Use AI for Session Notes Without Breaking Client Confidentiality

You'd love to stop writing session notes at night. Then the question hits: is it OK to put client info in AI? Most advice says "get consent and redact" and leaves you to figure out what that means. This workflow gives you the exact consent sentences, the settings to change in each AI tool, a list of what to strip before you paste, and a notes prompt that won't summarize the things it shouldn't. It's the safety layer under the session recap workflow.

Time check: about 45 minutes once to write your consent wording, change your settings, and make your strip list. After that, about 2 minutes per session to strip details before pasting. On its own it doesn't save time. It's what makes the recap workflow safe to use, and that one is where the 30 to 45 minutes a session comes back. These are estimates, not measured results.

Who it's for

Coaches, consultants, and service providers who talk with clients about personal or work situations and want AI to help with notes and recaps. It covers one-to-one, group, and sponsor-paid (company-paid) coaching. If you're a therapist or other licensed clinician, read the short answer above before anything else.

What you need

• Any AI chat tool: ChatGPT, Claude, Gemini, or Grok. You'll change one or two settings in step 3.
• Your coaching agreement or intake form: Wherever clients sign or tick a box before you start. You'll add one clause.
• A doc with find and replace: Google Docs, Word, Pages, or a notes app. This is where you strip details, before anything touches AI.
• A strip key, kept outside AI: A private note that maps real names to stand-ins (for example "Client A", "Manager M"). It never gets pasted anywhere.
• Optional: a recorder that makes transcripts: Zoom, Google Meet, Fathom, Otter, or your phone. Only if your consent covers recording.

The steps

1. Pick your level before you write any consent wording

   Consent only means something if you can say exactly what happens. So decide first which of these three levels you're using. You can use different levels for different clients.

   • Level 1, your notes only. No recording. You jot rough notes during or right after the session, strip them, and paste those into AI. The AI never hears the session. Lowest risk, and still saves most of the writing time.
   • Level 2, transcript, stripped. You record with a tool that makes a transcript, strip it, then paste. More detail, more stripping work.
   • Level 3, AI notetaker in the session. A notetaker (Zoom AI Companion, Fathom, Otter, Fireflies) joins the call and its vendor processes the audio. Most convenient, and the client's words go to that vendor before you can strip anything.

   If you're unsure, start at Level 1. Plenty of coaches never need more.

   Done when: you've written down which level you use, and for which clients.

2. Copy the consent wording: once in writing, once out loud, once a session

   Clients don't object to AI as often as coaches fear. They object to finding out later. Ask yourself the client's version of the question: if they later found out this was automated, would it feel helpful or disappointing? Saying it upfront is what makes it feel helpful. Use all three versions below. Swap the bracketed parts for your real tool and level.

   Copy this prompt

   Agreement clause (paste into your coaching agreement or intake form)

   CopyAI-assisted notes I may use [tool name, e.g. ChatGPT] to help me write session notes and recaps. Here is exactly what that means: - What goes in: [Level 1: my own written notes from our session] / [Level 2: a transcript of our session] / [Level 3: the audio of our session, processed by (notetaker name)]. - Before anything goes into the AI tool, I remove your name and details that could identify you or anyone you mention. - I have turned off the setting that lets the AI company train on my chats, and I delete the chats after [7] days. - Your notes are stored in [where, e.g. my private Google Drive] and only I can see them. If your company pays for coaching, they never see them. - You can say no now, or stop at any time, just by telling me. Nothing about our work changes; I'll write notes by hand.
   Copy this prompt

   Say this at the start of the first session

   CopyBefore we start: as we covered in the agreement, I use an AI tool to help with my notes so I can listen properly instead of typing. I take your name and identifying details out first, and only I see the notes. Are you still comfortable with that? You can change your mind anytime.
   Copy this prompt

   Say this at the start of every recorded session (Level 2 or 3)

   CopyQuick check before I hit record: still OK to record for my notes? If anything comes up today you'd rather keep out of the notes, just say "off the record" and I'll pause.

   Done when: the clause is in your agreement, and you've said the first-session version out loud at least once (to a mirror counts).

3. Change your AI tool's privacy settings (5 minutes)

   Do this once, on every device you use. Check the setting again every few months, because names and defaults change.

   • ChatGPT (Free, Plus, Pro): Settings > Data controls > turn off "Improve the model for everyone". For the most private option, start a Temporary Chat (the Temporary button at the top of a new chat). Temporary chats aren't used for training and don't stay in your history, though OpenAI says it may keep them up to 30 days for safety. Don't press thumbs up or down on a chat with client details: feedback can send the whole conversation for training.
   • Claude (Free, Pro, Max): Settings > Privacy > turn off "Help improve Claude". With it off, Anthropic says chats are kept up to 30 days. Incognito chats are never used for training. Reopening an old chat counts as "resuming" it under your current setting.
   • Gemini: Settings & help > Activity > turn off Keep Activity (it used to be called Gemini Apps Activity). Or use a Temporary Chat. Google says chats with Keep Activity off are still kept up to 72 hours.
   • Memory: in any tool, turn off memory for client work, or check what it has saved and delete client details. You don't want a client's situation showing up in next week's content draft.
   • Team or business plans (ChatGPT Business or Enterprise, Claude Team, Google Workspace with Gemini) generally don't train on your data by default. Check your plan's terms rather than assuming.

   Done when: training is off (or you're using a temporary or incognito chat), memory is off or clean, and you've noted the date you checked.

4. Strip details before you paste, using your strip key

   Strip in your own doc, not in the AI. Asking AI to remove names means you've already pasted the names. Open your rough notes or transcript in a doc, then use find and replace with your strip key.

   Replace:

   • The client's name, and everyone they mention: partner, kids, boss, coworkers ("Client A", "Partner P", "Manager M").
   • Company, school, and team names ("the school", "Company X").
   • Places smaller than a big city, plus addresses, dates of birth, and phone numbers.
   • Unusual details that point to one person: a rare job title, a news story they were in, a very specific number.

   Never paste, even stripped. Keep these in your own notes only:

   • Health, mental health, medication, or diagnosis details
   • Anything about abuse, safety, self-harm, or risk to anyone
   • Legal matters, immigration status, or anything they asked you to keep private
   • Money specifics tied to them (account numbers, exact debts, salary offers)

   Then run the check prompt on the stripped text. It catches the details you missed.

   Copy this prompt

   Prompt: Leftover-identifier check (run on stripped text only)

   CopyThis text has already had names replaced with stand-ins like "Client A". Don't summarize it. Only list anything still in it that could identify a real person or organization: names, places, job titles, dates, numbers, or unusual details. Quote each one exactly and say why it could identify someone. If you find nothing, say "Nothing found." TEXT: [paste stripped notes or transcript]

   Done when: the check prompt says "Nothing found", or you've replaced everything it listed.

*Time saved is an estimate based on typical manual effort for this task, not a measured result. Your mileage will vary. Always review AI output before sending or publishing.

Next: Are Prompt Packs Worth It? Turn the Few You'd Use Into Workflows →