AI-Assisted Imaging Codes Are Live in 2026 — What Radiology Billers Need to Know Before Submitting
Radiology billing has never exactly been simple, but 2026 has added a new layer that a lot of practices are still catching up on. So finally, the implementation of new codes for imaging studies assisted by artificial intelligence (AI) is here and if your team hasn’t updated their workflows yet, you’re missing out on reimbursement or worse, falling into traps that aren’t even taken in by your providers.
This is NOT a minor coding update. It’s a true change in the way radiology works. Fractures are now routinely identified by algorithms, aneurysms are measured, stroke cases are triaged, and mammography is read with the help of algorithms. Payers are now getting up to speed to develop payment pathways for that work, but the rules around when and how to use those codes are specific and that is where billing teams go wrong.
Why New Codes Exist for AI-Assisted Imaging
AI tools in radiology languished in a weird billing no-man’s-land for years. There was no clean system in place for the radiologists to separate out the AI portion of the interpretation and bill separately for it, while using computer-aided detection and diagnosis software. Some practices included the cost of the professional fee. Others didn’t charge anything for it and simply made the software cost their business.
To fill this void, the CPT Editorial Panel at the American Medical Association (AMA) added category I and III codes that are linked to AI-assisted image analysis in multiple modalities, such as chest computed tomography (CT) for pulmonary nodules, CT for coronary artery calcium scoring, and AI-assisted stroke detection on computed tomography (CT) and magnetic resonance (MR) imaging. These are not random numbers that could be anything. Each one has specific clinical restrictions on when it is necessary to use the AI tool output to achieve a meaningful contribution to the interpretation, rather than as quality control.
That’s more important than most billers realize. The use of AI software if it is used for incidental detection but not documented in the radiologist’s report as adding to the report is probably not eligible for the new code. It is precisely this kind of activity that will attract the attention of the payers.
See also: Mortgage Broker Expertise to Help You Navigate Complex Home Loan Choices
What’s Actually Required for Clean Claims
Payment of these claims on the first pass depends largely on the discipline of documentation. Some of the things that payers always look for:
The radiologist should state in the report the findings of the AI tool and their impact on the overall report. While a general reference to AI software is good, it is not enough for most payers; they will want to know the results were reviewed by the physician and validated, or refuted.
The AI software must be FDA-cleared or FDA-approved for the specific clinical use that is being billed. No general-purpose AI can be used for an application outside of its intended use, and lack of efficacy does not mean that it can be used in that manner.
There continues to be inconsistent use of modifiers among payers. Depending on the imaging context, some commercial carriers are mandating modifier 33 or a similar modifier to signify a preventive service, while others have their own requirements that are not easily translated to the CPT guidance. This is one of the largest causes of denials at the moment and one thing that can’t be solved with any singular rule. It really needs to be verified by the payer.
Many of these codes have local coverage determinations (LCDs) that differ by MAC jurisdiction, meaning that claims that pass in one area may fail in another for the same clinical situation. Billing teams with multiple states should already be monitoring this at the jurisdiction level, rather than at the Medicare level.
The Denial Patterns Showing Up So Far
Initial practice experience with the codes at implementation suggests some common coding denials. By a wide margin, missing or unclear AI-integration language is the most common radiology report language missing. The second one is coding the AI assisted code with the standard interpretation code without differentiating modifiers (some payers interpret this as an add-on code since no differentiation is made between the two).
There is a timing aspect too. It appears that some of the claims are being denied because the payer’s payer system has not yet finished its coding system’s update, although the code is valid by the CPT. This is frustrating, but short-lived, and typically will be corrected upon a resubmission once the payer’s system has caught up – provided the original documentation was sound.
Getting AI Imaging Billing Right from the Start
The practices navigating this most smoothly aren’t relying on ad hoc fixes. They’re building the AI-code checkpoint directly into their coding workflow rather than treating it as an afterthought. That typically means a coder-level review step that specifically confirms three things before a claim goes out: the AI tool used is cleared for that clinical application, the radiologist’s documentation ties the AI output to the interpretation decision, and the payer-specific modifier rules for that plan have been checked, not assumed.
Given how much variation exists between commercial payers and Medicare Administrative Contractors, keeping a living reference sheet of payer-specific requirements for these codes is proving to be one of the more effective internal tools practices are using right now. It’s not glamorous, but it’s the kind of unglamorous groundwork that keeps denial rates down. For practices without the internal bandwidth to track this level of payer-specific detail, this is often where dedicated radiology billing services earn their keep, treating jurisdiction tracking and AI-code documentation as a core part of the coding process rather than a side task squeezed in between everything else.
Where This Is Headed
AI imaging billing is an emerging solution and there’s little doubt that CPT guidance will continue to be refined as payers learn more about claims and as providers respond to inconsistent adjudication. The practices that are currently taking these steps are the ones taking documentation and payer-specific documentation verification seriously, as a necessary part of the process, rather than an add-on.
Radiology groups that are deciding whether to do this inside or out should consider the bandwidth their coding team has for keeping track of payer changes by jurisdiction. This is a coding environment that requires continuous updating. It’s a long-term agreement, and it will probably remain so for the coming few billing cycles.
