The envelope arrives at Maria's office on a Wednesday: a subpoena duces tecum from the attorney for her adult client's ex-husband, demanding "any and all records, notes, communications, and recordings" relating to her client, in a custody matter Maria has never been part of. Three feelings hit at once: the urge to protect her client, the fear of contempt if she ignores it, and the temptation to just send the chart and make the problem go away. All three, acted on raw, end badly. The subpoenaed treating therapist is not a forensic evaluator, not the court's expert, and not a free document warehouse for opposing counsel; she is a privilege-holder's clinician with specific legal duties and a narrow, learnable workflow. This lesson runs that workflow end to end for dependency court, family court, and custody matters: the subpoena-response cover letter that asserts privilege and minimum necessary, the attorney records-review package, the deposition-prep chart summary, and the over-disclosure audit before anything leaves the office. It also holds the lines that keep treating clinicians safe on the stand: the treating clinician does not opine on parenting capacity, and AI never drafts testimony. By the end you will build a Subpoena-Response Cover Letter asserting privilege and minimum necessary, ready for your own letterhead and your attorney's review.

A Subpoena Is a Question, Not Permission to Disclose

Start with the misunderstanding that generates most board complaints in this area: a subpoena is not a court order, and receiving one neither requires nor permits you to hand over a privileged chart. A subpoena is an attorney's demand, issued routinely and often overbroadly; the psychotherapist-patient privilege belongs to your client, and until the privilege is waived by the client or pierced by a judge's ruling, your job is to assert it. The statutes are state-specific and you cite yours, never a generalized "therapy is confidential": California Evidence Code §1014 establishes the patient's privilege to refuse disclosure of confidential communications between patient and psychotherapist; Texas Health and Safety Code §611 governs the confidentiality of mental health records and the narrow routes around it; New York CPLR 4504 codifies the physician, dentist, podiatrist, chiropractor, and nurse privilege structure under which protected clinical communications are shielded. Each statute has its own exceptions, its own waiver doctrine, and its own litigation-context carve-outs (a client who puts their mental state at issue may waive protection in ways that surprise everyone), which is exactly why step one of the workflow is never "respond" but "call."

Two calls, in order. First, your own attorney or your malpractice carrier's risk line, because the response deadline, the proper objection vehicle, and the local procedure are legal questions, and CPH-style carriers handle these calls daily. Second, with your attorney's guidance, notification to the client and the client's attorney, because the privilege is the client's to assert or waive, and the client's counsel may move to quash before you write a word. What you never do: ignore the subpoena (contempt exposure), or comply reflexively (privilege breach, board complaint, and a client who learns their trauma history was mailed to their ex-husband's lawyer because their therapist panicked).

Carry this lesson's analogy: your chart is a sealed diplomatic pouch, and you are the courier. A subpoena is someone at the border demanding the pouch. The courier does not open the pouch because the demand is loud; the courier produces credentials, states that the pouch is protected, and waits for the only authority that can lawfully open it: the sender's waiver or a judge's order, and even then, only the specific documents the order names. AI, in this picture, is the courier's typewriter: excellent for drafting the protest letter, the inventory, and the travel log; never the one who decides whether the pouch opens.

You Are the Treating Clinician: What That Role Cannot Do

The previous lesson built the Greenberg and Shuman distinction from the forensic evaluator's side; this lesson lives on the treating side, where the role limits are your shield on the stand. The treating therapist's competence is the treatment: what the client reported, what was observed in session, the diagnosis, the interventions, the course of care. The treating therapist's incompetence, in the formal sense, is the psycho-legal question: who should have custody, whether your client is a fit parent, what parenting plan serves the child's best interests. You never gathered the data those opinions require, no collateral verification, no evaluation of the other parent, no home visits, no forensic instruments, and the APA Guidelines for Child Custody Evaluations and the AFCC Model Standards both locate those opinions exclusively with the appointed or retained custody evaluator. The treating clinician does not opine on parenting capacity. Full stop, and saying so on the stand is not evasion; it is the professionally correct answer: "That question calls for a custody evaluation, which I have not performed and which would be outside my role as the treating therapist."

This role purity has a documentation corollary that starts long before any subpoena: your notes must not adjudicate. A treatment note that says "client is clearly the more stable parent" or "client's ex appears emotionally abusive" has, in one sentence, performed an evaluation you never conducted, about a person who was never your client, and that sentence will be projected on a courtroom screen. The discipline from the couples and dyadic lessons applies with full force: attributed report ("client reported that her ex-husband raised his voice during the exchange"), behavioral observation of your actual client, and zero characterization of the non-client parent. AI's legitimate role here is the audit: scanning existing notes and any document headed out the door for adjudicating language, fitness language, or non-client characterizations. AI's prohibited role is equally clear: AI does not summarize sessions in language that adjudicates fitness, and AI never sees the chart of the non-client parent, who may be another clinician's client with their own privilege and who never consented to your vendor's servers.

Dependency court adds a wrinkle: in child-welfare matters, statutes and court orders can compel disclosure more directly, mandated-report obligations may already have placed information into the system, and minors' privilege operates differently (often held or waived through a guardian ad litem or the court). The workflow is the same, counsel first, scope always, but the legal landscape under it shifts, which is one more reason the cover letter you build today routes everything through your attorney rather than your best guess.

The privilege is the client's, the scope is the judge's, and the only thing that is entirely yours is the discipline not to send one page more than the order compels.

Artifact One: The Subpoena-Response Cover Letter

The cover letter is the workhorse document, and it does three jobs in under a page: acknowledges receipt, asserts the privilege, and states your conditions for production. Its skeleton: paragraph one confirms receipt of the subpoena, identifying it by date, case caption, and requesting party. Paragraph two asserts that the records demanded are confidential communications protected by the psychotherapist-patient privilege under your state's statute, cited by section (CA Evidence Code §1014, TX Health and Safety Code §611, or NY CPLR 4504, as applicable), that the privilege belongs to the patient and has not, to your knowledge, been waived, and that you are therefore not at liberty to produce records absent the patient's written authorization or a court order compelling production. Paragraph three states the minimum-necessary position: that should production be authorized or ordered, you will produce only the records specified, within the date range specified, and that you request the opportunity to seek appropriate protections (in-camera review, a protective order limiting use and redisclosure) for material whose release would be clinically harmful. A closing line directs further communication to your counsel.

AI drafts this letter well, because it is structured legal-administrative correspondence with no clinical content: the prompt supplies the case caption, the statute citation you (or your attorney) selected, and the posture, and the model produces clean letterhead prose. Two guardrails. First, the letter contains zero clinical information, not the diagnosis, not the treatment dates beyond what the response procedurally requires, not even an implicit confirmation of the treatment relationship where your attorney advises that confirming it is itself a disclosure; the prompt must say so explicitly. Second, the letter is reviewed by your attorney before it is sent, every time, because objection procedure is jurisdiction-specific and a well-meaning letter that misses the local objection deadline protects nothing. The AI-drafted letter is a strong first draft for counsel, not a substitute for counsel.

Artifact Two: The Attorney Records-Review Package

Assume the next stage arrives: your client, advised by their attorney, signs an authorization, or a court orders production of defined records. Now the work is assembly under scope, and this is where AI earns its keep as inventory clerk. The attorney records-review package is the organized set you produce for counsel's review before anything goes to anyone else: a cover index listing every document (note, treatment plan, intake, correspondence, assessment) with date, type, and page count; the documents themselves within the ordered date range and category; and a flag list, items inside the scope that raise specific concerns: third-party information (the client's children, the non-client parent appearing in attributed report), psychotherapy notes if you keep them under the separate 45 CFR 164.501 carve-out (which are not "the chart" and typically demand separate, explicit treatment), mandated-report records with their own disclosure rules, and any 42 CFR Part 2 material if SUD program records ever entered the file, which, as the earlier lesson taught, do not move without their own compliant order.

The AI workflow: the clinician pulls the in-scope documents; AI builds the index from the document list, checks every item against the order's date range and categories, and produces the flag list against a checklist prompt: "For each document in this index, flag: any reference to persons other than the client; any psychotherapy-note material; any mandated-report content; any substance use disorder treatment content; any document outside [date range] or outside the categories [list]. Output a table: document, date, flag type, page reference. Do not summarize clinical content." The clinician then walks the flags with counsel: redact, withhold under separate objection, or produce. The decisions are legal-clinical judgments made by humans; the model's contribution is that nothing gets missed at 11 PM the night before the production deadline. The same discipline governs what AI sees: the package prompt operates on document titles, dates, and the clinician's flags wherever possible, and full-text processing happens only in a tool that satisfies the practice's BAA and retention requirements, because a chart being produced under privilege protest is the last data that belongs in a consumer chatbot.

Artifact Three: The Deposition-Prep Chart Summary, and the Testimony Line

If a deposition or hearing follows, you prepare, and preparation has a bright line through it. On the permitted side: AI can build you a chart-navigation summary, a chronological index of the record you will testify from: session dates, CPT codes billed, treatment plan dates and revisions, assessment administrations with dates, gaps in care, and where in the chart each lives. This is wayfinding, not narrative: when opposing counsel asks "when did you first document sleep disturbance," you find the page in seconds instead of flipping under oath. The prompt mirrors the records package: "From this document index and these clinician-provided entries, produce a chronological chart-navigation table: date, event type, document, page. Do not characterize, interpret, or summarize clinical content beyond the entry I provided."

On the prohibited side, state it the way you would to a supervisee: AI does not draft testimony. Not answers, not talking points, not "suggested phrasings for difficult questions," not a rehearsal script. The reasons are layered. Your testimony must be your recollection and your records, in your voice; a witness reciting model-polished language has a memorized quality that experienced litigators detect and dismantle ("Doctor, who wrote that sentence for you?"). Preparation materials can be discoverable in many circumstances, and an AI-generated answer sheet in your file is a cross-examination gift: it converts every answer you give into a question about authorship. And the line you must hold under pressure, "I cannot opine on parenting capacity; that requires an evaluation I have not performed", only holds if it is yours, understood from the inside, not a phrase you are carrying into the room. Prepare with your attorney, rehearse with a human, know your chart through the navigation summary, and walk in as the only author of your own voice. The Greenberg and Shuman boundary plus the AFCC Model Standards will carry you through every fitness question: you testify to treatment facts, attributed reports, and observations of your client; the custody opinion belongs to the evaluator the court appointed.

Artifact Four: The Over-Disclosure Audit Before the Box Leaves the Office

The final pass is the one most practices skip and most regret: auditing the actual production set, the physical or electronic stack about to leave, against the order, after all the assembly and redaction is done. Over-disclosure is the dominant failure mode of subpoena response, and it is rarely malicious; it is a duplicate page that escaped redaction, an EHR export that helpfully appended the full document history, a fax cover sheet with a diagnosis in the subject line, an intake packet that included the couples-session note naming the ex-spouse who is now the opposing party. Once it leaves, it cannot be unsent, and the client's remedy conversation starts with your name.

The audit is a two-layer check. Layer one, mechanical, and AI-suited: "Compare this production index against the court order's scope [paste scope]. Flag any document outside the date range, outside the named categories, any duplicate, and any document type on the withhold list (psychotherapy notes, mandated-report records, Part 2 material, third-party charts)." Layer two, human and final: the clinician (with counsel where stakes warrant) turns every physical page or screen of the actual export, because the model audits the index, and the index is not the box. The last items in the box are the cover letter for this production, restating that documents are produced pursuant to the order of [date], remain confidential, and are subject to whatever protective order exists, and your own production log: what left, when, to whom, under what authority, the treating clinician's version of the disclosure manifest the Part 2 lesson taught. The clinician decides, documents, and signs that log. The model never decides what ships; it only helps you prove what did.

The Applied Problem: The Subpoena-Response Cover Letter Asserting Privilege and Minimum Necessary

Your artifact is the cover letter, built as a reusable template with bracketed fields, plus its standing prompt, so the Wednesday envelope meets a Thursday letter instead of a panicked weekend. Step one, choose and verify your statute: CA Evidence Code §1014, TX Health and Safety Code §611, or NY CPLR 4504, whichever governs your license and venue, and confirm the citation with your attorney or carrier resource once, now, not under deadline. Step two, run the drafting prompt: "Draft a one-page subpoena-response letter from a licensed psychotherapist. Paragraph one: acknowledge receipt of the subpoena dated [date] in [case caption] from [requesting party]. Paragraph two: assert that the records sought are confidential communications protected by the psychotherapist-patient privilege under [statute], that the privilege belongs to the patient and has not been waived to my knowledge, and that records cannot be produced absent the patient's written authorization or a court order. Paragraph three: state that any authorized or ordered production will be limited to the specific records and date range identified, and that I reserve the right to request in-camera review or a protective order for clinically sensitive material. Close by directing further communication to my counsel, [name]. Include no clinical information of any kind and do not confirm or deny a treatment relationship."

Step three, the verification pass: read the draft for any sentence that leaks clinical content or concedes the treatment relationship; confirm the statute section is exactly right for your state; confirm the letter asserts rather than argues (the letter preserves the objection, the lawyers litigate it); and strip any AI-added legal flourishes your attorney has not blessed. Step four, route the template to your attorney or carrier for one review, then save it beside the three companion artifacts this lesson described: the records-review package checklist, the chart-navigation prompt, and the over-disclosure audit prompt. Done looks like this: a letter your counsel has pre-approved as a form, a workflow card taped inside your compliance binder reading "1. Call counsel. 2. Notify client and client's counsel. 3. Letter. 4. Nothing leaves without scope, audit, and my signature," and the quiet confidence that the next envelope is a process, not an emergency.

Key Takeaways

• A subpoena is an attorney's demand, not a court order: it neither requires nor permits disclosure of a privileged chart. The privilege belongs to the client under your state's statute (CA Evidence Code §1014, TX Health and Safety Code §611, NY CPLR 4504), and your first two moves are always counsel, then client and client's counsel.
• The treating clinician does not opine on parenting capacity. Per the Greenberg and Shuman role distinction, the APA Guidelines for Child Custody Evaluations, and the AFCC Model Standards, custody and fitness opinions belong to the appointed evaluator; you testify to treatment facts, attributed reports, and observations of your own client.
• Your notes must never adjudicate: attributed report for anything about the non-client parent, behavioral observation for your client, zero fitness language. AI's permitted role is auditing documents for adjudicating language; AI's prohibited role is summarizing sessions in language that adjudicates fitness, and AI never sees the non-client parent's chart.
• The subpoena-response cover letter does three jobs in one page: acknowledge receipt, assert the privilege by statute, and state the minimum-necessary position with a reservation for in-camera review or a protective order. It contains zero clinical content and goes through your attorney every time.
• The attorney records-review package is assembly under scope: an AI-built index and flag list (third-party information, psychotherapy notes under the 45 CFR 164.501 carve-out, mandated-report records, any 42 CFR Part 2 material), with every redact-withhold-produce decision made by humans with counsel.
• AI never drafts testimony: no answers, no talking points, no rehearsal scripts. AI-drafted talking points are a known cross-examination vulnerability, preparation materials can be discoverable, and the role-boundary line only holds on the stand if it is authentically yours. AI builds the chart-navigation summary; your voice stays your own.
• The over-disclosure audit is the last gate: AI checks the production index against the order's scope; the clinician turns every actual page, adds the production cover letter, and signs the production log recording what left, when, to whom, and under what authority. The clinician decides, documents, and signs; nothing ships on a model's judgment.