Will radiology techs be replaced by AI? Start with who does what
Most of the worry aimed at this job is really aimed at a different one. Will radiology techs be replaced by AI is a question about image reading, and image reading belongs to the radiologist. The technologist’s shift happens before any algorithm sees a pixel: greeting a patient who is in pain, checking the order against the body part, positioning a shoulder or a hip so the anatomy lines up, placing shielding, setting exposure, and holding the room calm for the two seconds that matter.
That work is physical, local and unpredictable. A trauma patient cannot lie flat. A child will not hold still. A portable chest film in the ICU has to be shot around lines, tubes and a bed rail. None of that is a prompt you can type. It is judgment applied with your hands, under time pressure, with radiation dose on your conscience.
Our split puts 61% of task time in work that still needs a person on site. The job is also not shrinking on paper: the Bureau of Labor Statistics counts about 230,490 radiologic technologists and technicians in the United States, with median pay of $80,110 and projected employment growth of 5% from 2025 to 2035 (BLS, 2025).
What software does, what it assists, what stays with the tech
The share of task time AI can handle on its own today is 17 out of 100 (higher means more of the work is covered). That coverage sits almost entirely on the screen side of the job: record and report handling, pulling prior studies, flagging an image as technically inadequate before the patient leaves the department, and the reconstruction and noise cleanup that happens inside the console.
A bigger slice is assistive rather than independent. Dose-optimization software suggests settings. Positioning aids and auto-collimation nudge the field. Scheduling and protocol tools sort the day’s list. In our task list that assisted group accounts for 39% of task time, and it changes how fast you work rather than whether you are there.
What is left is the part with a patient in it. Explaining a procedure to someone who is frightened. Immobilizing and positioning a body that does not cooperate. Watching for a reaction after contrast. Deciding, in the moment, that the image is not diagnostic and the patient needs one more exposure rather than a call-back next week. Those tasks do not get cheaper when the software improves.
How strong is the evidence?
Weaker than people assume. The evidence grade for this occupation is D, and a D means no direct, published head-to-head test of AI against a working technologist on this job’s tasks. So we publish no quality-parity number for it. Plenty of studies compare algorithms with radiologists on image interpretation, but that is the reading job, not the acquisition job.
What would settle it is specific: a measured comparison on positioning accuracy, repeat-exposure rates, patient dose and exam time, with and without an autonomous system, across trauma, pediatric and bedside work. Until something like that exists, treat confident claims in either direction with care. You can see how we grade evidence on the methodology page.
Good to know: the cost comparison on this page weighs software against wages, and leaves out the scanner, the shielded room and the service contract behind every exam.
When the picture could change
Most likely after 2042 (8 in 10 of our scenarios). What that window measures is explained on the replacement-year method page.
Two things could pull it earlier. First, self-positioning and auto-centering systems on fixed units are improving, and each step there shortens the hands-on part of a routine exam. Second, if hospitals push imaging volume onto fewer technologists with more automation per room, departments can grow throughput without growing headcount, which hits new graduates first.
Two things hold it back. Our robotics read places the hardware needed for the physical share of this job at the dexterous humanoid tier, which means a machine that can lift, angle and steady a human limb safely. That does not exist as a product you can buy for a radiology department. And the regulatory side is slow on purpose: state licensure, ARRT certification, radiation safety rules and liability all assume a qualified person is responsible for the exposure.
How to stay needed as a rad tech
Lean into the parts of the shift that only work with a person present. Difficult positioning, especially trauma, bariatric, pediatric and portable work. Patient communication and consent, including the patients who refuse, panic or cannot follow instructions. Image-quality judgment at the console, deciding what counts as diagnostic before the study leaves your hands.
Two skills compound from there. Cross-modality certification, because a tech who covers CT or MRI as well as plain film is harder to schedule around. And informatics literacy: knowing how AI triage flags behave, where they fail, and how to escalate when the software and the image disagree.
If you are weighing the next step, these jobs are close neighbors in equipment, patients and training:
- Magnetic Resonance Imaging Technologists
- Nuclear Medicine Technologists
- Diagnostic Medical Sonographers
You can put any two of them side by side on our job comparison tool, see the wider health technologist and technician family, or look at how exposure varies across the healthcare sector. For a broader view of hands-on roles, our list of jobs that mostly need a person shows where imaging work sits among them.