A client at Jordan's practice searches the scribe vendor's name after seeing it flash on her therapist's tablet, finds a tech blog headline about "AI listening to therapy sessions," and posts in a local parenting group: "Does anyone know if [practice name] records sessions with AI? Nobody told me anything." Within a week, two referring pediatricians have asked their staff to "look into it," one client has canceled, and the practice's silence is writing its story for it. The practice did everything else right: BAA signed, clinicians trained, rubric sign-offs filed, and it still lost control of the narrative because it treated communication as an afterthought to compliance. The rule this lesson teaches: in mental health care, where the entire product is trust, you disclose on your schedule or you explain on someone else's. By the end you will be able to draft the practice's public-facing AI statement, the consent addendum mailing that goes to every existing client, and the talking points for referring primary care physicians, psychiatrists, school counselors, and EAP partners, the three-audience communication kit that lets your practice say one true thing everywhere before anyone else says it first.

Disclosure Is a Clinical Act, Not a Marketing Act

Here is the controlling analogy for this lesson: communicating AI adoption is informed consent at the scale of the whole practice. Every clinician already knows how informed consent works in the room: you explain what you propose to do, in language the client can use, before you do it; you name the risks and the protections without varnish; you make the alternative real, not punitive; and you answer questions without defensiveness, because the consent conversation is itself part of the treatment relationship. Scale that to three audiences (the public, your existing clients, your referral network) and you have this lesson's entire method. Everything that makes consent valid in the room makes communication credible outside it: timing (before, not after discovery), plain language (no vendor adjectives), honest limits (what the AI does not do), and a real choice (opt-out without penalty).

The inverse also scales. What invalidates consent in the room (burying the disclosure, euphemism, springing it after the fact, making refusal costly) is exactly what converts a routine technology adoption into a trust story with the practice as the villain. Mental health clients are, definitionally, people who have decided to tell a stranger things they may have told no one; the implicit contract is that the room is the most protected space in their lives. A client who learns from a tech blog that a machine was part of that room experiences it as a unilateral change to the contract, and no retroactive FAQ repairs that. The communication kit exists so that no client, no referring physician, and no journalist ever knows something about your AI use that you did not tell them first, framed the way you would frame it in a consent conversation: this is what we do, this is why, this is what protects you, and here is your choice.

One True Story, Three Altitudes

Before drafting anything, fix the story, because the three documents are the same truth told at three altitudes, and any daylight between them will be found. The story has five fixed facts, each of which must already be true at your practice (this is the chapter's prior work paying off): what the tools do (assist with clinical documentation: capture, transcription, drafting); what the clinician does (reads and edits every word and signs every note; the signature is a legal attestation, and no AI output enters the chart unreviewed); what the AI never does (it never makes clinical decisions, never assesses risk, never determines diagnosis or treatment; every clinical judgment is made by the licensed clinician); what protects the data (a signed Business Associate Agreement, the retention configuration, in plain words like "the recording is deleted after the note is written" if zero retention is your configuration, and no use of client data to train AI models where your contract provides that); and what the client controls (consent before any use, opt-out at any time, and the rule stated in full: declining AI has no effect on access to care, scheduling, or the therapy itself).

Test every sentence of the story against the two failure modes of practice communication. Overclaiming: "state-of-the-art AI ensures the highest quality care" is vendor language, unverifiable, and the first sentence a board complaint or a journalist will quote back. Underclaiming by euphemism: "we use modern practice-management technology" is technically true and functionally a concealment, and concealment discovered reads worse than disclosure. The voice that works is the clinical voice this program has used throughout: direct, specific, unflinching about limits. "Our clinicians use an AI-assisted tool to help draft session notes. Your clinician reads and edits every word before anything enters your record. The AI never makes any decision about your care." Three sentences a twelve-year-old could repeat accurately, which is the actual standard, because your story will be retold by people who are not you.

In a profession where the product is trust, you disclose on your schedule or you explain on someone else's. Every audience that matters should hear about your AI use from you first, in the same plain clinical voice you would use for informed consent in the room.

Document One: The Public-Facing AI Statement

The public statement is a standing page on the practice website, linked from the footer and the FAQ, titled plainly: "How We Use AI at [Practice Name]." Its audience is everyone you cannot brief individually: prospective clients researching the practice, current clients' family members, journalists, the parenting-group thread. Structure it in five short sections matching the five fixed facts. What we use AI for: documentation assistance, named in plain words, with the boundary in the same breath ("AI helps our clinicians draft session notes; it does not provide therapy, make diagnoses, or assess risk"). Who is in charge: the clinician, always ("your therapist reads, edits, and signs every note; every clinical decision in your care is made by your licensed clinician"). How your information is protected: BAA, retention in plain words, the no-model-training commitment if contractually true, and a sentence inviting the reader to ask for details. Your choice: consent first, opt-out anytime, no effect on care. Questions: a named contact path, because a statement with no door reads as a wall.

Three drafting disciplines keep the statement durable. First, claim only what a document proves: every sentence must trace to the BAA, the vendor contract, the consent addendum, or the practice policy, because the public statement is discoverable in every future dispute and aspirational language becomes evidence. Second, write at the altitude of a worried parent, not a compliance officer: no statute citations on the public page (keep those in the policy), no acronyms unexplained, sentences short enough to survive being read aloud at a school board meeting. Third, date it and version it; technology and contracts change, and "last updated" plus a changelog signals that the practice treats the page as a living commitment, not a one-time press release. This page is also your staff's shield: when any clinician, biller, or front-desk coordinator gets the hallway question, "it's all on our website, and here's the short version" is an answer that scales.

Document Two: The Consent Addendum Mailing to Existing Clients

New clients meet the consent addendum at intake. Existing clients are the harder and more important audience: they formed their contract with the practice before AI existed in it, and they are owed an affirmative, individual communication, not a website update and a form thrust at them in the waiting room thirty seconds before session. The playbook names the instrument: the consent addendum mailed to existing clients. The mailing has three parts. The cover letter, one page, in the practice's clinical voice, ideally signed by the clinical director: what is changing and why ("our clinicians spend hours each week on documentation; this tool returns that time to clinical care"), the five fixed facts compressed to a paragraph, the explicit statement that nothing happens without the client's consent, and the invitation to discuss it with their own therapist before deciding, because the therapist, not the letter, is the relationship. The addendum itself, the same document new clients sign, enclosed for review, with the opt-out option printed as prominently as the consent option, visually equal, because a buried opt-out is a coerced consent. And the timeline with a no-pressure default: clients can sign at their next session after discussing it, mail it back, or do nothing yet, and no recording or AI processing of their sessions occurs unless and until they consent; silence means no, not yes.

Prepare the clinicians before the envelopes go out, in that order, always: clinicians first, then clients, then the public page goes live, then partners, all within the same week so no audience hears it secondhand. Clinicians get the mailing's text in advance, the Concept Card answers from training, and one explicit script for the session where a client brings the letter in: validate the question as exactly the right question to ask, answer from the five facts, and honor either decision without a flicker of pressure, because a client who opts out and feels respected stays in treatment, and a client who consents under perceived pressure has been harmed in the precise dimension this profession exists to protect. Log every consent and opt-out in the system the training lesson built, because the audit question is never just "did clients consent" but "can you show me."

Document Three: Talking Points for Referring Partners

The referral network is the audience practices most often forget, and the one whose trust converts most directly to revenue and reputation. Referring primary care physicians, psychiatrists, school counselors, and EAP partners send you their patients, students, and employees, and their professional credibility rides on the referral; when the parenting-group thread reaches a pediatrician's front desk, the pediatrician needs to already know the answer, or the next referral quietly goes elsewhere. The instrument is a one-page talking points document, delivered personally where the relationship warrants it (a call or a visit for your top referrers) and by professional letter for the rest, framed peer to peer: we are informing you as a colleague whose patients we share.

The content adjusts by audience, same facts, different emphases. For PCPs and psychiatrists: the clinical-responsibility line carries the weight ("every note is reviewed, edited, and signed by the licensed clinician; AI never makes diagnostic, risk, or treatment decisions"), plus what does not change for them (referral process, communication channels, release-governed records exchange all unchanged), and the documentation-quality point a physician audience hears clearly: more complete, more timely clinical documentation supports better coordinated care. For school counselors: lead with the minor-consent picture, that parents or guardians receive the consent addendum and that AI use follows the same consent rules as everything else in a minor's treatment, plus the assurance that nothing about AI changes what the school does or does not receive, which remains governed by releases. For EAP partners: contract language matters; confirm the AI vendor relationship sits inside the practice's HIPAA obligations under a signed BAA, that no EAP member data is used to train models where the contract so provides, and that opt-out is available to every member without affecting their sessions. All three versions end the same way: a named contact for questions, and the offer to provide the public statement and consent addendum for their files, because a partner who can show their own staff your documents becomes your advocate instead of your risk.

The Hard Questions Bank: Answering Without Flinching

Every audience will ask versions of the same hard questions, and the kit is not done until the answers are written, because improvised answers drift toward either overclaiming or defensiveness. Build the bank with the questions you least want asked. "Is a recording of my therapy session sitting on a server somewhere?" Answer with the actual retention facts: if zero retention, say plainly that the recording is deleted once the note is drafted and nothing is stored; if anything else is true, say what is true, because this is the question where euphemism dies loudest. "Can the AI's notes be subpoenaed?" The honest answer: the clinical record has always been subject to legal process, AI does not change what the record legally is, and the practice's records policies govern as before; do not promise immunity that does not exist. "Did you do this to see more patients and make more money?" The honest answer names the real economics without spin: clinicians were spending hours of unpaid evening time on documentation; this returns that time to clinical attention and to sustainable practice, and the practice will not pretend efficiency is not part of it. "What if I say no?" The answer is the rule, again, verbatim: nothing about your care changes, and your therapist will be glad you asked. And for the press version of any question, one operational rule: a single named spokesperson, the public statement as the spine of every answer, and no improvisation about technical details the spokesperson cannot verify, because "let me confirm and come back to you" beats a confident error in every news cycle that has ever run.

Sequencing the Launch and Maintaining the Story

The kit deploys in a fixed order over roughly one week, and the order is the strategy: clinicians briefed first (they are the front line and must never be surprised by a client holding a letter they have not seen), the client mailing second, the public statement live third (so the mailing's recipients who go looking find your page, not the tech blog), and partner talking points delivered the same week (so the pediatrician's front desk has the answer before the thread arrives). Internally, the champions from the previous lesson carry the question traffic, and the question-theme log becomes the early-warning system for which message is not landing; if five clients ask the same unanswered question, the hard-questions bank grows by one and the public page gets a clarifying line.

Maintenance is quarterly and event-driven. Quarterly: re-verify every factual claim in all three documents against the current vendor contract, BAA, and retention configuration, because a public statement that outlives its facts converts from shield to liability. Event-driven: any vendor change, subprocessor change, retention change, or new tool triggers the same kit update and, where material, a fresh client communication, because consent given to one configuration does not silently transfer to another. And once a year, run the parenting-group test on your own kit: have someone outside the practice read only your public statement and then try to write the alarming post; if the post writes itself around something your statement failed to address, the statement, not the poster, is what needs fixing. Practices that pass this test do not just avoid crises; they convert transparency into referral advantage, because in a market where most practices say nothing, the practice that explains itself plainly is the one the careful pediatrician trusts with her patients.

The Applied Problem: The Three-Audience Communication Kit

Your artifact is the Communication Kit: the public AI statement, the client mailing (cover letter plus addendum package), and the partner talking points in three audience versions, plus the hard-questions bank and the launch sequence. Build it in this order. First, write the five fixed facts as single sentences and verify each against its source document: the vendor contract for capabilities, the BAA and configuration for data protection, the consent addendum for the client's rights, the practice policy for the clinician-signature rule. Nothing else gets drafted until the five facts survive verification, because every document is an arrangement of these sentences. Second, draft the three documents with this prompt, run once per audience: "Draft a [public website AI statement / one-page cover letter to existing clients accompanying a consent addendum / one-page talking points document for referring (PCPs and psychiatrists / school counselors / EAP partners)] for a behavioral health group practice. Use only these verified facts: [paste the five facts]. Voice: a senior clinician explaining to a peer or a client, plain language, no marketing adjectives, no statute citations. Must state explicitly: the clinician reads and edits every note before signing; AI never makes clinical, diagnostic, or risk decisions; consent comes first and opting out has no effect on care. Include a named contact path." Third, write the hard-questions bank: minimum eight questions including the retention question, the subpoena question, the money question, and the what-if-I-say-no question, each answered from the facts without flinching.

Then run the verification pass, which for this artifact has a special edge: every claim in every document gets traced to a source document by someone other than the drafter, because the public statement is the most discoverable thing the practice will ever publish about its AI use. Run the twelve-year-old test on the public statement (can a layperson retell it accurately?) and the parenting-group test (can an outsider write the alarming post around a gap?). Have one clinician, one champion, and if possible one actual client advisory voice read the client letter for tone: respectful, unhurried, choice-forward. Have your most senior referring relationship preview the partner version and tell you what their front desk would still ask.

"Done" looks like this: three documents plus the bank, every claim source-traced, the launch sequence calendared in the correct order (clinicians, clients, public page, partners, one week), consent and opt-out logging confirmed live, the spokesperson named, and the quarterly re-verification recurring on someone's calendar with their name on it. When the next curious client searches the vendor's name, she finds your page first, written in your voice, saying exactly what her therapist already told her, and the parenting-group post never gets written.

Key Takeaways

• Communicating AI adoption is informed consent at the scale of the practice: explain before you act, in plain language, with honest limits and a real choice. In a profession where the product is trust, you disclose on your schedule or you explain on someone else's.
• Fix one true story with five verified facts before drafting anything: what the tools do, what the clinician does (reads, edits, signs every note), what AI never does (no clinical, diagnostic, or risk decisions), what protects the data (BAA, retention in plain words, no model training where contractual), and what the client controls (consent first, opt-out anytime, no effect on care).
• The public statement claims only what a document proves, is written at the altitude of a worried parent, and is dated and versioned; it is discoverable in every future dispute, so vendor adjectives and euphemisms are both liabilities. The twelve-year-old retell test is the real standard.
• Existing clients get an affirmative individual mailing: a one-page clinical-voice cover letter, the addendum with opt-out printed as prominently as consent, and a no-pressure default where silence means no; nothing is processed until the client consents, and either decision is honored without a flicker of pressure.
• Referring partners get audience-tuned talking points: clinical responsibility and unchanged coordination for PCPs and psychiatrists, the minor-consent and release-governed picture for school counselors, and BAA-anchored contract assurances for EAP partners, each ending with a named contact and the offer of your documents for their files.
• Build the hard-questions bank before launch: retention, subpoena, the money question, and what-if-I-say-no, each answered from facts without flinching; one named spokesperson, the public statement as the spine, and "let me confirm" over confident error.
• Sequence is strategy: clinicians first, client mailing second, public page third, partners the same week. Maintain quarterly by re-verifying every claim against the current contract and configuration, and re-communicate when the configuration materially changes, because consent to one configuration does not transfer to another.