AI Notetaker for Coaches: Confidentiality Decides the Audio Path

By Kyle Nelson, Founder, Fazit

ogTitle: "AI Notetaker for Coaches: The ICF Already Decided"

The best AI notetaker for a coach is the one that never creates an audio file. The International Coaching Federation's Ethical Insights and Considerations document requires client consent before any recording is made, and it names artificial-intelligence devices that automatically capture sound as exactly the kind of recording that needs that consent. A tool that transcribes in memory and writes a Markdown note into your own vault sidesteps the recording question entirely, because there is no recording to consent to.

That is not a marketing line. It is an architecture decision, and it maps directly onto what the ICF actually requires of you.

What the ICF says about recording

The ICF Code of Ethics, effective 2025, defines coaching confidentiality as protecting information obtained in or around the coaching relationship unless disclosure is legally required, necessary because of a threat of harm, or authorized by the client in writing. The same code requires coaches to maintain, store, and dispose of electronic files and communications in ways that promote confidentiality, security, and privacy and comply with applicable laws and agreements.

The recording requirement is more specific. According to the ICF's Ethical Insights and Considerations document, client consent must be obtained before audio or video recordings are made, including recordings made automatically by artificial-intelligence devices. The ICF's 2020 guidance already applied this to devices like Alexa and Siri that record sound in a room without a person pressing record. A cloud AI notetaker that joins your coaching call and captures everyone's audio is the same category of device, just wearing a different label.

The ICF also requires coaches to comply with applicable laws concerning personal data and communications. That pulls in state wiretap statutes, GDPR if you coach across borders, and professional-conduct rules that vary by jurisdiction. None of this is legal advice, and consent and professional-conduct duties apply to the person running the call. But the ICF's own text gives you a clear test: if the tool makes a recording, you need consent. If the tool never makes a recording, that specific obligation does not arise.

Bot-free is not the same as recording-free

Several notetakers now advertise bot-free capture. Granola, Krisp, Circleback, and now Fireflies and Fathom all capture meetings without a visible participant joining the call. That solves the client-friction problem: in a documented six-week test of 47 meetings, 32% of clients questioned a visible recording bot, while zero questioned bot-less capture. But bot-free describes what your meeting looks like, not where your audio goes.

Granola's public documentation says it records audio from your microphone and system audio, sends it to a transcription provider, and deletes the audio after transcription. The question for a confidential coaching call is what happens in between. The audio reaches Granola's systems and third-party services before it is deleted. Granola's public security FAQ names OpenAI and Anthropic as subprocessors, uses US-only AWS storage, and retains transcripts indefinitely by default. That is a recording that exists on someone else's infrastructure for at least part of its life.

The distinction matters because the ICF's consent requirement attaches to recordings, not to bots. A tool can be invisible to your client and still create a durable audio artifact that you now have to disclose, secure, and eventually delete. Bot-free is not the same as private, and for a coach the ICF's recording-consent rule is the sharper test.

The audio path is the only column that survives a security review

When you compare AI notetakers, the summaries have converged. Fathom, Otter, Fireflies, tl;dv, and Granola all write a clean note now. The comparison that still means something is architectural: where in the pipeline a durable artifact of your conversation is created, and who holds it afterward. Five of six create one on someone else's infrastructure.

Fazit is the sixth row. A Core Audio process tap reads the call app's output and the microphone directly, so nothing joins the meeting and no other participant sees anything. The audio exists only in a fixed-size RAM ring buffer with no write API, destroyed on every exit path including errors. Transcription runs on-device using Parakeet via CoreML, and note generation runs against a local model on localhost. The output is a plain Markdown file written into a folder of your Obsidian vault. There is no Fazit database holding a copy.

Every note records audio_retained: false in its frontmatter. That is not a promise about deletion. It is a statement that the audio was never eligible to be retained in the first place.

What a coach actually needs from a call note

A coaching call note should capture the client, date, objective, decisions, commitments, owners, due dates, and the follow-up email. The ICF's confidentiality standard applies to information obtained in or around the coaching relationship, which means the note itself is a confidential record you must store and dispose of properly. A Markdown file in your own Obsidian vault is a file you can grep, back up, and delete on your own schedule. A note living in a vendor's database is a row you can request to be deleted, with the deletion procedure and its gaps documented elsewhere.

The follow-up email is the part coaches screenshot. After a session, Fazit writes a paste-ready follow-up email in the client's own language and register, alongside the summary, key moments, and action items with owners. The note appears in your vault, not in a web app you have to export from.

The consent conversation gets simpler

The ICF requires consent before recordings. If your notetaker creates an audio file, you need to tell your client that a recording is being made, where it goes, and who processes it. If your notetaker never creates an audio file, the honest disclosure is shorter: the call is being transcribed on this Mac, the audio is discarded as it is transcribed, and the note lands in my own vault.

That is not a legal guarantee. California's CIPA is an all-party consent statute, and Germany's §201 StGB treats recording non-public speech as a criminal offense. German counsel has stated that a RAM-only pipeline that transcribes in volatile memory and never writes an audio file generally does not meet the elements of §201 StGB, because there is no recording on a sound carrier. But that is counsel's analysis, not a court ruling, and it does not remove your duty to be honest with the person on the other end of the call.

The ICF's own guidance points the same direction. It requires consent before recordings made automatically by AI devices. A tool that never makes a recording removes the recording from the conversation. What remains is ordinary professional transparency about how you keep notes.

The checklist for coaches

Run any AI notetaker through these four questions before you put it on a client call:

  • Does a bot join the call as a participant?
  • Does the tool create an audio file, even temporarily?
  • Does the audio or transcript leave your Mac for transcription or note generation?
  • Does the vendor hold a copy of the note after the call ends?

If the answer to any of the first three is yes, the ICF's recording-consent requirement is in play, and you need to disclose it. If the answer to the fourth is yes, you have a second record to secure under the ICF's storage and disposal standard.

Fazit answers no to all four. The capture is bot-free, the audio never becomes a file, transcription and note generation run on the Mac, and the note is a Markdown file in your vault. The ICF's Ethical Insights and Considerations document tells you what a recording is, and it tells you to get consent before you make one. The tool that never makes one is the one that keeps your confidentiality obligation simple.

Sources