Much of a Medicare annual wellness visit can now happen before the beneficiary sees a clinician. Software can administer the health risk assessment, reconcile the medication list and draft a prevention plan. CMS has confirmed that this already fits inside the rules, as long as a clinician reviews and approves each piece.
The confirmation sits in the CY 2027 physician fee schedule final rule (CMS-1848-F), published in the Federal Register on November 9, 2026 and effective January 1, 2027. Most of that rule concerns payment rates. Both 2027 conversion factors fall, by 1.83% for qualifying APM participants and 2.31% for everyone else. In a section on primary care, though, CMS answered its own request for comments on technology by walking through the wellness visit element by element.
The rule uses “SI,” for super intelligence, in place of AI, citing Executive Order 14434. The meaning is the same. We use AI here.
The clinician stays on every element
The annual wellness visit was created by the Affordable Care Act and is defined at 42 CFR § 410.15. It produces a personalized prevention plan, and it must be furnished by a physician or other health professional, or by a team working under a physician’s direct supervision. CMS did not change that. It described what AI may do within it.
| Visit element | What CMS says AI may do |
|---|---|
| Health risk assessment | Administer the questionnaire and summarize answers before the visit |
| Medical and family history | Extract information from the record for the clinician to verify |
| Current providers and suppliers | Reconcile the list for the clinician to verify |
| Medications and supplements | Reconcile the list for the clinician to verify |
| Depression risk review | Administer a screening tool the clinician selected and flag responses |
| Functional ability and safety | Administer screening questions the clinician selected |
| Screening schedule | Generate recommended screenings for the clinician to prioritize |
| Prevention plan | Draft the plan for the clinician to review and approve |
| Advance care planning | Prompt the discussion, at the beneficiary’s discretion |
| Routine measurements | No role identified |
| Cognitive assessment | No role identified |
The pattern is consistent. AI collects, drafts, reconciles or flags. A clinician verifies, prioritizes or approves. CMS said any use must preserve clinician autonomy, and it said its clarifications do not replace the relationship between beneficiary and clinician.
CMS identified no AI role for 2 elements. It said it knows of no way for AI to take a blood pressure or measure weight. It also said the regulation requires cognitive function to be assessed by direct observation, and it is not aware of how AI could meet that standard. Commenters identified digital cognitive assessment tools as not yet ready for reliable use in the visit.
Under the clarification, AI may collect and draft. A clinician must verify and approve every element.
What CMS declined to do
CMS did not create new payment for AI-supported visits. Most commenters opposed separate payment, treating AI as an enhancement to the existing service. CMS said it would consider the broader questions in future rulemaking, including how to value technology-enabled primary care and whether to let beneficiaries opt out of new technology in the visit.
It also flagged a legal risk. An organization that gives a beneficiary an app to carry out the prevention plan may implicate the federal Anti-Kickback Statute and the beneficiary inducements civil monetary penalty law. CMS said those arrangements would be judged case by case. Vendors that bundle a patient app with a wellness visit service now have a written warning to plan around.
CMS asked a pointed question in the proposed rule. How should it tell whether AI-enabled visits improve outcomes rather than merely increasing visit volume, documentation completeness, coding intensity or low-value follow-up care? The final rule does not answer it. The question matters because the evidence on the wellness visit itself is mixed. CMS cited studies finding more preventive service use after a visit, and others finding no effect on screening or spending.
What it means for plans and practices
A wellness visit drafted partly by software produces the same kind of record as one drafted entirely by a person. The difference is invisible unless someone records it. That is a problem for anyone who later relies on the record.
For practices, the safest reading of the rule is literal. If AI extracted the family history, the chart should show that the clinician verified it. If AI drafted the prevention plan, the chart should show what the clinician changed or approved. A note that cannot show the review step is hard to defend in an audit, because the rule’s whole permission rests on that step.
For Medicare Advantage plans, the stakes run through the diagnoses a wellness visit records. A health risk assessment that asks better questions will surface more conditions, some real and some not yet confirmed. CMS named coding intensity as a concern on its own. Plans that pay for in-home or vendor-run wellness visits should ask which elements were AI-drafted and how the clinician confirmed each diagnosis before it reached a claim. We argued in an earlier piece that upcoding is a design flaw in risk adjustment, and AI drafting will make that flaw easier to exploit unless verification is recorded.
For vendors, the table above works as a specification. It names the clinician action that must follow each AI action. A product that logs the AI output, the clinician’s edit and the approval for every element will fit the rule. A product that produces a finished note will be harder to defend.
Redlake builds tools for utilization review, and the principle here matches ours. Software can prepare the record. A person has to stand behind what it says.