within 90 days, the rollout needs a redesign. It does not need more training.
The lesson from three years of data is not that doctors are bad at adopting AI. They adopted faster than almost any profession on record. The lesson is that adoption speed and outcome improvement are two different projects, and only one of them has been measured well so far.
There is a governance angle here too. As ambient scribing and clinical AI tools spread, health systems need clear rules about what gets automatically drafted versus what always requires a human first pass. Practices that write that policy down now, before an error forces the conversation, are the ones avoiding both the burnout problem and the liability problem at the same time.