How to Talk to Clients About AI Documentation: A Therapist's Guide to the Consent Conversation

A practical guide for therapists on how to have the AI documentation consent conversation with clients. Sample scripts, answers to common client questions, how to handle refusal, and what to document.

Most therapists who hesitate to adopt AI documentation tools are not stuck on the technology. They are stuck on the conversation.

"How do I bring this up with clients?" is one of the most common questions before adoption. Not "how does the software work?" Not "will it be accurate?" The conversation itself is the sticking point. And it makes sense. You spend your professional life navigating difficult conversations with care and skill. The idea of explaining AI to a client who came to you carrying something tender feels like it deserves the same rigor.

This guide is not about consent forms. A separate article covers how to document informed consent in the clinical record. This guide is about the actual conversation: what to say, when to say it, how to answer the questions clients will ask, and what to do when a client says no.

Why This Conversation Matters

Informed consent for AI documentation is both an ethical obligation and, increasingly, a legal one. The APA 2025 Practitioner Pulse Survey found that 67% of psychologists' clients express concern about data breaches when they learn AI is involved in their care, and 60% name hallucinations as a worry. Those concerns are not irrational. They reflect the broader cultural anxiety about AI, and clients bring that anxiety into your office.

Beyond client concerns, state law is beginning to mandate disclosure. Illinois Public Act 104-0054, enacted in August 2025, requires written client consent before any AI tool is used to record or transcribe psychotherapy sessions. New York's proposed S.8484 mirrors this framework. Forty or more AI-related mental health bills are moving through state legislatures as of 2026. The landscape is changing fast.

But compliance is not the whole story. The therapeutic relationship is built on transparency. Clients who discover after the fact that AI was involved in their documentation may feel deceived, regardless of what any form said. The active consent model, meaning regular check-ins rather than a one-time signature, is now considered best practice by multiple clinical ethics commentators. The consent conversation is not a hurdle before therapy begins. It is part of the relational fabric of the work.

Timing: When to Introduce AI Use

New Clients at Intake

Intake is the right time to introduce AI documentation. You are already covering confidentiality, limits to confidentiality, your note-taking practices, and how records may be shared. AI use belongs in this same cluster of disclosures. Clients are in information-receiving mode. They expect to learn how your practice works. An AI disclosure at intake lands as a practice policy, not a surprise.

Introduce it after you have established some basic rapport, not as the opening line of the session. A few minutes into the conversation, once you have welcomed the client and explained what to expect, is natural.

Existing Clients Mid-Treatment

Introducing AI mid-treatment requires more care, because the relational frame is already established. Clients have developed an implicit understanding of how you work. Changing that understanding, even for something practical like documentation, can feel like a shift in the ground beneath the relationship.

The best approach is a brief, transparent check-in at the start of a session before you begin using the tool. Do not bury the disclosure in paperwork and hope the client reads it. Have the conversation. Give the client a chance to respond.

Here is how that might sound for an LMFT who has worked with a couple for two years:

"Before we get started, I want to let you both know about something I am adding to my practice. I have started using an AI tool to help me write my session notes after we meet. I still write a summary of what we cover, and the AI helps me organize it into the structure my notes follow. The session content comes from what I write, not from any recording of our conversation. I wanted to be transparent about that, and I am happy to answer any questions."

This framing covers what the tool does, what it does not do (record the session), and invites dialogue rather than demanding acceptance.

The 30-Day Trust Window

Research on AI documentation adoption shows most therapists feel genuinely confident with a new tool by month two. The first thirty days involve adjustment: adjusting the template, getting comfortable with the output, building the habit. This means the consent conversation you are having at intake with your first new client while still learning the tool may feel a bit uncertain on your end. That is normal. Clients respond to your genuine transparency about the tool more than to your polished fluency with it. You do not need to have all the answers before you start the conversation.

How to Frame It for Different Tools

The consent conversation changes depending on what kind of AI tool you use. There are two architecturally distinct categories, and how you explain each to a client is meaningfully different.

Recording-Based Tools (Ambient Scribes)

Some AI documentation tools work by recording the session. The audio is transcribed and the AI generates a clinical note from the transcript. These are sometimes called ambient scribes. Examples include tools that a therapist places in the room and that listen throughout the session.

If you use a recording-based tool, your disclosure must cover the recording itself. Clients in all-party consent states (California, Florida, Illinois, Massachusetts, Pennsylvania, and others) must give explicit consent before any recording begins. Even in one-party consent states, recording a therapy session without the client's knowledge raises significant ethical concerns. The consent conversation for an ambient scribe is more involved: you are telling the client that the session will be recorded, that the audio may be processed by a third-party service, and what happens to that audio afterward.

Clients with trauma histories, legal proceedings, domestic violence situations, or occupational reasons (military, government, law enforcement) may decline specifically because of the recording component. That is a clinically meaningful response, not just a preference.

Generation-Based Tools (Post-Session AI)

Other tools work differently. The therapist writes a brief summary of the session after it ends, and the AI structures that summary into the clinical note format (SOAP, DAP, BIRP, or whatever the clinician uses). No audio is captured. No transcript exists. The AI only sees what the therapist chose to write.

This is called a generation-based AI workflow, and it is what most solo private practice therapists now prefer, according to 2026 market research. The consent conversation for a generation-based tool is substantially simpler.

You are not telling the client that their words are being recorded. You are telling the client that the notes you write after your session are organized by an AI tool, and that the content comes only from your own written summary. The most anxious part of the AI consent conversation for recording-based tools (an AI is listening to us) does not apply.

NotuDocs, for example, is a generation-based tool. A therapist types their post-session summary, and the AI structures it into the note. There is no session recording, and nothing the client says in the room is processed by the AI. This does not make the disclosure optional, but it does make the conversation considerably less fraught.

What to Actually Say: First-Session Scripts

The following scripts are starting points, not scripts to read verbatim. Adapt them to your voice.

Script 1: New Individual Client, Generation-Based Tool

"As part of explaining how I work, I want to tell you about my note-taking process. After each of our sessions, I write a summary of what we covered and any clinical observations I have. I use an AI tool to help me organize that summary into the format my notes follow. The AI only works from what I write, not from any recording of our conversation. It helps me keep documentation consistent and frees me up to be more present with you during our time together. Do you have any questions about that, or any concerns I should know about?"

Pause and listen. Do not move on immediately. Give the client space to respond.

Script 2: Intake with a Client Who Seems Tech-Cautious

This is a fictional 47-year-old new client, James, who mentioned during phone intake that he is "not big on technology." You may pick up skepticism before the disclosure happens.

"I want to walk you through how I keep notes on our sessions, because it is a bit different from what some therapists do. After we meet, I write a summary of our session myself. I use a software tool to help format those notes, and that tool uses AI to do the formatting. But the tool does not have access to any recording of what you say here. It only processes my written summary. The session itself stays between us. I mention this because some people have questions or preferences around AI, and I want to give you the chance to ask anything before we get started. Does that work for you?"

James may have no reaction at all, or he may ask how his information is stored. Having a clear, honest answer ready (covered below) is the preparation the conversation requires.

Script 3: Existing Client Mid-Treatment

"I want to flag something before we dive in today. I have started using an AI tool to help me with my session notes. I still write the summary myself after each session, and the AI helps me put it into the structure my notes follow. Nothing about our actual session is recorded. I wanted to tell you directly rather than just adding it to paperwork, because I think transparency matters in how we work together. How does that land for you?"

The phrase "how does that land for you?" is intentional. It is open-ended, non-coercive, and gives the client clinical latitude to name an emotional reaction rather than just a logical yes or no.

Answering Client Questions

Here are the questions clients most commonly ask, with language you can use to answer them.

"Is this HIPAA-compliant? Who has access to my information?"

This question requires an honest, specific answer, not a blanket "yes."

First, clarify what the tool you use actually does with data. Does it have a signed Business Associate Agreement (BAA) with your practice? If you use a tool that offers a BAA, say so clearly. If the tool you use does not offer a BAA, be transparent about that and explain what privacy protections are in place.

For generation-based tools where the therapist writes the post-session summary: the only clinical information entering the AI tool is what you, the therapist, chose to write. The client's full session content, their name, their diagnosis, their history all remain in your control. You decide what goes into the summary.

A clear answer might sound like:

"The tool I use processes the summary notes I write, not a recording of our conversation. I am happy to tell you the name of the tool so you can look at their privacy policy if you would like. What I can tell you is that you control whether you want me to use it, and the information it sees is limited to my written summary, not the full session."

"Will a computer read my trauma?"

This is one of the most emotionally loaded questions clients ask, and it deserves a careful, humanizing response.

"That is a fair thing to wonder about. What the AI tool sees is the structured summary I write after our session, which is clinical language describing what we worked on. It does not see a transcript of what you said, and it does not have any way of knowing the emotional context behind the words. The writing is mine, the structuring is the tool's, and the clinical judgment about what to include is always my decision."

"Can I say no?"

Yes. Full stop. And say that clearly.

"Absolutely. If you prefer that I not use an AI tool for your notes, I will document your sessions manually. Your choice here does not affect your care. I will ask you to let me know, and I will note your preference in your chart."

Voluntariness is a core element of informed consent. A consent that comes with implied consequences is not genuine consent. Clients who ask this question may be testing whether you actually mean it. Say it plainly, and mean it.

"Who else reads my notes?"

This question is really about privacy and control, not just logistics.

"Your clinical notes are part of your protected health record. Aside from me, notes may be reviewed in specific situations: if you authorize a release to another provider, if there is a legal subpoena, in the event of an audit by your insurance payer, or in the rare clinical situations where safety concerns require disclosure, which I explained in our initial disclosure. The AI tool that helps me format my notes does not have independent access to your record, and it does not share information with any other party."

If you are using a tool that routes data through servers or is backed by a larger platform, give an honest summary. "I use [tool name], which processes my written summaries on their secure servers. I recommend looking at their privacy policy if you want the full technical picture."

Handling Hesitation and Refusal

A client who hesitates is not a client who said no. Hesitation is often a request for more information, or a moment of processing. Give it space.

If a client says "I am not sure about this," a useful response is:

"You do not have to decide right now. Let me know if you have more questions after today, and we can revisit it."

Do not push for a yes in the same session where you raised the topic. The therapeutic relationship takes priority over documentation efficiency.

If a client says no:

Document the refusal clearly. Return to manual note-writing for that client, or adapt your workflow so AI is not involved in their record. Do not shame the client for declining, and do not treat the refusal as a clinical problem to be solved. A client's wish to limit AI involvement in their documentation is not resistance in the clinical sense.

Consider whether the refusal opens a useful therapeutic door. For some clients (particularly those with histories of surveillance, violation of privacy, or loss of control over their own information) the question of who has access to their information is clinically significant. The disclosure conversation may reveal something worth holding.

You can also return to the topic later if circumstances change:

"A few months ago when we discussed AI documentation, you preferred that I keep notes manually. I want to check in about whether that is still your preference, or whether you have any new questions about how it works."

A client who said no six months ago may say yes today. That is ongoing consent, not pressure.

Ongoing Check-Ins

Active consent is not a one-time event. Best practice for AI documentation includes periodic check-ins to confirm clients still consent, particularly if the tool or your workflow changes. Once per year, or whenever you change tools, is a reasonable standard.

Framing for an annual check-in:

"I want to revisit something briefly. About a year ago, I told you I use an AI tool to help structure my session notes. I want to confirm that is still okay with you, and give you the chance to ask any questions or change your mind. Nothing has changed on my end, but I want to make sure we are still on the same page."

This takes thirty seconds. It reinforces that consent is ongoing, not contractual.

Written Disclosure Language

The verbal consent conversation should be supplemented by written disclosure. This does not replace the conversation; it confirms and documents it.

Language to include in your informed consent document or practice policies:

Use of AI-Assisted Documentation

This practice uses an AI tool to assist with formatting clinical session notes. [Describe: the tool name, whether it records sessions, what data it processes, and any applicable privacy certifications.] The use of AI in note documentation is voluntary. You may decline this practice at any time by notifying your therapist. Declining will not affect the quality or availability of your clinical care. If you have questions about how your information is handled, please ask.

Keep the language plain. Avoid technical jargon. If the tool you use does not offer a BAA, do not write language that implies it does.

The consent conversation needs to appear in the clinical record, not just the consent form. Document it in your intake note or in a session note if the disclosure occurred mid-treatment.

A brief, accurate note is sufficient:

"Informed consent for AI-assisted documentation reviewed verbatim with client. Client was informed that a generation-based AI tool is used to structure post-session written summaries into clinical note format. No session recording is involved. Client was informed of their right to decline. Client asked [specific questions if any]. Client consented to AI-assisted documentation / Client declined AI-assisted documentation and manual documentation will be used going forward."

If the client had questions, note the questions and your responses. If the client declined, note that clearly. This protects you clinically and legally, and it models the rigor you are asking the tool to support.

Making it too technical. Clients do not need to understand machine learning. They need to understand what happens to their information.

Rushing it. The consent conversation is a conversation, not a disclosure dump. Leave time for questions.

Burying it in paperwork. A signed form is not a substitute for a verbal explanation. Many clients sign intake paperwork without reading it closely. Say the important things out loud.

Treating refusal as a compliance problem. Some clients will say no. That is their right. The note-writing burden falls back on you, but the therapeutic relationship remains intact.

Not revisiting it when tools change. If you switch from a generation-based tool to a recording-based tool, the original disclosure is no longer accurate. Have a new conversation.

Overstating privacy guarantees. Do not say "completely private" or "totally secure." No data system offers guarantees that broad. Say what is true: what the tool does, what it does not do, and what privacy measures are in place.

A Pre-Session Rehearsal Script

Read this through before your next intake. Adjust the bracketed parts to match your tool and your voice.

"As I walk through how my practice works, I want to mention my documentation process. After our sessions, I write a summary of what we covered. I use [tool name] to help me organize that summary into a clinical note. [If generation-based: The tool works from my written summary, not from any recording of our conversation.] [If recording-based: The session is recorded, and the AI uses that recording to help generate my notes. The recording is [describe retention policy].]

You have the right to decline this, and I will ask you to let me know if you have any concerns. If you prefer that I not use AI in your documentation, your care will not be affected, and I will keep notes manually. Do you have any questions, or anything you want to tell me about your preferences around this?"

Pause. Wait. Let the client respond before moving on.

Before Your First AI-Assisted Session

Confirm what your AI tool does: does it record the session or work from your written summary?
Know your tool's data handling practices and be ready to describe them honestly
Update your informed consent document to include AI documentation disclosure
Know what you will do if a client declines (have a manual note-writing plan)

At Intake with New Clients

Raise the topic after initial rapport is established, not as the opening disclosure
Explain what the tool does and does not do (especially: does it record the session?)
Confirm that consent is voluntary and will not affect care
Give the client space to ask questions before moving on
Document the conversation and the client's response in the intake note

Mid-Treatment with Existing Clients

Introduce the change at the beginning of a session, not as a footnote
Use "how does that land for you?" or a similar open-ended response invite
Do not push for a yes in the same session if the client hesitates
Document the new disclosure and the client's response

Handling Client Questions

"Is it HIPAA-compliant?" — Answer specifically: what does the tool process, does it have a BAA?
"Will a computer read my trauma?" — Clarify that AI sees your written summary, not session audio or transcripts
"Can I say no?" — Yes, plainly, with no implied consequences
"Who reads my notes?" — Explain the standard list: provider, authorized releases, subpoenas, audits

Schedule an annual check-in to reconfirm consent
Have a new conversation any time you change tools
Document each check-in in the clinical record

Related reading: