Lesson objective
Distinguish the clinician review and signature duty, the coding professional query and documentation duty, the front desk minimum necessary decision and the manager ownership of administrative output, and build separate practice for each.
Plain-language explanation
Why this matters now
A session that treats "check the output" as one skill has merged four different duties that live under four different authorities and fail in four different ways. A clinician checking a drafted note is doing something quite unlike a front desk lead deciding how much of a record belongs in a message. The words in the objective look similar. The obligations underneath are not.
Merging them has a particular cost. The person whose duty is heaviest hears their obligation described in the general terms that suit somebody else's job, decides it does not apply with any force to them, and goes back to work.
Here are the four, each with what actually stands behind it.
Professional guidance (not binding by itself): Federation of State Medical Boards, "Navigating the Responsible and Ethical Incorporation of Artificial Intelligence into Clinical Practice", adopted by the FSMB House of Delegates in April 2024. It says state medical boards "do not regulate tools or technologies, only the licensed physicians that use those tools"; that the physician "is ultimately responsible for the use of AI and should be held accountable for any harms that occur"; that "failure to apply human judgement to any output of AI is a violation of a physician's professional duties"; and that physicians "retain their duty to review records created with AI", with AI-generated records lacking proper oversight likely to produce inaccurate documentation for which the physician will be accountable. It reports that failure to maintain adequate medical records was the basis for 6% of all state medical board disciplinary actions from 2015 to 2019. It is guidance and recommendations to state medical boards and licensees. It binds nobody by itself, and has effect only where a state board adopts rules or applies its existing disciplinary authority. As of 2026-09-15 no state medical board AI rule and no AI disciplinary case against a physician is recorded here; do not let a room's discomfort with that silence turn into an invented rule.
Professional standard (binds AHIMA members and credential holders): AHIMA Standards of Ethical Coding, the 2016 version linked from AHIMA's ethics page. The Standards apply to coding professionals "whether credentialed or not", including coding staff, auditors, educators and clinical documentation improvement professionals. Standard 3 says assign and report "only the codes and data that are clearly and consistently supported by health record documentation". Standard 4 says query or consult the provider for clarification before final code assignment, and 4.4 says not to query when no clinical information in the record necessitates a query. Standard 5.3 says not to misrepresent the patient's clinical picture through incorrect coding, omission or unsupported codes to increase reimbursement, justify medical necessity, improve publicly reported data or qualify for coverage. Standard 9.2 says to "Recognize that computer assisted coding (CAC) and/or electronic encoders are only tools and are not a substitute for the coding professional's judgment." The document predates generative AI and refers to computer-assisted coding and encoders, so applying 9.2 to a large language model tool is an inference, and a session must say it is one. It is enforceable through AHIMA processes for its members and credential holders, and is not law.
Regulation (binds covered entities and business associates): 45 CFR 164.502(b) and 164.514(d). Section 164.502(b)(1) requires a covered entity or business associate, when using or disclosing protected health information or requesting it from another covered entity or business associate, to "make reasonable efforts to limit protected health information to the minimum necessary to accomplish the intended purpose". Section 164.514(d)(2) requires identifying the persons or classes of persons in the workforce who need access, the categories of information they need and any conditions of access, and limiting access accordingly. Section 164.514(d)(3) allows routine and recurring disclosures to be limited by policies or standard protocols, while all other disclosures need criteria and individual review.
The fourth duty has no external authority, and that is the point. Administrative output belongs to the practice manager or another named administrative owner because the practice says so in writing. A session that cannot point to a rule sometimes treats the administrative owner's duty as optional. It is not optional; it is simply local.
A question to sit with
A biller and a front desk lead both tell you, in the same session, that they "check the AI output before it goes". They are describing two different acts, against two different sources, with two different failure modes. What did your slide say that let them both nod at the same sentence?
What goes wrong
1. The general verb that covers four duties. "Verify before sending" is heard by the clinician as reading a note, by the coder as glancing at a suggestion, by the front desk as a spelling check, and by the manager as somebody else's job.
2. The clinician's duty softened. An instructor says a clinician "should probably glance at" a drafted note. FSMB's language is stronger and more specific: the duty to review records created with AI is retained, and failure to apply human judgement to any AI output is described as a violation of a physician's professional duties. Quote it as guidance to boards, but quote it at full strength.
3. The coding duty taught as code selection. The instructor starts explaining which code fits a fictional scenario. That is not this programme's content, it is not an instructor's role in front of a room, and CPT descriptors are AMA copyright and are never reproduced. The teachable duty is different: a code rests on documentation that clearly and consistently supports it, and a query goes to the provider when the record needs clarification.
4. The query taught as a way to get a better code. Standard 4.4 says not to query when no clinical information in the record necessitates one. A session that presents querying as a revenue technique has taught the opposite of the standard.
5. Minimum necessary taught as a slogan. The front desk duty is a decision about how much of a record travels, made every time a message is drafted. Teaching it as a principle rather than as a decision leaves nothing to practise.
6. The manager's duty left to goodwill. Nobody is named, so nobody samples anything, and the first review happens after a complaint.
The practical answer
Build four separate practice rounds, each on its own duty, each on fictional material.
| Duty | Who owns it | The act practised | What it is checked against | The classic failure |
|---|---|---|---|---|
| Review and signature | The licensed treating clinician | Read the fictional record, then the drafted note, then sign or return it | The record itself, including what the draft omitted | Signing a tidy paragraph that contains a detail the record does not carry |
| Code assignment and query | The coding professional, working from the clinician's documentation | Decide whether a fictional record clearly and consistently supports the reported code, and whether a query is warranted | The documentation, and Standard 4.4 on when not to query | Querying to improve a figure rather than to clarify the record |
| Minimum necessary | The workforce member making the disclosure, under the practice's policy | Cut a fictional drafted message down to what the purpose needs | The purpose of the message, and the practice's protocol for routine disclosures | Pasting a whole fictional summary because the tool produced it |
| Administrative ownership | The practice manager or named administrative owner | Sample fictional AI-assisted outputs against the written rule and record the result | The practice's own written rule | No named owner, so no sample ever happens |
Best practice: teach the boundary between the four out loud. Ask each round who owns the output and what would happen if the wrong role performed the act. The answer is not always dramatic, and saying so honestly protects the times when it is.
Best practice: give the room the honest silences. A session that teaches the coding duty will be asked what CMS says about AI-suggested codes. As verified on 2026-09-15, no CMS rule addressing AI-suggested codes is recorded here, and the duty does not depend on one: the code rests on documentation and judgement regardless. Say the silence plainly rather than inventing a rule to fill it.
Remember this
- Four duties, four owners, four failure modes: review and signature, code and query, minimum necessary, administrative ownership.
- FSMB is guidance to state medical boards, binding only where a board adopts or applies it, and no state board AI rule or AI disciplinary case is recorded as of 2026-09-15.
- AHIMA's Standards bind members and credential holders, cover coding professionals whether credentialed or not, and predate generative AI, so applying Standard 9.2 to a large language model is a labelled inference.
- Minimum necessary is a decision made message by message under 45 CFR 164.502(b) and 164.514(d), not a slogan.
- No instructor teaches code selection, reproduces a CPT descriptor, or rules on a real claim.
Medical practice example
The fictional Larkspur Orthopedic Associates PLLC asks Ifeoma Castellanos to run one session covering "checking AI output" for eleven staff. She refuses the single session and proposes four rounds inside one hour, each on its own duty and its own fictional material. For the two surgeons and the physician assistant, she builds a fictional operative day record and a drafted note that contains a laterality detail appearing nowhere in the fictional record. Six of the seven reviewers catch it. For the two billers and the part-time coder, she builds a fictional chart in which the documentation supports less than the reported service, plus a second fictional chart where nothing in the record raises a question at all. The teachable moment is the second one: two of the three want to send a query, and Ifeoma reads them Standard 4.4, which says not to query when no clinical information in the record necessitates a query. Nobody in the room selects a code, no CPT descriptor appears anywhere in her material, and when the billing lead asks her to rule on a real claim from that morning, Ifeoma declines and says the determination rests on the treating clinician's documentation and the coding professional's judgement at that practice. For the two front desk staff and the referral coordinator, she supplies a fictional drafted message 340 words long answering a one-line question about a fictional appointment, and asks them to cut it to what the purpose needs; the shortest surviving version is 31 words. For the practice manager, she supplies ten fictional administrative outputs and the practice's own draft written rule, and the round is simply performing the sample and recording the result. In the debrief, the physician assistant says the four rounds made it obvious that his duty and the biller's duty are not the same duty, which is the sentence Ifeoma designed the hour to produce. The managing partner asks what CMS says about AI-suggested codes; Ifeoma says she has no CMS rule addressing that, that she will not invent one, and that the duty stands on documentation and judgement regardless.
Responsible-use warning
Never merge the four duties into one instruction to verify: the clinician, the coding professional, the front desk and the manager are doing different acts against different sources. Never teach code selection, never reproduce a CPT descriptor, and never rule on a real claim from the front of the room. Never present FSMB guidance as a rule every state has adopted, and never let a question about state board AI rules be answered with an invented one. Never present querying as a way to improve a figure: Standard 4.4 says not to query when nothing in the record necessitates it.
Human-review requirement
The instructor owns the accuracy of every duty described and the fictional material each round uses. A licensed clinician at the practice reviews any material touching the record or the signature before it is used. The coding professional or billing lead reviews the coding round to confirm it teaches documentation support and query practice rather than code selection. The named privacy official reviews the minimum necessary round. The practice owner decides the written rule behind the administrative ownership round, and owns compliance for the practice.
Practical exercise
Build four separate practice rounds for one fictional practice, one per duty, on fictional material only.
- Name a fictional practice and list the roles attending, with a count.
- Round one: write a fictional record extract and a drafted note that contains one detail the record does not support. Write the review act you will ask clinicians to perform and the two sentences you will use to state the FSMB position accurately, including its status as guidance to boards.
- Round two: write two fictional charts, one where documentation does not clearly and consistently support the reported service, and one where nothing in the record necessitates a query. Write the question you will ask about each. Include no code and no CPT descriptor anywhere.
- Write the exact sentence you will say when a learner asks you to decide a code for a real claim.
- Round three: write a fictional drafted message of at least 250 words answering a one-line question, and the instruction that asks learners to cut it to what the purpose needs. Record your own shortest defensible version and its word count.
- Write two sentences stating the minimum necessary standard accurately, naming the sections and whom they bind.
- Round four: write the practice's fictional one-page written rule for AI-assisted administrative output, and the sampling act the manager will perform.
- For each round, write the failure mode in one sentence, and the name of the role that owns the output.
- Write one sentence for each round saying what would happen if the wrong role performed the act.
- Write the sentence you will use when the room asks what CMS says about AI-suggested codes.
- Read the whole design and strike anything that states a state scope, supervision or board rule you cannot cite.
Knowledge check
-
Why can a session not teach "check the output" as a single skill?
Answer guidance: Because four different duties sit underneath it, owned by the clinician, the coding professional, the workforce member making a disclosure and the named administrative owner, each checked against a different source with a different failure mode.
-
What is the status of the FSMB 2024 AI policy?
Answer guidance: It is guidance and recommendations adopted by the FSMB House of Delegates in April 2024, binding nobody by itself, with effect only where a state board adopts rules or applies its existing disciplinary authority.
-
What does AHIMA Standard 4.4 say, and why does it matter in a session?
Answer guidance: It says not to query when no clinical information in the record necessitates a query, which contradicts any teaching that presents querying as a technique for improving a reported figure.
-
What must an instructor add when applying AHIMA Standard 9.2 to a large language model tool?
Answer guidance: That the 2016 Standards predate generative AI and refer to computer-assisted coding and encoders, so extending the standard to a language model is an inference and must be labelled as one.
Sources and further reading
- Federation of State Medical Boards — Navigating the Responsible and Ethical Incorporation of Artificial Intelligence into Clinical Practice. Accessed 2026-09-15.
- American Health Information Management Association (AHIMA) — American Health Information Management Association Standards of Ethical Coding [2016 version]. Accessed 2026-09-15.
- U.S. Department of Health and Human Services (Electronic Code of Federal Regulations) — 45 CFR 164.502(b) Standard: Minimum necessary; 45 CFR 164.514(d) Implementation specifications. Accessed 2026-09-15.
- AI Coalition Network — Certification Educational Standard: Human Review of AI-Assisted Professional Work. Accessed 2026-09-15.