A clinician at a community clinic sees the same kind of client twice in one afternoon: a 67-year-old Medicare beneficiary at 1:00 and a 34-year-old Medicaid member at 2:00. Same diagnosis category, same modality, same 55 minutes of psychotherapy. She writes the same style of note for both, because nobody ever told her she had crossed a border between sessions. Eighteen months later, two audits arrive in the same quarter: a Medicare reviewer flags her 90837 notes for missing time-in-session and medical-necessity language, and the state Medicaid program flags the other chart for missing service start and stop times and a signature line that does not show her credential. Same clinician, same skill, two different failures, because Medicare and Medicaid are not one government program with two names. They are two audit universes with different physics. This lesson teaches you to write for both: what each universe inspects, why an AI scribe trained on one will quietly fail the other, and how to build the parallel note pair that survives each reviewer. By the end you will produce that pair for a single session, line by line.

Two Border Crossings, One Suitcase

Hold one controlling analogy through this lesson: your progress note is a suitcase, and Medicare and Medicaid are two border crossings with different inspection manuals. The suitcase contains the same clinical reality either way, the session that actually happened. But the Medicare checkpoint opens it looking for specific items: medical-necessity language tying the service to the diagnosis and to expected improvement, the time in session that justifies the code, and a signature dated within the regulatory window. The Medicaid checkpoint, in most states, searches for different contraband entirely: missing service start and stop times, missing location of service, an unnamed modality, and a signature line that does not carry the rendering clinician's credential as the state's provider rules require. A note that sails through one checkpoint can be confiscated at the other, and the painful part is that both inspections happen retrospectively, years after the suitcase was packed, when your only defense is what the note already says.

Why are the manuals different? Because the programs are built differently. Medicare is a federal program with national coverage architecture, administered through contractors, and its documentation expectations flow from a single question: was this service medically necessary and actually furnished as billed? Hence its obsession with necessity language and time. Medicaid is a federal-state partnership in which each state writes its own provider manual, sets its own documentation elements, and runs its own program integrity unit, and its historical battle has been a different one: phantom billing and credential mismatches, services billed that were never rendered, rendered by someone unqualified, or rendered somewhere other than claimed. Hence its obsession with start and stop times, place of service, and credentialed signatures. Neither universe is irrational. Each is auditing against the fraud pattern it has actually seen, and your note must answer both histories at once when your caseload spans both programs.

One framing correction before the details: "Medicaid documentation rules" is not a single thing you can learn once. Medicaid varies state by state, and the elements this lesson teaches, start and stop time, location, modality, credentialing-aligned signatures, are the typical core that recurs across state manuals, not a substitute for reading your own state's. Treat this lesson as the map of the border region; your state's provider manual is the actual inspection checklist, and the applied problem at the end has you pull it.

The Medicare Universe: Necessity, Time, and the Signature Clock

Walk the Medicare checkpoint first, because its three inspection items are the cleanest to learn. Item one: medical-necessity language. A Medicare reviewer reads the note asking whether it demonstrates that psychotherapy at this intensity was reasonable and necessary for this diagnosis: symptoms active, functional impairment present, treatment targeted at improvement or stabilization, response being monitored. "Supportive listening provided, client doing well" fails that reading even when the session was excellent, because a note that describes wellness without active treatment of impairment reads, to this universe, as a service that was not necessary. The necessity sentence, taught earlier in this chapter, carries the weight: the diagnosis, the active symptoms with their functional cost, the intervention applied, and the clinical reason care continues at this frequency.

Item two: time in session. For 90837, the code itself is a time claim: 53 or more minutes of psychotherapy. Medicare expects the note to substantiate the time, and the defensible standard is time in session, the psychotherapy minutes actually delivered, not appointment time, not the calendar block. This single distinction has decided real audits: a 60-minute calendar slot with ten minutes of scheduling chatter and paperwork is not 53 minutes of psychotherapy, and a reviewer who finds "appointment: 3:00-4:00" where minutes should be treats the time element as undocumented. Write the minutes: "55 minutes of psychotherapy provided in session." Commercial payers run parallel scrutiny here, and Optum's 90837 high-utilization audit program has been among the most aggressive in commercial behavioral health since 2022, but in the Medicare universe the time element is not a utilization flag; it is a condition of the code.

Item three: the signature and its clock. A Medicare-acceptable note is signed and dated by the rendering clinician within the regulatory window, because an unsigned note is, for audit purposes, an unauthenticated document, and a note signed long after the service invites the question of when it was actually written. This is where the program's cardinal rule does double duty: the clinician signs the note, and the signature is a legal attestation, not a formatting step. In the Medicare universe the attestation has a timestamp, and a backlog of unsigned AI drafts sitting in a queue is not a workflow inconvenience; it is an accumulating authentication problem. If your scribe drafts twenty notes you sign on Sunday night, the record shows exactly that, and a pattern of late signatures is itself an audit finding waiting to be made.

The Medicaid Universe: Start, Stop, Where, What, and Who

Now the second checkpoint. The typical state Medicaid manual wants the note to answer five blunt questions, and the bluntness is the point. When did the service start and when did it stop? Not a duration, two clock times: "Service start 2:04 PM, end 2:59 PM." The start-stop pair exists because Medicaid program integrity units detect phantom and overlapping billing arithmetically: two services billed by one clinician at overlapping times, or more service hours in a day than a day holds, are caught by comparing clock times across claims, and a note without them cannot be cleared. Where was the service rendered? Office, school, home, community setting, or telehealth, matching the place-of-service on the claim, because location mismatches are a classic integrity flag, and HCBS and school-based work make location an eligibility question, not a detail. What service was delivered? The modality, named: CBT, behavioral activation, family therapy with the client present. A Medicaid reviewer reading "met with client" cannot distinguish psychotherapy from a friendly visit, and the state is paying for the former.

And who rendered it, with what credential? The signature line in this universe carries the rendering clinician's name, credential, and where required, provider identifiers, and it must align with the credential the service was billed under. This is where supervised settings concentrate risk: an associate-level clinician rendering services billed under a supervisor's credential is a lawful arrangement only where the state's rules permit it and the documentation reflects the actual arrangement, supervisory co-signatures included where required. A note signed "C. Reyes" over a claim billed under an LMFT supervisor, in a state whose manual requires the rendering provider's credential and the supervisor's countersignature, fails the checkpoint twice. Carmen's situation from earlier lessons returns here with money attached: her hours, her supervisor's countersignature obligations, and the agency's Medicaid billing all meet on the signature line.

Two environments add a third layer on top of state rules. CCBHCs, certified community behavioral health clinics, and HCBS programs carry daily progress documentation expectations and prospective payment system (PPS) reporting requirements, which means the note is not only a clinical record but an input to a payment-rate reporting structure; missing or late documentation in those settings ripples into the clinic's cost reporting, not just one claim. If you work in a CCBHC, the documentation manual is longer than this lesson, and the discipline it demands, every encounter documented, every element present, every day, is exactly the discipline the parallel note pair builds.

Medicare asks whether the service was necessary and how long it lasted; Medicaid asks when it started, when it stopped, where, what, and by whom. Write the note that answers all seven questions and both auditors go home empty-handed.

Where an AI Scribe Quietly Fails Each Universe

Here is why this is an AI lesson and not just a billing lesson. AI scribes learn note-writing from their training distribution and their templates, and most are tuned to the commercial-outpatient middle of the market. Put one to work across a Medicare-Medicaid caseload and it fails each universe in a characteristic, quiet way. In the Medicare universe, the failure is fluent thinness: the scribe produces a beautiful narrative of the session that never states the necessity logic, because necessity is an argument, not a transcript feature; no client says "my symptoms cause functional impairment warranting continued weekly psychotherapy" out loud, so a transcription-based tool has nothing to transcribe. And it cannot know the minutes. A scribe that listened to audio knows roughly how long the recording ran, which is not the same as psychotherapy time in session, and a scribe working from your typed shorthand knows nothing at all. The minutes are the worked example of this chapter's standing rule: the verifiable detail AI cannot supply. You looked at the clock; the model did not.

In the Medicaid universe, the failure is structural omission: the template simply has no field for service start and stop times, place of service, or a credentialed signature block, because the commercial-outpatient notes the tool was built on do not require them. The draft reads complete, which is the trap; completeness to the eye is not completeness to the checkpoint. Start and stop times are clock facts only you hold. The place of service is a fact about the encounter the model may guess wrong, especially in hybrid telehealth weeks. The modality must be named as delivered, not as the model's best label for the conversation; a reviewer comparing the note's "processed emotions" against a treatment plan specifying CBT sees drift, and drift reads as either an undocumented plan change or an inaccurate note.

The fix is not abandoning the scribe; it is template engineering plus a verification habit. Build two note templates, one per universe, each with the checkpoint items as mandatory fields the draft cannot be signed without: Medicare template with a necessity-sentence slot, a minutes line, and the signature date; Medicaid template with start-stop times, place of service, named modality, and the credentialed signature block, plus the supervisory countersignature line where your state requires it. Then instruct the scribe, in its standing prompt, to leave those fields as explicit blanks, [MINUTES], [START-STOP], [LOCATION], [MODALITY AS DELIVERED], rather than inventing values. A blank you must fill is a safety feature. A plausible invented value is a false claim with your signature on it.

The Parallel Note, Side by Side

Now run the worked example slowly: one clinical hour, written for each universe. The session: a client with recurrent major depressive disorder, moderate, seen for 55 minutes of individual psychotherapy, CBT with behavioral activation, PHQ-9 administered this session scoring 13, down from 18 six weeks ago. The Medicare-acceptable note leads with necessity and time: diagnosis and active symptoms ("depressed mood, anhedonia, and impaired occupational functioning persist; client missed two workdays this month"), the intervention as delivered ("CBT with behavioral activation; scheduled three specific activities with accountability plan"), the measurement anchor ("PHQ-9 today 13, from 18 on [date], improving but above remission threshold"), the necessity close ("continued weekly 90837 is medically necessary to consolidate gains and address residual occupational impairment; risk of relapse with reduced frequency given prior episode history"), the time line ("55 minutes of psychotherapy provided in session"), and the signature with today's date, inside the window.

The Medicaid-acceptable note for the same hour reorders the suitcase: "Service start 2:04 PM, end 2:59 PM. Place of service: office [or telehealth, matching the claim]. Service: individual psychotherapy, CBT with behavioral activation." Then the clinical content, symptoms, interventions, response, the same PHQ-9 anchor, and the plan. Then the signature block: full name, credential as the state recognizes it, identifiers as required, and the supervisor's countersignature line if the rendering clinician is associate-level and the state requires it. Read the two notes side by side and notice what the exercise teaches: roughly seventy percent of the content is shared clinical substance, and the remaining thirty percent is checkpoint-specific packaging. The skill is not writing two notes from scratch; it is knowing which packaging each border requires and never shipping the suitcase with the wrong one.

Notice also which elements appear in both notes because they are simply true: the PHQ-9 delta (18 to 13, with dates) and the minutes. Those two facts are this chapter's recurring payer rule made concrete: the verifiable details AI cannot supply are exactly the details both audit universes weigh most heavily, which is why the clinician, not the scribe, owns them in every workflow this program teaches.

When the Audit Comes: How Each Universe Reviews

Understand the review mechanics, because they explain the documentation rules better than any manual. Medicare review arrives through contractors conducting prepayment or postpayment review, frequently triggered by data analytics: a clinician whose 90837 percentage towers over peers, or whose claims pattern deviates from specialty norms, gets selected, and a records request follows for a sample of dates. The reviewer scores each note against the elements: necessity demonstrated, time documented, signature authenticated and timely. Notes that fail are overpaid claims, and overpayment findings extrapolate: a sample failure rate projected across the review period is how a stack of thin notes becomes a five-figure recoupment, the $14,200 letter Maria watched a peer receive. The defense is never argument after the fact; it is the note that already contains the elements, written the week the service happened.

Medicaid program integrity runs on different fuel: claims-data cross-checks (overlapping times, impossible daily hours, place-of-service mismatches), credential verification against provider enrollment files, and on-site or desk audits of full charts in which every element of the state manual is a line item. The findings read differently too: where Medicare findings speak the language of medical necessity, Medicaid findings speak the language of failure to document required elements, and a chart that is clinically rich but missing start-stop times can fail wholesale on elements alone. Federal-state enforcement also gives Medicaid findings a longer tail, because documentation failures in integrity reviews can escalate beyond repayment into program exclusion questions for patterns that look like misrepresentation. The practical translation for a busy clinician: in the Medicare universe, write the argument; in the Medicaid universe, complete the form; in both, do it contemporaneously, because the signature date is itself evidence.

And in both universes, the AI-era version of an old rule applies: the audit reviews the note you signed, not the draft the model produced. No reviewer in either universe accepts "the scribe template did not have that field" as a defense, any more than they ever accepted "the EHR template did not prompt me." The tool shapes the draft; the signature owns the note.

Building the Two-Universe Workflow in a Mixed Caseload

For the clinician or group practice serving both populations, the durable solution is workflow, not memory. Step one: flag the universe at scheduling, so every encounter carries its program label into documentation, because the worst failures happen when a clinician does not consciously register which checkpoint a given note must clear. Step two: bind the template to the flag. Medicare encounters open the necessity-and-time template; Medicaid encounters open the elements template with start-stop, location, modality, and signature block; CCBHC and HCBS settings layer their daily-documentation and PPS-reporting requirements on top. Step three: configure the scribe per template, with the standing instruction to leave the verifiable details as named blanks. Step four: a signing checklist of five seconds per universe. Medicare: necessity sentence present, minutes stated, date today. Medicaid: start-stop present, location matches the claim, modality named as delivered, signature block carries the right credential, countersignature if required. Step five: a monthly self-audit of three charts per universe against the checklist, logged, because the clinic that finds its own element gaps fixes a template; the clinic that waits for the integrity unit funds a recoupment.

Supervisors carry one extra duty in this workflow: the countersignature is not a rubber stamp. In the Medicaid universe a supervisor countersigning an associate's note is attesting within the supervisory framework the state's rules define, and a countersigned note missing start-stop times fails with two names on it. Build the associate's template to the state manual first, then teach the associate why each field exists, using this lesson's two-checkpoint frame, because a clinician who understands what each universe is hunting for completes the fields without being chased.

The Applied Problem: Write the Parallel Medicare-Medicaid Note Pair

Your artifact is the Parallel Note Pair: one session, two audit-ready notes, plus the two templates they generate. Step one: take the worked session from this lesson, 55 minutes, recurrent major depressive disorder moderate, CBT with behavioral activation, PHQ-9 13 from 18 with dates, or substitute a fully de-identified session of your own. Write the fact line first, the facts only you hold: minutes in session, start and stop clock times, place of service, modality as delivered, instrument scores with dates, signature credential, and whether a countersignature applies.

Step two: draft both notes, with AI assistance if you choose, using a prompt like: "Draft two versions of a progress note for the same psychotherapy session, using only the facts I provide and leaving [BRACKETED] blanks for any fact not provided. Version A, Medicare-acceptable: include a medical-necessity statement linking diagnosis, active symptoms, functional impairment, intervention, and rationale for continued weekly 90837; state psychotherapy minutes in session; end with a dated signature line. Version B, Medicaid-acceptable: open with service start and stop times, place of service, and the named modality; include the clinical content and measurement; end with a credentialed signature block and a supervisory countersignature line. Facts: [paste your fact line]." Step three: the verification pass, against this lesson's two checklists: Version A answers necessary-and-how-long; Version B answers when-started, when-stopped, where, what, who; both carry the PHQ-9 delta with dates; no bracketed blank remains; nothing in either note is a value the model supplied for a fact you did not give it.

Step four: strip the case specifics and save the skeletons as your two standing templates, then pull your actual state's Medicaid provider manual documentation section and reconcile Version B's fields against it, adding anything your state requires that the typical core misses. "Done" is four documents: two completed notes, two reconciled templates, and the habit they encode. The next time you cross the border twice in one afternoon, the suitcase will be packed for the right checkpoint both times.

Key Takeaways

• Medicare and Medicaid are two audit universes with different physics: Medicare inspects for medical-necessity language, documented time in session, and a signature dated within the regulatory window; typical state Medicaid manuals inspect for service start and stop times, place of service, named modality, and credentialing-aligned signature lines. One note style cannot serve both unexamined.
• The universes differ because their fraud histories differ: Medicare audits the necessity and duration of services billed, while Medicaid program integrity hunts phantom billing, overlapping clock times, location mismatches, and credential misalignment, which is why it wants clock times, not durations.
• For 90837 in any universe, time in session is the standard: 53 or more minutes of psychotherapy actually delivered, written as minutes, never "appointment time." Optum's commercial 90837 high-utilization audits enforce the same discipline from the commercial side.
• AI scribes fail each universe characteristically: fluent thinness in Medicare notes, because necessity is an argument no transcript contains, and structural omission in Medicaid notes, because commercial templates lack start-stop, location, and credential fields. The fix is two templates with mandatory fields and a standing instruction to leave verifiable details as bracketed blanks.
• The verifiable details AI cannot supply, the minutes, the start-stop clock times, the PHQ-9 delta with dates, the modality as delivered, are exactly the elements both universes weigh most heavily. The clinician owns them in every workflow, and a plausible invented value is a false claim with your signature on it.
• CCBHC and HCBS settings add daily progress documentation and PPS reporting on top of state Medicaid rules, and supervised settings put two names on every countersigned note: the countersignature attests the note, so an element failure fails the supervisor too.
• Your artifact is the Parallel Note Pair: one session written Medicare-acceptable and Medicaid-acceptable, verified against the seven-question checklist (necessary, how long, when started, when stopped, where, what, who), then stripped into two standing templates reconciled against your own state's Medicaid provider manual.