Choose by setup, not brand: record on your own device, or use a tool whose pages promise no training on recordings, with short retention. Fireflies says it never trains on customer data, Fathom offers an opt-out, and Otter's policy reserves training on de-identified audio. Clients must consent in writing to what is captured, where it goes, who sees it and when it is deleted.
No tool is private by design, and a bot-free mode does not remove the need for consent; it only hides the extra participant. Your coaching agreement is where confidentiality is promised, so the tool has to fit that promise. The ICF Code of Ethics points the same way: its confidentiality standard asks for the strictest level with all parties, and another standard extends the duty to any technology a coach uses, artificial intelligence included.
Three ways to capture a session, and what each puts in the cloud
The first decision is how the audio is captured, because that settles what leaves your computer. Which brand you pick comes second.
| Method | What leaves your device | Effort |
|---|---|---|
| Local capture, local transcription: record the call on your own computer and run a speech-recognition model on that machine | Nothing | Highest: you write the notes or summarise by hand |
| Local capture, cloud transcription: upload the saved file to a vendor | The whole audio file | Medium; Otter's Pro plan lists 10 imported files a month and its free plan 3 for the lifetime of the account |
| Cloud capture: a notetaker joins, a desktop app records, or the platform summarises | Audio or transcript, processed and stored by the vendor | Lowest; fastest to a usable summary |
For the first route, OpenAI's Whisper is a speech-recognition model released under the MIT licence. You install it on your own computer with a command-line tool and run it on audio files, and its repository lists memory needs from about 1 GB to about 10 GB of video memory depending on model size. The catch comes next. Pasting the transcript into a cloud chatbot to summarise it sends the content out again, so a fully local route means summarising yourself or using a model that also runs on your machine. Whatever you choose, consent is still needed to record, and all three methods need it for the same reason.
Where each tool stands on training, retention and visibility
The table uses each vendor's own pages, read in November 2026, and quotes only short phrases. Wording changes, so reread the source before you rely on it. Prices are USD list prices per user.
| Tool | Training, in the vendor's words | Retention controls | Capture without a joining bot | Price |
|---|---|---|---|---|
| Otter | Its privacy policy lists "training our proprietary AI technology on de-identified audio recordings" as a use, names data-labelling providers among recipients, and describes no AI-training opt-out | A "custom data retention" row sits on the pricing page without showing which plans have it; data kept "as long as necessary", with deletion on request | Yes: desktop recording "without bots joining the call", plus mobile capture | Pro $8.33 a month billed annually ($16.99 monthly); Business $19.99 annual ($30 monthly) |
| Fireflies | "Your data is never used for AI training or any purpose outside your direct business needs"; its policy says vendors are contractually barred from training on it | Custom retention is an Enterprise feature; you can delete data at any time | Yes: a Chrome extension records "without Fireflies bot", alongside desktop and mobile apps | Pro $10 a seat annual ($18 monthly); Business $19 ($29) |
| Fathom | Its policy says it "may use and create de-identified data" from meeting content, offers an opt-out "in your account settings", and bars third parties from training on it | Custom data retention is an Enterprise feature; on team accounts, deletion requests go through the account's administrator | Yes on the free plan: bot-free capture, still in beta on Mac | Free; Premium $16 annual ($20 monthly); Team $15 ($19); Business $25 ($34); Team and Business need two users |
| Zoom meeting summary | Its privacy statement says it does not use audio, video, chat or other communications-like content to train its own or third-party AI models | The pages I read give no storage location or period for summaries | Built into the meeting, so no extra participant joins | Basic: summaries for 3 hosted meetings a month; Pro $14.16 a month billed annually, unlimited summaries |
Read the table as three tiers. Fireflies and Zoom make flat no-training statements. Fathom allows de-identified use unless you opt out in settings. Otter's policy lists training on de-identified recordings as a purpose, which you can read for yourself in its privacy policy. "De-identified" is the vendor's term, and a conversation about a restructuring or an acquisition can point to one company by its content alone, so the label settles less than it appears to for an executive coach. Fathom's opt-out is described in its privacy policy too.
Two cautions on the Fireflies row. Its privacy policy says it does not cover content processed for business customers, which falls under its terms of service and data processing agreement, so read those for your plan. And retention controls that matter sit on the top plans: custom retention is an Enterprise feature at Fireflies and Fathom, and Otter's page does not show which tier has it. Fireflies' Enterprise plan, at $39 a seat billed annually, also lists a "Transcript + Summary only mode", worth asking about if you never want audio stored. A solo coach on a standard plan is therefore the retention policy, and has to delete on a schedule (a workable one appears below).
If your clients host on Teams, Microsoft states that prompts, responses and data accessed through Microsoft Graph are not used to train its foundation models. Whether you can use the meeting recap depends on your licence and the host's settings, so check both. For Google Meet, check which Workspace plans include the notes feature and where it saves them, because both depend on your plan. For the general case, the tutorial on whether AI meeting note-takers are safe for client calls covers the checks that apply to any profession.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA 25-sessions-a-month coach, priced and rated
Price it for a solo coach who runs 25 one-hour sessions a month, which is 1,500 minutes of audio (an invented but typical load). Annual costs at list prices:
- Otter Pro: 12 × $8.33 is about $100 a year, but the plan lists 1,200 meeting minutes a month per user and 90 minutes per conversation, so 1,500 minutes does not fit. Business at 12 × $19.99, about $240 a year, lists unlimited meeting minutes.
- Fireflies Pro: 12 × $10 is $120 a year. Pro lists unlimited transcription but 8,000 minutes of storage per seat, so at 1,500 minutes a month the account fills in about five months unless you delete on a schedule, which you should do anyway.
- Fathom: the free plan lists unlimited recordings and transcriptions at $0; Premium at 12 × $16 is $192 a year.
- Zoom: the Basic plan's summaries stop at 3 hosted meetings a month, so 25 sessions needs Pro, which lists unlimited meeting summaries at 12 × $14.16, about $170 a year. ZoomMate, Zoom's separate $20-a-month agent product, is not needed for this.
- Local route: $0 in software, plus your time.
Now the sum that matters. Assume writing up a session takes 20 minutes without a tool and about 8 minutes checking and correcting an AI summary with one. The saving is 12 minutes × 25 sessions = 300 minutes, which is 5 hours a month. At $120 an hour of your time (use your own rate), that is $600 a month, against $10 to $20 for the subscription. Price therefore does not decide the choice. What you give up in confidentiality is the real cost, so rate the options on that.
| Kind of session | Capture method | Reason |
|---|---|---|
| Board-level, restructuring, grievance, health or legal matters | Local capture and local transcription, or handwritten notes only | Content could harm the client or others if it leaked, and no vendor promise removes that |
| Typical development coaching | Cloud tool with a no-training statement, short retention, written consent | Good time saving at a risk the client has agreed to |
| Group workshops and open webinars | Any of the tools, with attendees told in the invitation | Content is not private to begin with |
| Clients who decline or hesitate | Handwritten notes, then dictate your own summary | Trust outweighs convenience |
Settings to change before the first session
Every tool ships with defaults tuned for sales teams, not for coaches. Before a client joins, work through these, then test with a five-minute fake conversation and look at everything the tool produces and sends.
- Stop automatic joining. Otter's home page describes scheduling its agent to join all your Zoom, Teams and Meet meetings automatically, and any notetaker connected to your calendar can behave the same way. Disconnect the calendar, or restrict recording to meetings you choose by hand.
- Set the training option. On Fathom, switch the opt-out on in account settings. Otter's policy describes no AI-training opt-out, so ask support in writing or keep confidential sessions out of it. On Fireflies, read the data processing agreement for your plan as well as the policy.
- Check who receives the notes. Look for sharing defaults for attendees, calendar invitees and workspace teammates, and set the visibility to you alone.
- Switch off pushes to other apps. Summaries sent on to a CRM, a chat channel or email create copies that your retention schedule cannot reach.
- Turn on two-factor sign-in wherever the tool offers it. A coach's account holds a client list's worth of private conversations.
- Ask the client about their platform. Some organisations block third-party notetakers on calls they host. If the client hosts, find out before you rely on a tool that connects to their meeting.
The tutorial on adding Otter to Microsoft Teams and controlling who can invite it goes into the invitation side for one platform.
The client consent clause, in words you can adapt
Put this in the coaching agreement, or send it as a one-page addendum before the first session, and keep the signed copy. It is an illustration, not legal advice: have a lawyer who knows your market check it, because recording and data rules differ by place. For a general-purpose consent form that is not specific to coaching, the tutorial on AI note-taking consent wording for client meetings has more variants.
Recording and AI notes
1. With your agreement, I record our sessions and use [tool name]
to produce a transcript and a summary. This is for my own
preparation and follow-up only.
2. What is captured: audio, a transcript, and an AI-written
summary. None of it is shared with your organisation or any
third party unless you ask me to in writing.
3. The tool is run by [vendor], which stores the recording and
transcript on its servers in [location, as confirmed by the
vendor]. [Vendor] states that it [does not use your content to
train AI models / offers an opt-out, which I have switched on].
I will tell you if that changes.
4. Access: only I can open the recordings. [Or: my assistant can
open them for scheduling only.]
5. Retention: I delete the audio within [7] days of each session,
the transcript within [30] days, and the summary within [7]
days, after moving key points to my own notes. You can ask me
to delete anything sooner, and I will confirm in writing.
6. Your choices: you can ask me to pause or stop recording at any
time, mark a passage as off the record, or decline recording
altogether. Declining changes nothing about our work.
7. Limits: AI summaries can be wrong, and I check each one against
my own notes. Any duty to disclose is as set out in our
coaching agreement.
Name: ________________ Date: ____________Three clauses carry the weight. Clause 3 forces you to read the vendor's current statement and name it, so the promise to the client is as accurate as your source. Clause 5 turns a vague "I delete things" into dates you can be held to. Clause 6 protects the relationship, because a client who can say "off the record" mid-session will say more than one who cannot.
What to say at the start, in the middle and when a client declines
The consent form is not a substitute for saying it aloud. At the start of each recorded session: "Before we begin, I'm recording this one for my own notes, as we agreed. If you say 'off the record' at any point, I'll pause it. Shall I start?" Waiting for the yes matters; a recording started over a silence is not consent.
In the middle, when a client signals they are about to say something sensitive, pause the recording yourself and say so: "Pausing now." Resume with: "Recording again." Note the time in your own notes so the transcript has a visible gap that matches. When a client declines altogether, thank them and carry on: "That's fine. I'll take handwritten notes and send you the actions after." Then do exactly that. Do not mention the tool again unless they raise it.
The same principle covers the sponsor. In a three-way contract, the organisation paying often expects progress reports. Say in the agreement what the sponsor receives, usually movement against agreed goals, and say that session recordings, transcripts and AI summaries are not part of it. If the sponsor asks for them later, the answer is already written down.
An AI summary and what you keep from it
An illustrative before and after shows why retention matters. An AI summary of a session reads: "[First name] discussed concerns about the planned restructuring of the operations team, including likely redundancies in the second quarter, and a difficult relationship with the finance director. Action items: prepare talking points for the board meeting; consider raising the finance director issue with the chair."
That paragraph contains a company's unannounced plans, a named colleague and a hint of a dispute. The note you keep in your own system is shorter and about the person: "Theme: managing own response to upcoming operational change. Commitment: draft board talking points by Friday. Next: explore leadership approach to peer conflict." The summary is deleted within the 7 days in the clause, and the coaching record holds only the second version. This is also where the ICF code's standard on keeping records securely, and on destroying them in a way that protects confidentiality, becomes concrete.
A retention schedule that fits a solo practice
| Item | Keep for | Before you delete |
|---|---|---|
| Audio | 7 days, or until the transcript is checked | Confirm the transcript matches what was said on any passage you will quote |
| Transcript | 30 days | Copy the lines you need into your own notes |
| AI summary | 7 days | Write your own note first; do not file the summary itself |
| Your own notes | As your agreement and any professional or legal rules require | Check the rule with your adviser; I would not guess |
On a plan with no retention controls, the schedule is a Friday routine of about ten minutes. Open the tool's list sorted by date, delete audio older than 7 days and transcripts older than 30, empty any trash or recycle area, and write the date in a log. Ten minutes a week is cheaper than a breach conversation with a client. If you want more on turning notes into follow-ups once the recording question is settled, the tutorial on AI session notes and follow-ups for coaches sets out a weekly workflow, and the sibling piece on AI for life coaches and the ICF code covers between-session support.
The notetaker that joined the wrong call
Failures here are usually quiet, as in this invented example. A coach connected a calendar to a notetaker so that sessions were captured without clicking anything. Months later a client asked for an informal chat about a colleague's performance, put on the same calendar as a normal session. The notetaker joined, the transcript went into the cloud account, and nobody noticed for three weeks. Nothing leaked, but the coach had recorded a conversation the client expected to be unrecorded, and had to explain that.
The fix had four parts: delete the recording and transcript, tell the client what happened and what was deleted, disconnect the calendar and switch to starting each recording by hand, and add a line to the consent clause saying recording happens only when announced at the start of the session. The cost of the shortcut was a few seconds saved per session. Convenience settings are where confidentiality usually leaks, which is why the settings list above begins with automatic joining.
Running the local route end to end
For the sessions in the top row of the tier table, here is a workable local routine. Record the call on your own computer if your platform lets the host save a recording locally; otherwise use a separate recorder on the desk and tell the client. After the session, run the speech-recognition model on the file while you do something else, then read the transcript with the audio open for any passage you may quote. Write your own note from it, and delete the audio and transcript on the retention schedule. Nothing in that chain reaches a vendor's servers.
Two practical limits. A local model's transcript is only as good as the audio, so test it with a client-style call from a quiet room before you trust it, paying attention to names and figures. And the routine costs you time; for most coaches the sensible compromise is to use it for the handful of sessions where the content is highly sensitive and a cloud tool with a no-training statement for the rest.
Six questions for a vendor's support team
Send these in writing and keep the reply with the client's consent form. Support teams often answer the question they expected, so ask each one plainly.
- Is audio or transcript content ever used to train or evaluate models, yours or anyone else's, and is that on or off by default for my plan?
- Which outside AI providers and data-labelling services can receive my content, and under what restrictions?
- Where is my data stored, and can I choose the location?
- When I delete a recording, transcript or summary, how long until it is gone from backups, and will you confirm deletion in writing?
- Who can see my content by default, including support staff, and how is that access logged?
- What happens to my data if I cancel, and can I export everything first?
A vendor that answers all six clearly is rarer than the marketing suggests. If the replies differ from the public pages, trust the written reply and copy its wording into clause 3 of the consent form.
Questions coaches ask about recording sessions with AI
Is bot-free recording more private than a visible notetaker?
No. Otter, Fathom and Fireflies all describe ways to capture without a bot joining the call, but the audio still goes to the vendor's servers and the client still has to know. Bot-free capture only removes the extra participant on screen. If the point is confidentiality, the questions that matter are where the recording goes, who can open it, how long it stays, and whether it feeds model training.
Do I need consent if I only use the notes myself?
Yes, because the recording is made and stored whether or not anyone else reads the summary. Tell the client before the first session what is captured, which tool processes it and when it is deleted, and record their agreement in writing. Some places also require every party to a call to consent to recording, so ask a legal adviser which rules apply to you.
What if a client says no to recording?
Take handwritten notes or dictate your own summary afterwards, and say plainly that nothing about the relationship changes. Do not make the tool a condition of coaching. A coach who can work either way is easier to trust than one who needs the recording, and the refusal itself is useful information about how much the client wants to say.
Should the sponsor's organisation see the transcripts?
Not unless the coaching agreement says so and the client has agreed to it. In a three-way contract, define what the sponsor receives, usually progress against goals rather than session content, and state that transcripts and summaries are not part of it. Keep the sponsor's reporting separate from anything the notetaker produces.
Further reads
- AI Meeting Note-Takers Compared for Small Teams — A broader comparison of note takers for small teams
- ChatGPT Prompts for Executive Coaching: Prep, Reflection and Goals — Prompts for session prep and goal reflection
- AI Tools for a Coaching Business: Notes, Programmes and Follow-Up — The wider toolset around notes and follow-up
- Can Lawyers Use AI Note Takers in Client Meetings? — How another confidential profession handles note takers
- How Much Does an AI Note-Taker Cost per Seat? — Per-seat costs across the main note takers
- Life Coaching Rates During Automation: How to Reprice Packages — How to reprice life coaching packages once AI handles prep and admin: hours-based versus milestone-based structures, with a worked example.
- AI 360 Feedback for Coaches: Collect, Theme and Report — A coach's 360 process from question set to debrief: anonymity thresholds, AI theme coding with a human check, a report clients trust, and what tools cost.
- AI Prompts for Career Coaching: A CV and LinkedIn Review Workflow — A career coach's review workflow: consent, intake form, six prompts with a before-and-after example, and the judgements that stay with the coach.
- AI for Mediators: Summaries and Agreements That Stay Neutral — A neutrality checklist for AI session notes, a balanced summary prompt, agreement drafting rules and a consent clause for every party.
- How to Use AI in a Small Consultancy: Proposals, Research and Delivery — Stage-by-stage AI use for a three-person consultancy: proposals, research and delivery, with the confidentiality rules that decide which tools you can use.
- AI Coaching for Leaders: What to Offer Clients Alongside Your Sessions — Design a bounded AI coaching add-on for managers: rehearsal prompts, privacy terms, human check-ins and a pilot scorecard.
- Best AI Model for Life Coaching: What Session Notes and Privacy Need — How a life coach should choose between ChatGPT, Claude and Gemini for session prep and follow-ups, with privacy settings, ICF Code limits and a test routine.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Otter, Fireflies, Fathom and Zoom pricing and privacy pages; Otter home page; Microsoft Learn Copilot privacy page; ICF Code of Ethics; OpenAI Whisper repository (all checked November 2026).