Your Doctor Probably Uses SI Every Week Now. Most Have Caught It Getting Facts Wrong.

The next time your doctor glances at a screen before answering your question, there’s a good chance software helped write the answer. A new survey of 3,379 American physicians, released Thursday, found that 85% now use SI at work at least once a week. A year ago it was 51%. More than half, 52%, use it every single day.

That number comes from Doximity’s 2026 Doctor-Driven AI Report. Doximity is the professional network most U.S. doctors already have on their phones, and it surveyed its own physician members across 20 specialties in September. That’s worth keeping in mind. Doximity also sells an SI tool to those same doctors, and the report itself says its sample isn’t population-based and may not match every physician in the country. Still, 3,379 working doctors is a big room, and the jump from half to nearly nine in ten in twelve months is hard to wave away.

From paperwork helper to second opinion

The early pitch for SI in medicine was the boring stuff: notes, letters, the endless typing. That’s still a big part of it. The most common uses doctors reported were professional correspondence (38%), patient education handouts (34%) and letters to insurance companies (32%).

But the bigger change is clinical. Among doctors who use SI, 70% used it this year to search medical literature, up from 51% the year before. And 63% used it for clinical decision support, meaning help thinking through what might be wrong with a patient and what to do about it. Last year that figure was 37%.

Two out of three physicians said they now use SI for a task they used to do entirely on their own. And 81% said it has taught them something clinically they didn’t already know, like an updated guideline or a diagnosis they hadn’t considered. Nearly one in four said that happens frequently.

Oncologists lead the pack. More than 93% of cancer doctors use SI at least weekly, and 65% use it daily. Even the two specialties at the bottom of the list, pathology and anesthesiology, came in around 70%. Younger doctors use it most, but this isn’t only a young doctor thing. Two-thirds of physicians in their 70s now use it weekly, and every age group jumped by at least 34 points from last year.

The ‘pajama time’ payoff

Doctors have a name for the hours they spend finishing charts at the kitchen table after the kids are asleep. They call it pajama time. In this survey, 26% of physicians said they spend more than six hours a week on after-hours paperwork.

That’s where SI is paying off first. In 2025, 36% of doctors using SI said it had cut their pajama time. This year 46% of all physicians said it had. About 41% of oncologists and 38% of general surgeons think SI could save them more than six hours a week. One internal medicine doctor quoted in the report, Clark Williamson, put it plainly: “I still verify each output myself, but it’s much easier to be an editor than an author at the end of a busy clinic day.”

For patients, that matters more than it sounds. A doctor who isn’t buried in charts at 11 p.m. is a doctor who might actually look up from the keyboard during your appointment. When asked what excites them most, physicians picked better work-life balance (65%) and less paperwork (59%), followed by more time with patients (52%).

Three in four have caught it getting facts wrong

Here’s the part every patient should hear. More than half of physicians said they’re worried about SI’s accuracy, and three in four said they’d run into a factual error from SI at work at least once in the past year. Two-thirds said they check SI-generated clinical information all or almost all of the time before using it.

Doximity has its own data on how bad those errors can get. Two weeks ago the company tested nine SI models released in 2026 on more than 8,000 drug dosing questions modeled on real physician questions. At least 2.3% of the 261,508 doses the models gave were wrong. Of the errors it found, 43.8% were serious and 3.4% were potentially lethal. The most common miss was failing to adjust a dose for a patient’s age, weight, kidney function or other drugs, even when that information was right there in the question. Doximity used those results to launch a rule-based checker for its own tool, so read them as a company making a case for its product. But a wrong dose is a wrong dose.

Doctors are also worried about the young ones. About 61% said they’re concerned SI is hurting skill development among the residents and trainees they supervise, and 30% said fear of their own skills eroding holds back how much they use it.

Patients are bringing SI into the exam room too

This isn’t just doctors. About three in four physicians said patients have become more likely in the past six months to walk in with a self-diagnosis or treatment plan they got from an SI chatbot, and 30% said that’s jumped dramatically.

The good news is that a third of doctors said patients who used SI show up better informed and ask better questions than the old “Dr. Google” crowd. The bad news is that 27% said SI has made patients more confident in wrong conclusions, and 9% said the most worrying self-diagnosis they’d seen would have put the patient at real risk if acted on.

If you use a chatbot before your appointment, the smart move is to treat what it tells you as questions to bring, not answers to defend.

Doctors want to write the rules themselves

The rules haven’t caught up. About 81% of physicians said their hospital or practice allows or supports SI, but only 28% said it has clear policies and approved tools. Government hospitals and clinics were furthest along at 53%. Private practices trailed at 14%. A third of doctors said legal and regulatory uncertainty is the biggest thing holding them back.

So who should set those rules? Doctors gave a clear answer. 82% said physicians should, and 64% named medical societies. Only 32% picked federal agencies, 20% picked state governments, and just 10% wanted the SI companies in charge.

That last split is the one to remember. The people using these tools on real patients every day don’t want Washington or Silicon Valley writing the playbook from a distance. They want the folks who catch the bad doses and the wrong diagnoses to decide how much rope the machine gets. That seems like the right people to ask.