The concurrent-review nurse at the payer has six minutes for your file, and your treatment plan is the document she reads first. If it says "client will reduce anxiety and improve coping," she has a denial template open before she reaches paragraph two, and your complex trauma client, the one carrying F43.10 with co-occurring F43.12, just lost authorization for the next twelve sessions you both know she needs. This lesson teaches you to build an AI treatment plan a therapist can actually defend: a DSM-5-TR aligned, medical-necessity-justified plan for complex C-PTSD, anchored to validated PCL-5, PHQ-9, and DES-II scores, written with SMART goals tied to evidence-based interventions, an expected duration, and a review schedule. You will use AI to draft the structure and the language. You will supply every number, every clinical judgment, and every diagnosis yourself. By the end you will have a payer-surviving treatment plan template you can adapt for the hardest chart in your caseload.

Why Treatment Plans Fail Concurrent Review

Start with the scene Maria knows too well. Tuesday, 9:54 PM, Oakland. Her 56-year-old client with complex C-PTSD has been authorized for twenty sessions, and the payer's concurrent review hits at session sixteen. The reviewer is not hostile. She is a utilization-management nurse working a queue, and she is looking for four things: a diagnosis that maps to a covered condition, measurable baseline severity, goals that could in principle be achieved and verified, and evidence that the treatment being delivered is the kind that treats the diagnosis. A plan that says "process trauma, build coping skills, weekly sessions" gives her none of the four. A plan that says "PCL-5 reduced from 58 to below 33 within 24 weeks via Cognitive Processing Therapy delivered weekly under 90837" gives her all of them, and she moves on to the next file.

Complex C-PTSD makes this harder, not easier. DSM-5-TR does not contain a standalone complex PTSD diagnosis; ICD-11 does, but your claim runs on ICD-10-CM codes mapped from your DSM-5-TR formulation. So the chart for a complex trauma presentation typically carries PTSD coded as F43.10, and where there is a co-occurring adjustment-disorder picture from a recent stressor layered on the chronic trauma, F43.12 (adjustment disorder with mixed anxiety and depressed mood is coded differently; F43.12 is adjustment disorder with depressed mood under some mappings, so confirm against your current ICD-10-CM manual and your payer's accepted code list, because the reviewer will). The point a senior supervisor would press here: the code pairing is a clinical decision you make and you defend. AI can remind you what F43.10 denotes. It cannot decide that this client meets criteria, and if you let it guess, you have just delegated diagnosis, which no board in any state permits.

The expensive failure mode is not a rejected plan. It is an approved plan that collapses later. A plan with vague goals gets rubber-stamped at intake, then at concurrent review the nurse asks "what progress has been made against the goals," and there is nothing measurable to point to. The recoupment letter that follows a post-payment audit cites exactly this: services billed against a plan that never established medical necessity in measurable terms. Maria watched a peer absorb $14,200 that way. The treatment plan is not paperwork. It is the load-bearing wall between your clinical work and the payer's right to claw back every 90837 you billed.

The Plan as a Building Permit, Not a Diary

Carry this analogy through the lesson: a treatment plan is a building permit, not a diary. A diary records what you feel like doing. A permit states what will be built, to what specification, by what method, inspected on what schedule, and it is approved by someone who never visits the site but trusts the document. The concurrent reviewer is the inspector who never visits. She cannot see your client's affect soften over eight months. She can only see what the permit said would change and whether your measurements show it changing.

Every component of the payer-surviving plan maps to the permit. The diagnosis section is the zoning declaration: what condition justifies treatment at all. The baseline scores are the site survey: validated instruments, PCL-5 for PTSD symptom severity, PHQ-9 for the depressive load that almost always rides along with complex trauma, DES-II for dissociative symptoms that change your phase-one decisions, taken before work begins, so progress has a denominator. The SMART goals are the build specification: specific, measurable, achievable, relevant, time-bound, each one tied to an instrument or an observable behavior. The interventions section is the construction method: named, evidence-based modalities with literature behind them for this diagnosis. The expected duration and review schedule are the inspection calendar: the dates on which you will re-measure and either justify continuation or document completion.

AI's role in this analogy is the drafting service that turns your engineering decisions into clean permit language. A drafting service is enormously useful: it knows the format, never forgets a section, and writes faster than you do at 9:54 PM. But a drafting service that invents the load calculations gets buildings condemned. The numbers come from you, the instruments, and the client. Always.

The Measurement Foundation: PCL-5, PHQ-9, DES-II

Before AI touches the plan, you need three baseline numbers, because a goal without a baseline is a wish. The PCL-5 is the 20-item PTSD Checklist for DSM-5, self-report, each item scored 0 to 4, total range 0 to 80. A common screening threshold for probable PTSD sits in the low 30s, and a reduction of roughly 10 points or more is widely treated as reliable change; your plan should state the client's actual score and the target. For our worked client, call her R.L., the intake PCL-5 is 58: severe, with the avoidance and negative-cognition clusters loaded heaviest.

The PHQ-9 captures the depressive picture: nine items, 0 to 27, with 5/10/15/20 marking the mild, moderate, moderately severe, and severe bands. R.L. scores 17, moderately severe, with item 9 (thoughts of being better off dead or self-harm) scored 1, which you assess directly in session; AI never interprets an item-9 endorsement, and you document your own risk assessment in your own words. The DES-II, the 28-item Dissociative Experiences Scale, matters specifically for complex trauma because a high dissociation score changes sequencing: trauma-processing work attempted before stabilization in a highly dissociative client can destabilize rather than heal. R.L.'s DES-II mean is 24, elevated but below the range that typically flags a dissociative disorder workup, which supports a phased plan with explicit stabilization work before formal trauma processing.

Notice what just happened: three instruments produced three numbers, and the numbers made three clinical decisions visible: severity justifying weekly 90837, a depressive comorbidity worth its own goal, and a sequencing decision about when exposure-based work begins. That is what the reviewer means by medical necessity, and it is also simply good care. The AI prompt you write later will include these scores verbatim, because AI cannot supply them, and an AI that is not given them will fabricate plausible ones. A fabricated baseline in a signed treatment plan is a falsified record. There is no soft way to say that.

A treatment plan survives payer review for the same reason it serves the client: every goal has a number, every number has an instrument, and every intervention has evidence. AI can draft the sentences; only you can supply the truth they carry.

Writing SMART Goals That Survive a Reviewer

Here is the difference between a goal that dies in review and one that survives, using R.L.'s chart. The dying version: "Client will process trauma and develop healthier coping mechanisms." Nothing is measurable, nothing is time-bound, nothing could ever be marked complete. The surviving version, the worked artifact of this section:

Goal 1 (PTSD symptom reduction): "Reduce PTSD symptom severity as measured by PCL-5 from baseline 58 to below 33 (below probable-PTSD threshold) within 24 weeks, with interim target of 48 or lower at week 8, via weekly Cognitive Processing Therapy." Objective 1a: "Client will complete CPT impact statement by session 3 and identify three stuck points by session 5." Objective 1b: "Client will report using cognitive worksheets between sessions at least twice weekly by week 6, per session self-report." Goal 2 (depression): "Reduce PHQ-9 from 17 to below 10 within 16 weeks, re-administered every fourth session." Goal 3 (stabilization/dissociation): "Client will demonstrate two grounding skills (5-4-3-2-1 sensory anchoring, paced breathing) in session by week 4 and report independent use during one dissociative episode by week 8; DES-II re-administered at week 12 with target mean below 20."

Walk through why each survives. Specific: the instrument and skill are named. Measurable: PCL-5 58 to below 33 is arithmetic, not adjectives. Achievable: a 25-point PCL-5 drop over 24 weeks of CPT is consistent with published outcomes for the modality, not a fantasy of cure in six sessions. Relevant: every goal maps directly to the F43.10 diagnosis or its documented comorbidity. Time-bound: weeks 4, 8, 12, 16, 24 give the reviewer the inspection calendar. The interim targets are the underrated move: at concurrent review you will not have hit the final number, and the interim target is what lets you say "on trajectory" with evidence instead of adjectives.

Naming the Interventions: CPT, PE, EMDR, IFS

The interventions section is where many AI drafts go quietly wrong, because a generic model will list every trauma modality it has ever read about, which reads to a reviewer like a clinician who has not chosen one. Your plan names the primary modality you are actually trained in and actually delivering, states its evidence base for PTSD, and gives the dose. For R.L.: "Primary intervention: Cognitive Processing Therapy (CPT), a structured, evidence-based protocol for PTSD, delivered in weekly 53-minute individual sessions (CPT protocol typically 12 sessions; extended course planned given complexity and comorbid depressive symptoms), billed 90837. Adjunct: skills-based stabilization drawn from phase-oriented treatment of complex trauma during weeks 1 to 6 prior to formal trauma processing, indicated by elevated DES-II."

If your training and the client's presentation pointed elsewhere, the same structure holds for prolonged exposure (PE), EMDR, or IFS. PE and EMDR carry strong evidence bases for PTSD and are recognizable to reviewers; IFS is increasingly used for complex trauma and may need one extra sentence of clinical rationale in plans submitted to conservative utilization-management shops. Two supervisory cautions. First, list only modalities within your scope and training; a plan that promises EMDR from a clinician with no EMDR training is a board problem waiting for a transcript. Second, name the modality you will actually use in session, because the payer-content rule that governs all your documentation applies here too: the verifiable detail AI cannot supply, the modality actually delivered, the minutes actually spent, the score actually obtained, is exactly the detail that makes the record defensible. When your progress notes later say "CPT session 7, stuck point work on self-blame," they corroborate the plan, and the chart hangs together under audit.

Close the section with duration and review: "Expected duration: 24 weeks of weekly individual psychotherapy, with formal treatment plan review at weeks 8, 16, and 24 or sooner if clinically indicated. Continuation beyond 24 weeks contingent on documented medical necessity, including instrument scores and functional status." That sentence tells the reviewer you are managing the episode of care, not drifting through it.

The AI Drafting Workflow, Step by Step

Now, and only now, AI enters. The workflow has five steps, and the order is the safety mechanism. Step one: you decide. Diagnosis (F43.10, with the co-occurring F43.12 picture documented and code-checked), baselines (PCL-5 58, PHQ-9 17, DES-II 24), modality (CPT with phase-one stabilization), duration (24 weeks), review cadence (8/16/24). These decisions exist before the prompt, on paper or in your head, because anything you leave blank, the model will fill, and what it fills is fiction.

Step two: you prompt inside a HIPAA-appropriate tool, meaning a platform under a BAA (your EHR's integrated drafting feature, or a scribe platform such as Mentalyc, Upheal, Eleos, or Heidi operating under your executed BAA), never a free consumer chatbot tier, and with identifiers minimized regardless. The prompt supplies your decisions and asks only for structure and language: "Draft a treatment plan in [my EHR's] format for an adult client, diagnosis PTSD (F43.10) with co-occurring adjustment-related presentation (F43.12, code to be confirmed), baseline PCL-5 58, PHQ-9 17 (item 9 = 1, risk assessed separately by clinician), DES-II mean 24. Primary modality CPT, weekly 90837, 24-week expected duration, phase-one stabilization weeks 1 to 6. Write 3 SMART goals with the interim and final score targets I provide: [your targets]. Do not add diagnoses, scores, modalities, or risk language I have not provided."

Step three: the model drafts. Step four, the verification pass, is where your license lives: check every number against your source documents, confirm no invented symptoms or history crept in, confirm the goals say what you decided and not a smoothed paraphrase, strip any hedge language that weakens medical necessity ("client may benefit from" becomes "client meets criteria for and requires"), and confirm the risk content is yours alone. Step five: you sign. The signature is a legal attestation that every statement in the plan is your clinical judgment, not a formatting step at the end of a software workflow. Read every word before signing. Every time.

Reading Your Own Plan Like the Reviewer Will

Before the plan goes in the chart, run the reviewer's six-minute read against it. Question one: does the diagnosis justify the level of care? F43.10 with PCL-5 58 and PHQ-9 17 justifies weekly outpatient psychotherapy without argument; the same diagnosis with a PCL-5 of 21 would not, and pretending otherwise invites the audit. Question two: can progress be verified from the chart alone? If every goal has an instrument and a date, yes. Question three: does the intervention treat the diagnosis? CPT for PTSD, yes; "supportive therapy" for PTSD at 90837 frequency, the reviewer has seen that denial a hundred times. Question four: is there an end in sight, or a documented reason there is not? The 24-week duration with reviews answers it; for legitimately longer complex-trauma courses, your week-24 review documents the medical necessity for the next authorization period in fresh numbers.

Then run the second pass nobody teaches: the deposition read. If this chart is ever in front of a board or a plaintiff's attorney, the treatment plan is exhibit one. Does it match what your notes say you did? Does it reflect a risk assessment in your voice? Is every score in it traceable to an administered instrument with a date? An AI-drafted plan that sailed through payer review can still sink you in a board matter if it asserts things you never assessed. The two audiences, payer and board, are satisfied by the same discipline: nothing in the plan that did not come from you, the client, or a scored instrument.

Finally, calendar the reviews like clinical appointments, because a plan with a week-8 review that never happened is worse than no review schedule at all; it documents your own non-compliance with your own plan. The re-administration dates for PCL-5 and PHQ-9 go in your scheduling system the day you sign.

The Applied Problem: The Payer-Surviving C-PTSD Treatment Plan

Your deliverable is the Payer-Surviving Treatment Plan for R.L., built start to finish. First, assemble the clinical inputs on one page before opening any AI tool: diagnosis F43.10 with the co-occurring F43.12 presentation noted and code-verified against your current ICD-10-CM manual and the payer's code list; baselines PCL-5 58, PHQ-9 17 with item 9 = 1 and your separately documented risk assessment, DES-II mean 24; modality CPT with six weeks of stabilization first; weekly 90837; 24-week duration; reviews at weeks 8, 16, 24.

Second, in your BAA-covered tool, run the drafting prompt from this lesson, pasting your inputs and your three SMART goals with their interim targets (PCL-5 to 48 by week 8, below 33 by week 24; PHQ-9 below 10 by week 16; two grounding skills demonstrated by week 4, DES-II below 20 at week 12). Include the closing instruction verbatim: "Do not add diagnoses, scores, modalities, or risk language I have not provided."

Third, the verification pass with a printed copy and a pen: circle every number and tick it against your source sheet; strike any symptom, history detail, or intervention you did not supply; rewrite any goal the model softened; confirm the risk paragraph is word-for-word yours; confirm duration and review dates survived intact. Fourth, run the reviewer's four questions and the deposition read from the previous section. "Done" looks like this: a one-to-two page plan in which every assertion traces to an instrument, a session observation, or your documented judgment; three SMART goals with baselines, interim targets, instruments, and dates; one named evidence-based modality with dose and rationale; an expected duration with a review calendar already entered in your scheduler; and your signature applied after a full read, not before. Save the structure, not the scores, as your reusable template; the scores are new for every client, every time.

Key Takeaways

• A treatment plan is a building permit, not a diary: the concurrent reviewer never meets your client and can only approve what the document makes verifiable. Vague goals do not just risk denial at review; they set up the post-payment recoupment that arrives after years of billing against a plan that never established measurable medical necessity.
• Complex C-PTSD is documented in DSM-5-TR terms and billed on ICD-10-CM codes, typically F43.10 with a co-occurring F43.12 presentation where clinically supported. The code pairing is your clinical decision to make and defend; AI may explain codes but never assigns a diagnosis, and you verify codes against the current manual and the payer's list.
• Three validated instruments anchor the plan: PCL-5 for PTSD severity (R.L.: 58), PHQ-9 for depressive comorbidity (17, with item 9 assessed directly by you), and DES-II for dissociation (mean 24), which drives the sequencing decision to stabilize before trauma processing. AI cannot supply these numbers, and an ungiven model will fabricate plausible ones; a fabricated baseline in a signed plan is a falsified record.
• SMART goals survive review when each one names an instrument, a baseline, an interim target, a final target, and a date: PCL-5 from 58 to 48 by week 8 and below 33 by week 24 via CPT. Interim targets are what let you show "on trajectory" at concurrent review with arithmetic instead of adjectives.
• Name one primary evidence-based modality you are trained in and actually delivering (CPT, PE, EMDR, or IFS with rationale), with dose, billing code, and duration, rather than letting AI list every trauma therapy it has read about. Your progress notes must later corroborate the plan, because a chart that hangs together is the audit defense.
• The five-step workflow is the safety mechanism in order: you decide, you prompt inside a BAA-covered tool with the instruction "do not add anything I have not provided," the model drafts, you verify every number against source documents, and you sign. The signature is a legal attestation, not a formatting step; read every word first.
• Read the finished plan twice before filing: once as the reviewer (diagnosis justifies level of care, progress verifiable, intervention treats the diagnosis, end in sight) and once as a deposition exhibit (everything traceable, risk assessment in your own voice). Then calendar the week-8, week-16, and week-24 reviews the same day you sign.