AI Scribe for Counsellors: A Safe Trial Plan With Consent and Opt-Out

Trial an AI scribe with three to five clients who have each given written consent, over about six weeks. Keep writing your own notes alongside the draft, compare the two after every session, let any client opt out at any point without explanation, and set the audio deletion rule before the first recording.

Counselling is not the same as a medical visit. Clients say things they have never said aloud, and a recording can feel very different from a clinician typing. That is why the trial has to be small, slow and built around the client's freedom to say no, with your own notes remaining the record until it ends.

Follow me on Instagram@sagnikteaches

Before any client hears about it

Do the groundwork alone first. It takes a couple of hours and spares you from improvising later.

Connect on LinkedInSagnik Bhattacharya

1. Tell your supervisor. A supervisor can spot risks you have not, and it protects you if anyone later asks how the decision was made.
2. Ask your insurer and membership body. Some publish guidance on recording and on AI tools. Save whatever they send you.
3. Read the vendor's data terms, not the home page. You want to know where audio goes, how long it stays, whether it trains models, and whether a business associate agreement (or its local equivalent) is available.
4. Pick the tool and the plan. A trial is a good use of a free tier or free trial. Mentalyc lists a 14-day free trial with no card, covering 15 notes. Jane gives each practitioner five free notes a month, and Heidi lists a free plan. Berries says it does not store session recordings, and its free Trial plan covers the first 20 sessions, then 10 a month, with no card. Five free notes a month will not cover four clients weekly, so check the cap against your plan (the sums are worked through below).
5. Set the deletion rule. Choose the shortest retention that suits you. In Jane, a clinic-wide policy can delete recordings immediately after the draft is generated, and the help guide says that is the default when you switch the scribe on. Mentalyc lists audio deleted once the note is ready. Confirm the same setting is on before you record anyone.

If your tool is part of a practice-management system you already pay for, my notes on Jane for counsellors, with costs and data location cover that case in detail.

Subscribe on YouTube@sagnikteaches

Who to ask, and who to leave out

Pick clients, not sessions. Ask the people with whom the working relationship is steady and the content is ordinary. Leave out anyone for whom recording could do harm or confuse the work.

Say a counsellor with 18 clients in a private practice. After applying the filters above, perhaps six remain. She approaches five and keeps the sixth in reserve. Two say no straight away, one wants time to think, and two say yes. She does not chase the doubters. A trial with two clients is thin, so she widens the filter slightly and asks one more, ending with three consenting clients. That is a respectable trial. These numbers are illustrative.

The consent form, written for this situation

Do not reuse a generic technology form. A counselling client needs to know what is recorded, who gets it and what happens if they change their mind. Here is wording to adapt, written in plain language. Your own regulator, insurer or lawyer should review it.

TRIAL OF AN AI NOTE-TAKING TOOL

I am trying an AI tool that listens to our session and drafts my
notes. I would like to ask whether you are willing to take part.

What it does
- It records the audio of our session, turns it into text, and
  drafts a session note. I read the draft, correct it and decide
  what goes into your record. I also write my own notes during
  the trial so I can check the tool.
- The tool is called [name], run by [company]. Your session
  content is processed by [company and any named sub-processors].
- The audio is deleted [immediately after the draft is created /
  after __ days]. The tool does / does not use your recordings to
  train AI [confirm from the vendor's terms].

What it does not do
- It does not make decisions about your care.
- It is used for [number] weeks with a small number of clients.
  After that I will decide whether to keep using it.

Your choices
- You can say no, and nothing about our work will change.
- You can say yes for some sessions and not others.
- You can stop at any point, including in the middle of a session.
  Just say "please stop" and I will switch it off.
- You do not have to give a reason.
- You can ask me to delete any recording or draft from the trial.

[ ] I agree to take part in the trial
[ ] I would like more time to think
[ ] I do not wish to take part

Name: __________   Date: __________   Signature: __________

Give clients the form at least a day before the session, by hand or by secure message, so they can think without you watching. Do not ask in the room at the start of a session, because the pressure to agree is greater there. Keep a signed copy with your trial records, separate from the client notes.

The opt-out has to be easy in the moment

A written form is not enough on its own. The real test comes in session, when a client mid-sentence realises they would rather not be recorded. Prepare for that with three things.

• A start-of-session check. Before pressing record, say: "Are you still happy for me to record today? You can say no whenever you like." Then wait.
• A stop phrase. Agree one with the client: "Please switch it off." You do it immediately, tell them it is off, and write the rest by hand.
• A no-penalty rule for yourself. If a client withdraws, respond warmly and move on. Do not ask why, do not sigh, and do not hint that it was useful.

Decide in advance what you will do with a recording that is part-made when the client says stop. Delete it, check that the deletion took effect, and note the date. Do not assume the tool's deletion rule catches it. Jane's help guide, for instance, says a recording started from the Chart tab that has not yet been connected to a chart entry is not covered by the clinic's deletion policy and stays saved until a summary is generated, so a recording you stopped halfway can sit in the account until someone deletes it by hand. Stop a recording halfway through your mock session in week 1 and look at the recordings list, so you know what your tool does before a client relies on it. If a client withdraws after the session, delete their recordings and drafts from the trial and tell them it is done.

Mentalyc and Jane both describe routes that do not need a recording at all, such as dictating a summary after the session, or typing a note and refining it with the tool. These are useful for a client who declines audio but still lets you try a form of AI help on your own words. Treat that as a separate, clinician-only trial and say so in your records.

The conversation where you ask, with a sample script

How you ask shapes the answer. Keep it short, neutral and unhurried, and hand over the written form for the decision itself. An illustrative version, in your own words:

"I'm trying out an AI tool that could draft my session notes from
a recording. I'm testing it with a few clients first, and I'd only
include you if you're comfortable. I'll give you a form that
explains it, and you can take a week to decide. If you say no, or
you say yes and then change your mind, nothing about our work
changes. Is it okay if I send you the form?"

Notice what is missing: any claim that it will help the client, any hint that you would like a yes, and any request for an answer on the spot. If the client asks "What happens to the recording?", answer from the vendor's terms and your deletion setting. If you do not know, say so, and tell them you will find out before they decide.

Three clients, three outcomes

A trial rarely goes the way the plan reads on paper. These illustrative cases show what to do when it does not.

Client A agrees and stays comfortable. She signs, forgets the tool is running after ten minutes, and her sessions produce drafts that need three or four edits. You still write your own note first, and you ask at the start of each session whether she is happy to continue. By week four you have six pairs of notes and a clear sense of the tool's habit of flattening her phrases. Her only comment is that she liked being asked each time.

Client B agrees but looks uneasy. He signs the form, yet in session two he glances at the device twice and speaks more guardedly than usual. You name what you noticed: "I wonder whether having this on is changing how it feels to talk." He admits it is. You switch it off, finish by hand, and offer him the choice again next week with no expectation. He declines. You log it as a withdrawal, delete the recording and draft, and your trial now has two clients. Nothing about the work was damaged, which is the point of the opt-out.

Client C says no. She would rather not, and says so briefly. You thank her, do not probe, and carry on as before. She is not in the trial and does not appear in any score. If you notice yourself feeling disappointed, that is useful information about how neutral you are being.

Across the three, the trial has produced two clients' worth of paired notes and a lesson in how consent works in practice. That is a good return for six weeks.

When a risk disclosure happens while the tool is running

Sooner or later a recorded session may include a statement about harm to the client or someone else. The scribe does not decide what to do about it, and it must never be the thing you rely on to notice. You respond as you always would, under your own professional duties and policies. Afterwards, three checks apply to the draft.

1. Is the disclosure in the draft, in the client's words and at the right point in the session?
2. Is your response in the note, written by you, not suggested by the tool?
3. Has the draft softened or reframed anything? A phrase such as "client expressed some difficult thoughts" is not an adequate record of a risk statement.

If a disclosure ever gets lost in a draft, treat that as a stop-or-adjust signal for the trial and tell your supervisor. A rule worth adopting from day one: for any session containing a risk statement, your own handwritten or typed note is the record, and the draft is only a cross-check.

Will a free tier last the trial? The sums for three clients

The routine below has live sessions in weeks 2 to 6. With three clients seen weekly that is 15 sessions; with five, 25. Free tiers rarely line up with that, so do the count in week 0 rather than discovering in week 3 that the tool has stopped drafting.

• Jane's free tier: five notes a month per practitioner, roughly seven or eight across six weeks. Enough for a mock session and one client, not three. Unlimited is listed at $15 a month per practitioner.
• Mentalyc's trial: 14 days and 15 notes, which covers about the first two live weeks. Its Mini plan, listed at $19.99 a month (or $14.99 a month billed annually) for 40 notes, would cover the rest of a three- to five-client trial.
• Berries' Trial plan: the first 20 sessions free, then 10 a month. Three clients for five weeks fits inside it; five clients would pass 20 sessions during week 5.
• Heidi's free plan: it lists unlimited AI documentation, so the session count is not the constraint. Its templates and features are.

Here is how that plays out for an illustrative counsellor with three trial clients on Mentalyc. The 14-day trial covers weeks 2 and 3, six sessions, well inside the 15-note allowance. She then pays $19.99 for one month of Mini to cover weeks 4 to 6, nine more sessions, and cancels or continues at the week-6 review. The subscription cost of the whole trial is about $20. The larger cost is her time. During the trial she writes her own note (nine minutes in the scoring example further down), reviews the draft (four minutes) and fills in the scoring sheet (two minutes): 15 minutes a session instead of nine, so 15 paired sessions add about 90 minutes over six weeks. If the tool earns its place, the saving afterwards is the gap between reviewing and writing, five minutes a session in that example, or 100 minutes a week across 20 weekly sessions. Neither figure is a promise; your own scoring sheet supplies the real ones.

A six-week routine you can follow

The plan below uses weekly sessions, but adjust it if you see clients fortnightly. The aim is to produce about 12 to 18 paired notes (yours and the tool's), enough to judge quality.

Writing your own note first matters. If you read the draft first, you will anchor to it and agree with things you might have written differently. The point of the comparison is to see what the tool adds or gets wrong, not to check that it sounds plausible.

Scoring each pair of notes

Keep the sheet short enough to complete in two minutes. Here is a filled-in example for one illustrative session (no real client).

Client code: C2   Session: 3   Date: [date]
Draft minutes to review: 4   My own note minutes: 9

Accuracy (1 = poor, 5 = excellent): 4
  Wrong or invented items: 1 ("client has trouble sleeping" - not said)
  Missed items: 1 (client's mention of a bereavement anniversary)

Tone (does it sound like me and the client?): 3
  Draft was clinical and flat; lost the client's own phrase
  "everything feels loud".

Risk content (did it handle risk statements correctly?): 5
  No risk disclosures this session; check performed.

Client response at end: "Fine. I forgot it was on after ten minutes."
Would I have signed this unedited? No. Edits needed: 3.

What the sheet tells you across several sessions: whether the draft is faster to review than your own note is to write, whether invented or missed items recur, and whether the tone fits your way of working. Pay special attention to missed items. A draft that quietly drops the most emotionally important moment of the session looks tidy and fails the client.

Comparing the draft with your own note

Lay the two notes side by side and look for four kinds of difference.

1. Things only the tool wrote. Check each one against the session. Anything you cannot confirm is an invention and should be struck out.
2. Things only you wrote. These are usually your clinical impressions and plans, which the tool has no business creating. Notice whether it tried and got them wrong.
3. Things both wrote but differently. Wording differences are fine. Differences in fact, such as dates, names, relationships or the order of events, are not.
4. Things neither wrote. A good prompt for yourself at the end: what was the most important moment of this session, and does either note show it?

Here is a short before-and-after to show the kind of difference to look for. The client said: "I haven't been sleeping since my mum moved in, and I keep thinking I'm failing her." A draft that reads "Client reports insomnia related to family caring responsibilities and low mood" is not wrong, but it replaces the client's own meaning with labels, and "low mood" was never said. Your own note might read: "Sleep poor since mother moved in; client voiced fear of failing her." For many counsellors, that gap is the whole question about whether the tool suits their way of working.

Data location and retention, written down

Before the trial ends, make a one-page record of where each piece of data went. It will be useful whether you adopt the tool or not.

Transcripts are the item most often forgotten. Mentalyc, for example, deletes audio once the note is ready on every plan but keeps transcripts and notes until you delete them, and lists automatic transcript deletion only on its Super and Team plans, so on a cheaper plan the transcript row in your table is a manual job for the end of each trial week.

If you practise where the law restricts moving client data abroad, also record the region where the vendor stores it. Jane, for example, says UK clinics' recordings and drafts are stored in the UK and not used to train models. Whatever the tool, write the answer you were given, the date, and who gave it.

The decision at week six

Hold a short review with your supervisor and decide using rules you wrote in week 0, not feelings in week 6. Here is a rule set to adapt.

• Continue to a wider trial if: at least two-thirds of drafts needed three edits or fewer, nothing serious was nearly signed, no client withdrew under pressure, review time was clearly shorter than writing time, and the vendor's data terms held up.
• Adjust and repeat if: accuracy was mixed but the errors were of one kind that a template change could fix.
• Stop if: a draft contained a serious error you nearly signed, a client felt pressured, the vendor's answers changed, or you found yourself steering sessions to suit the tool.

Stopping is a good result. You have spent six weeks and a handful of free notes to learn that the current way suits you better. If you decide to continue, widen slowly: add a few more clients using the same form, and keep the first scoring sheet going for a month. For a view of what changes when several therapists adopt the tool, the group practice roll-out plan picks up where this trial ends, and the solo-practice cost sums help you decide whether a paid plan is justified.

If it would help to talk the plan through before you start, that is what my AI implementation consultation does on a 1:1 call, with no obligation to adopt anything.

Counsellor questions before a trial

How many clients should I include in a trial?

Three to five is enough to learn how the tool handles your style without exposing many people to something unproven. Choose clients with straightforward presentations, who understand the idea, and who you expect to attend regularly so you can compare several sessions.

Should I tell my supervisor and insurer?

Yes. Tell your supervisor before the first session, and ask your professional indemnity or liability insurer and your membership body whether they have guidance on AI note-taking. Keep their replies with your trial records.

What if a client agrees but seems uneasy?

Take the unease seriously. Offer a session without recording, and make clear that nothing changes in the work if they say no. A client who agrees to please you has not really consented, so check in again at the next session.

Can I keep the recording for supervision or training?

Only with separate, specific consent. Consent to a scribe is consent to draft a note, not to keep audio for other uses. If your tool deletes audio after drafting, keep it that way, and ask the client before any other use.

When should I stop the trial early?

Stop if a client withdraws and the process feels pressured, if a draft contains a serious error you nearly signed, if the vendor's data terms turn out different from what you were told, or if you notice yourself editing the session to suit the tool.

Further reads

• AI Scribe for Occupational Therapists: A Buyer's Checklist — A buyer's checklist from a neighbouring allied-health role.
• How to Roll Out an AI Scribe Without Losing Patient Trust — Wording and habits that help clients feel comfortable about a scribe.
• AI Scribes for Doctors: Six Note Errors to Catch Before You Sign — Six error types to look for when you review any draft.
• AI Scribe Patient Consent: Implied or Explicit for UK Clinics? — Whether consent can ever be implied for a scribe.
• AI Scribe Security: Audio, Retention and BAA Vendor Questions — Questions to put to a vendor about audio, retention and BAAs.
• AI Scribe Consent Form: US Wording and State Recording Laws — A US consent form template and a state recording-law map.
• Heidi Health for Dentists: Setup, Templates and Patient Consent — A step-by-step Heidi setup for a dental practice: the plan to start on, template tweaks, getting notes into the PMS, and patient consent wording.
• How to Use AI on SimplePractice: Note Taker or Care Aide? — A decision rule for solo therapists: when SimplePractice Note Taker is enough, when Care Aide earns its extra $24, and how to trial each safely.
• Starting a Private Counselling Practice in the UK: What AI Can Do — The admin a new UK private counsellor sets up first, which parts AI can draft safely, which it must never touch, and what first-year running costs look like.
• What's the Best AI Tool for Therapists? A Rule by Practice Size — A rule for choosing a therapy AI tool by practice size, with list prices, BAA checks and worked costs for a solo, three-clinician and nine-clinician practice.
• AI Charting for Dietitians: Drafting ADIME Notes With a Review Step — Draft Assessment, Diagnosis, Intervention and Monitoring notes with AI, then check each line before it enters the record.
• AI Note Generator for Counsellors: Turn Your Own Bullets Into Notes — Type your own session bullets, strip identifiers, and have AI draft the note you then edit and own. Prompts and examples included.
• Therapist Intake Form Template: Build It in ChatGPT, Then Check It — Draft a therapy intake form with AI from de-identified prompts, then run consent, safety and lawyer checks before any client sees it.
• Therapist Termination Letter to Client: Wording and an AI Draft — A therapist termination letter template with wording for reasons, referrals and notice, plus a safe AI drafting step and clinician review.
• AI for Counsellors: Can It Draft a Professional Will and Handover Plan? — Draft a counsellor's practice continuity plan with AI: the sections, a filled-in template, who must approve it, and why no client data goes in.
• Counselling GDPR: Can You Use AI With Client Information? — A counsellor's step-by-step vetting of any AI tool before client information goes in: basis, contract, location, retention and privacy notice.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Mentalyc pricing and security pages; Berries home page; Jane App pricing, AI Scribe feature page and help guides (deletion policy, using the scribe without recording, UK FAQ); Heidi pricing page. Prices and settings as read on 4 November 2026.