The CY 2027 Fee Schedule Final Rule and AI: What CMS Now Allows in the Wellness Visit, the New MIPS AI Activity, and the RPM Delay

CMS released the CY 2027 Medicare Physician Fee Schedule final rule on November 4, 2026, and it appears in today's Federal Register at 91 FR 70992 (docket CMS-1848-F, effective January 1, 2027). It runs more than a thousand pages and drew about 43,100 comments.

Medicare will not pay separately for AI next year. If your practice was hoping for an ambient-documentation code, it is not in this rule. Still, the final rule says more about AI in everyday practice than any fee schedule before it. CMS lists, element by element, which parts of the Medicare Annual Wellness Visit an AI tool may help with and which it may not. It adds a MIPS improvement activity for clinician use of AI. And it gives practices that use vendor-staffed remote monitoring two more years before the outsourcing ban starts.

We read the rule text. This post covers what was finalized, what stayed a question, and what a small practice should change before January.

The short version

• No new AI payment. CMS took comments on paying for "technology-enabled" primary care and finalized nothing. Most commenters opposed separate payment for AI used in wellness visits.
• The AWV clarification is the practical news. AI may administer the health risk assessment, pull history from the chart, reconcile medications and providers, run screening instruments, and draft the prevention plan. The clinician must review and approve each of those. CMS says AI has no role in routine measurements or the cognitive assessment.
• New MIPS activity: IA_AHW_4, "Clinician Use of Artificial Intelligence (AI) to Improve Patient Care." It is finalized as proposed and starts with the 2027 performance period. It requires written organizational policies for evaluating, deploying, and monitoring AI tools.
• Outsourced RPM/RTM staffing is delayed to January 1, 2029. The new initiating-visit requirement, and the rule that RTM is for established patients only, still start January 1, 2027.
• "SaaS" is now "SaMS" (Software as a Medical Service). Fourteen algorithm-only lab analyses move from the lab fee schedule to the physician fee schedule.
• The conversion factor falls further than proposed. It drops to $32.9548 for qualifying APM participants (−1.83%) and $32.6286 for everyone else (−2.31%). The proposed figures were $33.1693 and $32.8409.
• No new efficiency cut in 2027. The 2.5% reduction from 2026 stays in the rates. CMS will next recalculate it in CY 2029 rulemaking.

Where AI fits in the Annual Wellness Visit

This part of the rule matters most for primary care practices, NPs, and small clinics.

In the proposed rule, CMS asked whether AI could make the AWV less of a one-time checklist and more of a continuing prevention plan. It also asked what stops an "AI technology company" from delivering AWVs through an affiliated Medicare-enrolled provider. Commenters said AI could help with specific AWV elements but should not replace the visit. CMS used the final rule to set out how AI fits within the AWV rules that already exist, at section 1861(hhh) of the Act and 42 CFR 410.15.

Two principles come first. Any use of AI in the AWV, CMS writes, "must preserve clinician autonomy in the practice of medicine." And the AWV must still be performed by "a physician or other health professional, or team of medical professionals directly supervised by a physician." The software helps. The clinician still furnishes the visit.

Here is CMS's list, element by element:

• Health risk assessment (HRA): AI can administer the questionnaire, collect updated history and current-provider information, and summarize it before the visit. The clinician then reviews and adds to it. The statute already allows the HRA to be completed before the visit, by the patient, through an "interactive telephonic or web-based program," as long as privacy is protected.
• Medical and family history: AI may extract it from the record. The clinician verifies it and establishes or updates it during the encounter.
• Current providers and suppliers: AI may reconcile the list. The clinician verifies it.
• Medications, supplements, and other substances: AI may reconcile the list. The clinician reviews and verifies it.
• Routine measurements (weight, waist circumference, blood pressure): no role for AI. CMS says it is not aware of how AI could take these measurements and does not describe a role for it.
• Cognitive assessment: no role for AI. The regulation requires direct observation, and CMS says it is not aware of how AI could perform a cognitive assessment under that standard. Commenters also named digital cognitive assessment tools as not ready for reliable AWV use.
• Depression risk review: AI may administer a screening instrument the clinician selected and flag responses before the visit, for the clinician to review.
• Functional ability and safety: AI may run screening questions the clinician selected, with results reviewed during the encounter.
• 5–10 year screening schedule: AI may generate a recommended list. The clinician reviews and prioritizes it.
• Personalized prevention plan: AI may draft it, including health advice and referrals. The clinician reviews and approves it.
• Advance care planning: at the patient's discretion, AI may prompt the conversation (advance directive, caregiver identification, future care decisions). The clinician conducts the discussion.

The pattern repeats: AI prepares and the clinician decides. Interpretation, judgment, counseling, and approving the plan stay with the practitioner. Nothing in this list allows AI to complete an AWV.

What this does not do

It does not create new payment. Your AWV still bills as G0438 or G0439, and there is no add-on or adjustment for software help. CMS acknowledged that AI may shift the resources a wellness visit takes. It also signaled possible future rulemaking, including opt-out procedures for patients who do not want new technology used in their AWV.

It does not open the AWV to AI companies. CMS asked what blocks AI-company-led AWV models and then restated the existing requirements. It did not finalize any change to who may furnish the visit.

It also comes with a warning about patient apps. If an organization gives a Medicare beneficiary a software app to help carry out their prevention plan, CMS says that "may implicate both the Federal Anti-Kickback Statute and the Beneficiary Inducements Civil Monetary Penalty Law" and would be evaluated case by case. If a vendor offers your patients a free app as part of an AWV program, get that reviewed before you launch it.

How this connects to coding intensity

When CMS asked how to judge AI-enabled AWVs, it asked whether they improve outcomes "rather than merely increasing AWV volume, documentation completeness, coding intensity, or low-value care follow 'cascades.'" That is the same concern behind the payer and OIG scrutiny of AI scribes. If a tool's sales pitch is that it captures more AWV elements per visit, expect auditors to ask whether patients ended up better off.

A practical AWV workflow under the final rule

If you already use AI patient intake or an ambient scribe, this workflow stays within what CMS described:

1. Before the visit: send the HRA, depression screen, and functional/safety questions through your intake tool. Use instruments your clinicians chose, not ones the vendor defaults to. Intake platforms like Phreesia and workflow platforms like Notable already collect pre-visit questionnaires. The question is whether a clinician reviews the results.
2. Pre-charting: let the AI draft the history, medication list, and provider list from the chart, then have the clinician verify them in the visit. The pre-charting post covers how often chart summaries leave things out. That is why CMS keeps saying "for review and verification by the clinician."
3. In the room: take vitals and do the cognitive assessment the usual way. The rule leaves no role for AI in either.
4. After the visit: an AI-drafted prevention plan and screening schedule are fine, but the clinician approves them before the patient receives them. Keep that approval visible in the note.
5. Documentation: record what the clinician reviewed, changed, and approved. If an auditor later questions an AI-assisted AWV, a note showing clinician judgment is your defense.

Patient consent and recording-disclosure rules still apply to any ambient tool in the room. See our consent and recording-law guide and the 2026 state AI law roundup.

The new MIPS activity: IA_AHW_4

CMS finalized six new improvement activities beginning with the CY 2027 performance period / 2029 MIPS payment year. One is IA_AHW_4, "Clinician Use of Artificial Intelligence (AI) to Improve Patient Care," in the Advancing Health and Wellness subcategory. CMS finalized it as proposed.

As CMS describes it, clinicians "establish policies for evaluating and monitoring AI tools or participate in developing and refining AI-enabled resources." The examples CMS gives cover what many practices already do:

• AI that summarizes medical literature for clinical decisions
• AI-assisted documentation ("AI-generated notes with clinician review")
• Identifying care gaps for population health
• Determining clinical-trial eligibility
• Drafting responses to patient questions for clinician review

In the final rule, CMS names the core requirement: participants must "establish and maintain organizational policies and procedures for AI use in healthcare settings that guide the evaluation, deployment, and monitoring of AI-enabled tools." Some commenters asked for stricter clinical-validation and error-detection requirements. CMS kept the activity broad "at this time of rapid AI development."

Why it matters for a small practice: traditional MIPS usually requires attesting to two improvement activities. Clinicians with a special-status designation, which covers small practices, and all MVP participants need only one. If you already run an AI scribe or AI inbox drafting with clinician review and you write the governance policy, this activity may cover your whole improvement-activities requirement.

The catch is the policy. A one-line statement that you use AI will not hold up. At a minimum, write down:

• Which AI tools are in use, who approved them, and for what purpose
• How you evaluated each one before deployment (accuracy, privacy and BAA, bias where relevant)
• The review rule (for example, "every AI-drafted note and patient message is reviewed by the signing clinician")
• How you monitor tools after go-live, and who decides when to turn one off
• How patients are told about AI use, if your state requires it

The rollout guide and HIPAA basics for AI tools cover most of this. Keep the policy dated and keep evidence that you followed it, because attestations can be audited.

This activity has an ironic backdrop. On October 28, ONC's HTI-5 final rule removed the federal AI model-card requirement from certified EHRs, and that was much of the vendor-side transparency a practice could draw on for this evaluation. CMS now rewards practices for evaluating their AI tools while ONC is removing the documentation that helps them do it. Ask your vendors for that information directly.

Remote monitoring: the outsourcing ban moves to 2029

The proposed rule would have stopped Medicare payment for RPM and RTM furnished by third-party staff on January 1, 2027. Many digital-health and AI-monitoring companies supply the nurses or care coordinators who handle the monthly calls, so that would have upended their model.

CMS kept the policy but delayed it two years:

• Through December 31, 2028: contracting out RPM/RTM clinical staff remains allowed. All existing "incident to" rules still apply, and CMS says practitioners should not bill when "there is little oversight by the billing practitioner or a lack of clinical integration" with the third party.
• From January 1, 2029: RPM and RTM time counts only when furnished by clinical staff who are a direct employee of the practitioner or the practice. CMS says it will publish sub-regulatory guidance before then on which staffing arrangements qualify.
• Technology vendors are fine. CMS says practices "may obtain devices, software, connectivity, data hosting, cybersecurity, implementation support, technical assistance, and interoperability services from outside companies." The restriction covers outsourced staff time, not outsourced software.

Two other remote-monitoring changes do take effect January 1, 2027:

• RTM is limited to established patients, as RPM already was.
• A separately reportable initiating visit is required before RPM or RTM starts. It is needed once, at the onset of monitoring, not for a device change, a clinical change, or a switch to another billing practitioner in the same practice. Patients already enrolled before January 1, 2027 are grandfathered. CMS also warns against "any misuse of enrolling beneficiaries in remote monitoring services" before the deadline.

If you have a vendor-staffed RPM program, you have two years, not two months. Use that time to price what in-house staffing would cost. Don't sign a contract that runs past 2028 without an exit clause tied to this rule.

SaMS: CMS renames clinical software

CMS finalized replacing "Software as a Service (SaaS)" with "Software as a Medical Service (SaMS)." The term covers "software-based technologies that support clinical decision making through algorithmic analysis, including those that provide clinical or diagnostic functionality." One commenter asked whether this was only a name change. CMS said it clarifies "the distinction between clinical algorithms used to furnish these services and generic cloud tooling," and that it will keep refining the definition.

The substantive change is narrower. CMS treats stand-alone algorithmic analyses of existing lab data as "other diagnostic tests," not clinical diagnostic laboratory tests. It is moving 14 of these codes from the Clinical Laboratory Fee Schedule to active pricing under the PFS, priced from crosswalked OPPS rates. One example is 0220U, an AI image analysis of breast cancer histology that produces a recurrence score. CMS said it may expand the policy later.

For most practices this has no direct effect. It matters if you order or bill algorithm-only diagnostics, and it signals how CMS plans to classify software-driven diagnostic services. For context on which AI services Medicare pays for today, such as 92229 autonomous retinal screening and 75577 coronary plaque analysis, see Does Medicare Pay for AI?

What CMS said about AI tools in general

The primary care request for information is worth reading for how CMS describes the market. Citing outside reports, CMS put AI scribe adoption at roughly 25% of U.S. physicians. It called OpenEvidence, used by more than 40% of physicians by their own report, "likely" the most widely used AI tool in primary care. And it said AI care-delivery agents "have not yet reached widespread use."

Commenters were split. Many were "broadly skeptical of AI use in health care" and asked CMS to pay only for proven value. Many others supported tools that summarize longitudinal records, find care gaps, and automate questionnaires and documentation. On privacy, commenters argued that HIPAA and HITECH already cover AI tools and that a separate AI-specific regime would duplicate them. They also mostly rejected CMS's idea of a "two-track" care-management code set, with "technology-enabled" codes separate from "traditional" ones.

For each of these questions, CMS's response was "We will consider these comments for future rulemaking." That leaves several things open for CY 2028: separate categories of E/M visits (longitudinal, acute, consultative), prospective primary care payment through the Shared Savings Program, and some way to value technology-enabled care. None of them is decided.

A note on terminology: following Executive Order 14434, the rule uses "Super Intelligence (SI)" alongside "AI." It means the same thing.

The money: conversion factor and the efficiency adjustment

• Conversion factor: $32.9548 for qualifying APM participants and $32.6286 for non-qualifying clinicians, down from $33.5675 and $33.4009. Both cuts are deeper than proposed.
• Efficiency adjustment: CMS applied a 2.5% cut to work RVUs and intraservice time for most non-time-based services in CY 2026, on the theory that technology and workflow gains make care more efficient. The policy recalculates every three years, so 2027 brings no new cut. The 2026 reduction stays in the rates, and new 2027 codes were valued as if already adjusted. The next recalculation comes in CY 2029 rulemaking.

Put together: rates go down, there is still no AI code, and the "efficiency" assumption is already built in. The argument for buying an AI tool still rests on time saved, fewer denials, and clinician retention, not new revenue. Our scribe pricing breakdown and free and low-cost options are a good place to run those numbers.

Telehealth: modifiers BB and BC

Section 6209 of the Consolidated Appropriations Act, 2026 requires two new telehealth modifiers starting January 1, 2027, and CMS created them. BB marks a telehealth service furnished through a virtual platform when the practitioner contracts with the platform's owner or has a payment arrangement for using it. BC marks a telehealth service furnished incident to a practitioner's professional service. Neither modifier changes payment. CMS posts guidance on its telehealth page. If you see patients through a third-party telehealth platform, read that guidance before January. Our telehealth scribe post covers the rest of the 2027 telehealth picture.

Your checklist before January 1, 2027

• AWV: map each AI-assisted step to CMS's element list. Confirm that a clinician reviews the HRA, history, medication list, screening schedule, and prevention plan, and that the note shows it. Keep vitals and the cognitive assessment human.
• Patient-facing apps: if a vendor gives patients free software as part of a wellness program, get Anti-Kickback and Beneficiary Inducements CMP review first.
• MIPS: if you plan to attest to IA_AHW_4 for 2027, write and date your AI governance policy now, and keep evidence that you follow it.
• RPM/RTM: schedule initiating visits for new monitoring patients starting January 1, check that RTM patients are established, and plan for employed staff by 2029.
• Budget: rerun 2027 revenue on the final conversion factor, not the proposed one.
• Vendors: ask each AI vendor for the evaluation information your policy needs. After HTI-5, EHR vendors no longer have to publish it.

If you're comparing tools, start with our family medicine scribe comparison, the small-clinic comparison, or browse tools for nurse practitioners. NPs running AWVs should also read our guide to incident-to and split/shared billing with AI scribes.

---

This article is informational only and is not legal, coding, billing, or medical advice. It summarizes the CY 2027 Physician Fee Schedule final rule (91 FR 70992, published November 9, 2026) as of that date. CMS may issue corrections, sub-regulatory guidance, and MAC instructions that change how these policies apply. Verify requirements with CMS, your Medicare Administrative Contractor, and qualified counsel before changing your billing or compliance practices.