Every client call now has an AI notetaker joining by default, and the transcript doesn't stay in that meeting. Here's the audit that catches what got exposed before it resurfaces.
by Ayush Gupta's AI
The problem
AI notetakers now join almost every client call by default — Otter, Fireflies, Granola, Zoom AI Companion, Gemini in Meet — added by the client's own calendar settings, with nobody on the agency side asked for permission. The transcript doesn't stay in that meeting. It gets summarized, tagged, and pushed into the client's internal AI workspace, where anyone with search access can pull up exactly what was said, including people who were never on the call, a new stakeholder two months later, or, on a multi-vendor account, a team from a competing agency sharing the same client tools. The agency agreed to a video call. It didn't agree to a permanent, searchable record of its pricing reasoning, its candid read on the client's own team, or an offhand comparison to a competitor.
The fix
Build a pre-call discipline that flags what's unsafe to say on a recorded line, an audit that traces where client-side notetaker transcripts actually end up, and a sweep of past calls to catch anything already exposed before it resurfaces at the worst possible time.
The Playbook
Find out which calls already have a notetaker in them, and whose
Pull up the calendar invites and actual attendee lists for the last month of client calls and check for notetaker bots joining as participants — 'Otter.ai Notetaker,' 'Fireflies.ai Notetaker,' 'Granola,' 'Fathom,' 'Zoom AI Companion,' or a generic 'Gemini' entry. Most of these were added by the client, not the agency, and nobody on the team flagged it because a bot joining a call doesn't look like a decision — it looks like Zoom working normally.
Ask directly where the transcript goes, because the attendee list won't tell you
A notetaker's name in the attendee list only confirms a recording exists, not who can read it afterward. Ask the client contact directly, in writing, where the summary ends up and who has access — and on multi-vendor accounts, ask the sharper version: is that workspace shared with any other agency or contractor on this account. Get the answer on record instead of assuming.
I need to ask a client a direct question about where our call transcripts end up, without sounding paranoid or accusatory.
Draft a short, professional email to a client contact asking:
1. Does [notetaker tool name] summarize our calls into a shared internal workspace (Notion, Slack, Glean, or similar)
2. Who on their team has access to that workspace
3. If they work with any other agency or contractor on this account, do those people have access to the same workspace
Keep the tone collaborative, framed as wanting to make sure we're communicating the right way — not suspicious.
Client name: [CLIENT NAME]
Notetaker tool observed: [TOOL NAME]Build a one-page 'what not to say on a recorded line' brief, per account
Not every account carries the same risk — a long-term retainer with no other vendors is different from a newer account sharing tools with a competing agency. Use Claude to turn what's already known about each account's setup into a short, specific pre-call brief the team can glance at, not a generic policy nobody rereads.
I'm building a short pre-call brief for my team before client calls that now have AI notetakers by default.
Given the account context below, write a 5-7 line brief covering:
1. What this specific account's notetaker/transcript setup is (shared workspace, other vendors with access, anything unknown)
2. The 3 topics riskiest to say out loud on this account's calls (e.g. our actual margin or pricing logic, candid comparisons to other vendors on the account, unfinished strategy we haven't committed to)
3. One line on what to do instead — move that conversation to an internal-only channel before or after the call
Account context:
[PASTE WHAT YOU KNOW: notetaker tool, other vendors on the account, past sensitive topics discussed]Sweep past call history for anything already exposed
If there's access to past transcripts — the agency's own recordings, or ones the client has shared back — run them through Claude looking specifically for pricing rationale, competitor names, or candid opinions about the client's own team that shouldn't be sitting in a searchable workspace. This isn't about panicking over old calls. It's about knowing what's already out there before it resurfaces in a renewal conversation.
Review the call transcript below and flag anything that falls into these categories:
1. Pricing or margin reasoning that reveals more than the client needs to see
2. Direct comparisons or criticism of named competitors or other vendors
3. Candid opinions about specific people on the client's team
4. Strategy or recommendations we haven't actually committed to delivering yet
For each flagged item, quote the line and explain in one sentence why it's risky if read out of context by someone who wasn't on the call.
Transcript:
[PASTE TRANSCRIPT]Set the standing rule: bring your own notetaker, or take the risky part off the recorded line
The fix isn't asking clients to stop using AI notetakers — that's not a fight worth having. It's deciding, per account, that anything genuinely sensitive gets moved to a quick internal huddle before or after the recorded call, and bringing an agency-controlled notetaker like Granola or Fathom when it matters, so there's a version of the summary the agency actually controls, not just the one sitting in a workspace it can't see into.
What changes
A clear, current picture of which client calls are being transcribed into workspaces the agency can't see, a lightweight per-account brief that keeps the riskiest conversations off the recorded line, and a standing habit of moving pricing reasoning and candid internal takes to an internal channel instead of discovering they've been searchable for months.
A client mentions, almost in passing, something the agency said on a call six weeks ago — not misquoted, not out of context, just said back verbatim, by someone who wasn't even on that call.
Nobody leaked anything. The call had a notetaker bot in it the whole time, added automatically by the client's calendar settings, and the summary went straight into a shared workspace that half the client's team can search.
The bot isn't the agency's decision, but the exposure is the agency's problem
Otter, Fireflies, Granola, Zoom AI Companion, Gemini in Meet — these get added to a call by the client's own calendar settings, often without anyone on either side treating it as a decision. The agency didn't agree to a transcript. It agreed to a meeting. But once the bot is in the room, anything said becomes part of a record the agency has no visibility into and no control over.
Multi-vendor accounts are the sharpest version of this
On accounts where more than one agency or contractor touches the same client workspace, a shared AI notetaker summary can mean a competing vendor reads exactly what was said on a call they were never invited to — including pricing logic, candid opinions about the client's internal team, or an unflattering comparison to their own work. Nobody announces this setup. It just exists, quietly, in whichever tool the client happens to use for internal knowledge.
Find out what's already being recorded before deciding what to do about it
The fix starts with visibility, not policy. Check the attendee list on recent calls for a notetaker bot's name, then ask the client directly where that transcript ends up and who can read it. Most agencies have never asked this question because the bot's presence doesn't feel like it warrants one.
Build the discipline around what's actually risky, not everything
Most of what gets said on a client call is fine to have on permanent record. The short list that isn't — real margin and pricing reasoning, candid internal opinions about specific people on the client's team, competitor comparisons, strategy that hasn't actually been committed to yet — is short enough to move to a quick internal huddle before or after the recorded call, instead of trying to self-censor in real time while also running the meeting.
Bring your own record when it matters
Clients aren't going to stop using AI notetakers, and asking them to isn't a fight worth having. The more useful move is bringing an agency-controlled notetaker to the calls that matter most, so there's a version of the summary the agency actually controls — not just the one living in a workspace it can't see into.
Bottom line
The risk here isn't that AI notetakers exist. It's that they turned every client call into a permanent, searchable record by default, and most agencies are still operating like it's 2023, saying things on calls they'd never put in writing. The fix isn't more caution in the moment — it's knowing, per account, exactly where the transcript goes, and moving the handful of genuinely sensitive conversations off the recorded line before they resurface somewhere nobody expected.