Yes, with four conditions: the client agrees at the start of each meeting (recording-consent laws differ, and some require everyone's consent), the tool is on a business contract that doesn't train on your data, recordings and transcripts sit in the matter file with a set retention period, and a lawyer checks the summary before it becomes the attendance note.
The bigger risk isn't the software; it's the record. An AI summary that turns "I'll think about it" into "client instructed us to accept the offer" becomes, once filed, evidence of instructions that were never given. Add to that bot-based note takers that join every meeting in a calendar on their own, and a meeting with the other side can end up recorded by accident. Both problems are fixed by settings and habits, not by avoiding the tools.
Meetings where a note taker fits, and where it doesn't
| Meeting | AI note taker? | Why |
|---|---|---|
| Internal file review or supervision | Yes | Low risk; saves writing up actions |
| Initial fact-finding with a new client | Yes, with consent | Lots of detail, and the transcript helps the lawyer listen instead of writing |
| Taking instructions on a decision (settle, accept, sign) | Yes, but the lawyer writes the instructions line personally | This is where summaries overstate agreement |
| Meeting with a vulnerable or distressed client | Usually no, or only if they clearly want it | Recording can inhibit them; consent may not be fully free |
| Without-prejudice or settlement discussion with the other side | No, unless all parties expressly agree | Their consent, the confidentiality of the discussion and who receives the output all become issues |
| Hearings, tribunals, mediations | No, unless the court or mediator permits it | Recording is often restricted or prohibited without permission |
Take a campsite owner's first meeting about a dispute with a contractor. The lawyer uses the note taker and gets a full record of dates, names and amounts that would have taken twenty minutes to write up. A week later, when the owner decides whether to accept a settlement figure, the lawyer keeps the note taker running for the detail but types the instruction into the file herself, in the client's words, and reads it back to him before the call ends.
Built-in meeting features versus bots that join the call
There are two kinds of tool, and they carry different risks.
Built into the meeting platform. Zoom's paid plans include a meeting summary feature. Microsoft Teams offers AI recaps and Copilot in meetings, depending on the firm's licences. Google Meet's "Take notes for me" is available on Workspace Business Standard and above. These run inside a platform the firm already has a contract with, show participants that notes or recording are on, and keep the output in the firm's own account. Google added a control in 2026 that lets Workspace admins require every participant to actively agree before notes, recording or transcription start; it rolled out from May and is off by default, so an admin has to switch it on. Settings for the other two are covered in Zoom's meeting summaries and privacy settings and Copilot in Teams meetings.
Third-party bots. Tools such as Otter join the meeting as a separate participant and work across platforms. Otter's Business plan is $19.99 a user a month billed annually ($30 monthly). The risks are the defaults: bots can connect to a calendar and join meetings automatically, and some can share notes with all attendees. Otter's own settings let you limit its note taker to meetings you pick and switch off auto-sharing. Otter is also defending a consolidated privacy class action, filed from August 2025, alleging that its note taker recorded meeting participants who hadn't consented and used recordings to improve its models. Those are allegations, not findings, but they show exactly the kind of exposure a firm wants to avoid.
For a small firm, the platform's built-in feature is usually the safer starting point, and often the cheaper one. A rough sum for a four-lawyer firm already on Google Workspace Business Starter (illustrative; confirm prices on the vendors' pages): moving to Business Standard to get "Take notes for me" costs about $7 more per user a month, so about $28 a month for the firm. Four Otter Business seats billed annually come to about $80 a month, on top of the Workspace bill. The time side is similar either way. If each lawyer has eight client meetings a week and a hand-written attendance note takes about 20 minutes, while checking and correcting an AI draft takes about 8, that is roughly 96 minutes a week back per lawyer. The saving only holds if the checking is real; a lawyer who skims a draft in two minutes has saved more time and bought a worse file.
A comparison of the main options is in AI meeting note-takers compared for small teams.
Consent wording at the start of a call
Ask every time, at the start, even with clients who agreed last time. The engagement letter tells clients the firm may use the tools; the meeting is where they agree to it for this conversation. Words that work:
"Before we start, I'd like to use an AI note taker so I can concentrate on what you're saying. It's for our file only. It won't be shared with anyone outside the firm, and I'll check the notes myself. Are you happy with that? You can ask me to pause it at any point."
If the client says no, switch it off and take notes by hand, and record the preference on the matter so nobody asks again at every meeting. If a client asks to pause it for part of the conversation (often when money, health or family comes up), pause it and note in the file that the notes have a gap and why. A filled-in example of both entries (illustrative):
Matter note, 14 Oct: AI note taker used with client's consent,
asked at 10:02 and agreed verbally. Paused 10:41-10:53 at client's
request while discussing her separation; that part is recorded in
my handwritten note below, not in the transcript. Resumed with
consent at 10:53.
Client preferences: happy with AI notes for routine updates; ask
again before any meeting about the family finances.The second entry matters more than it looks. When a colleague covers the matter later, they know what the client agreed to without having to guess or ask again at an awkward moment.
Two details trip firms up. First, recording laws aren't uniform: some jurisdictions allow a call to be recorded with one participant's consent, others require everyone's, and when participants join from different places the stricter rule is the safe one to follow. Asking everyone, every time, avoids having to work out which applies. Second, the person you asked may not be the only person listening. A client joining from home or a car may have a spouse, a colleague or a child in the room. If someone else speaks, ask whether they're happy to be recorded too, or pause the note taker while they're involved.
A second person on one microphone also confuses the transcript. Speaker labels follow voices, and two people sharing one laptop often end up under a single label. In an illustrative family-business meeting, the transcript showed the client saying "we'll sell the van and put the money into the business", when it was his wife who said it and he had disagreed a minute later. Filed as it stood, the note would have recorded an instruction he never gave. The habit that catches this: whenever a second voice joins, say so aloud ("I can hear your wife has joined us"), then check every speaker label for that stretch when you correct the draft.
Clients sometimes ask for a copy of the transcript. The safer offer is the checked attendance note or a summary letter, because a raw transcript contains the lawyer's thinking aloud, half-formed options and transcription errors that can be misread later. If a client insists, send it clearly marked as an unchecked machine transcript.
A line for the engagement letter to go with it:
We may use AI transcription or note-taking tools in meetings to help prepare accurate file notes. We will always ask before using them, you may refuse or ask us to stop at any time, and any notes produced are checked by a lawyer before being placed on your file.
From transcript to an attendance note you'd stand behind
A raw AI summary isn't an attendance note. It's a first draft of one. Ask for a format that separates what was said from what was decided, and forces the model to admit uncertainty:
From the transcript below, draft an attendance note with sections:
1. Attendees and date
2. Facts the client told us (bullet points, client's words where possible)
3. Advice given by the lawyer (only what the lawyer actually said)
4. Client's instructions (ONLY explicit instructions; if the client
did not clearly instruct us, write "No instructions given")
5. Actions and who owns them
Mark anything unclear as [CHECK]. Do not infer decisions.
[transcript]A sample of what comes back from a meeting with a tour operator's director about a supplier claim (illustrative):
4. Client's instructions
- Client instructs us to issue proceedings against the coach company
if no response by 30 September.
- Client agreed to accept 6,500 in settlement. [CHECK]
5. Actions
- Lawyer to send formal letter of claim by Friday.
- Client to send booking records for the June tours.The lawyer's corrections: the director said he'd "probably want to issue" but wanted to talk to his business partner first, so the first instruction becomes "No instructions given; client to confirm after speaking to co-director". The 6,500 line was the model reading the other side's offer as the client's acceptance; it is deleted and replaced with "Other side has offered 6,500; client considering." Two errors in five lines is not unusual in the instructions section, which is why that section is the one the lawyer rewrites rather than skims.
Names, dates and figures need a separate pass, because transcription errors in them look perfectly plausible. A second prompt, run on the same transcript, makes them easy to check:
List every sum of money, date, deadline, person's name and company
name in this transcript. For each, give the timestamp and the exact
words spoken around it. Do not correct or interpret anything.
[transcript]Illustrative output from the same tour-operator meeting:
- 6,500 (12:14) "they've come back at six and a half"
- 30 September (18:40) "if we hear nothing by the thirtieth"
- 16,500 (21:05) "the total loss is about sixteen five"
- Coach company name (04:10) [spelled as transcribed]Checked against the recording, "sixteen five" was right, but the draft note had rounded it to "about 16,000", so the note is corrected to the client's figure. The company name was misheard: the transcript's spelling did not match the client's booking records, so the note uses the spelling from the records and cites them as the source. The pass takes about five minutes and catches the errors a lawyer reading for sense tends to miss.
Meetings held through an interpreter need one more line. The transcript captures the interpreter's words, not the client's own, so the note should say the meeting was interpreted, give the interpreter's role, and record that the instructions were read back through them. Quoting a transcript line as "the client's words" would be wrong in that case.
The corrected note, not the AI draft, goes on the file. Do this the same day, while the lawyer still remembers what was said.
Retention, deletion and what a records request can reach
A firm needs a written answer to "what do we keep, and for how long?" before the first recorded meeting. A filled-in rule for a small firm (illustrative, to be checked against your own retention policy and regulator's rules):
| Item | Kept where | Kept for |
|---|---|---|
| Checked attendance note | Matter file | Same as the rest of the matter file |
| Raw transcript | Matter file, marked "unchecked AI transcript" | Until the matter closes, then deleted, unless a dispute is likely |
| Audio or video recording | Meeting platform only | 30 days, then deleted |
| Copies in the note-taker's own app | Nowhere | Deleted once saved to the matter file |
Two reasons to decide this deliberately. Transcripts and recordings are records like any other: if a person makes a request for their data under data-protection law, or a dispute leads to disclosure, they can be within scope. And if a dispute is likely, deletion has to stop; the matter's hold overrides the routine rule.
A realistic slip, and how it came to light
Here is the scenario the settings are there to prevent. A lawyer at a small firm connects a note-taker bot to her calendar in its default mode. The next week she joins a settlement call with the lawyer for the other side. The bot joins too, as an extra name in the participant list. The other lawyer notices, objects, and asks for confirmation that the recording will be deleted. The firm can give it, but the call starts ten minutes late on a bad footing, and she now has to write to her own client explaining the delay.
The fixes take fifteen minutes and should be done firm-wide before anyone uses a bot: auto-join set to "meetings I choose"; auto-share switched off; calendar connections reviewed; and a rule that meetings with anyone outside the client relationship never have a note taker without everyone's express agreement. The general risks are covered in whether AI meeting note-takers are safe for client calls, and the confidentiality basics in whether lawyers can use ChatGPT without breaching client confidentiality.
Five questions for any note-taker vendor
- Is our meeting content used to train or improve your models, and can we switch that off contractually, not just in settings?
- Where are recordings and transcripts stored, and can we set our own retention period?
- Can admins control auto-join, auto-share and calendar connections for every user in the firm?
- How do participants know they're being recorded, and can we require their consent before it starts?
- Will you sign a data processing agreement, and who are your sub-processors?
A vendor that answers all five clearly, in writing, is one a firm can defend. One that points you to a marketing page for each answer isn't.
A monthly spot-check that shows the rules are being followed
Settings drift. People reconnect calendars, new starters install their own tools, and the "delete once saved" step gets skipped when a week is busy. A twenty-minute check once a month, by whoever supervises the files, shows whether the rules above are holding:
- Pick three matters with a recorded meeting in the last month. Open each filed attendance note and look for any [CHECK] marker left in it. There should be none.
- For one of those meetings, compare the instructions section of the note with the transcript at the point the decision was discussed. It should match the client's words, not the model's paraphrase.
- Open the note-taker app for two users and count the transcripts still held there. Anything already saved to a matter file should be gone.
- In the admin console, check auto-join and auto-share for every user, not just the ones you expect to use the tool.
- Check that each sampled matter has a consent entry for that meeting.
An illustrative first month in a four-lawyer firm: two of the three notes were clean, and one still had "[CHECK] client may prefer mediation" in the filed copy. One user's app held 11 transcripts that had already been filed, and a trainee had connected a free note-taker account to her calendar with auto-join on. None of that had caused a problem yet, which is the point of looking. The fixes were a two-line reminder to the team, a deletion, and a rule that nobody connects a note taker to a work calendar unless the firm's admin sets it up.
Further reads
- AI Note Takers for Financial Advisers: Compliant Meeting Records — How another regulated profession handles AI meeting records.
- How Much Does an AI Note-Taker Cost per Seat? — What note-taker tools cost per seat once you pick one.
- How to Summarise Bundles and Transcripts With AI Safely — Summarising longer transcripts and bundles safely.
- AI Acceptable Use Policy for a Small Professional Firm — Put the note-taker rules into the firm's AI policy.
- How to Turn Meeting Notes Into Tasks Automatically With AI — Turn checked attendance notes into tasks automatically.
- Which Legal Tasks Should a Small Firm Never Hand to AI? — Where AI help stops in legal work more generally.
- ChatGPT or a Legal AI Tool: Which Should a Small Firm Use? — Which legal tasks a business ChatGPT plan handles safely, which need a legal research tool, and how a three-lawyer firm split the budget.
- AI Implementation Plan for a Small Law Firm: The First 90 Days — A day-by-day 90-day AI plan for a six-person law firm: policy first, two scored pilots, shadow testing, an error log and a go/no-go decision.
- AI Time Capture for Solicitors: Recover Hours You Never Billed — How passive AI time capture works for solicitors, the three ways a small firm can start, a daily review routine and honest arithmetic on what it recovers.
- Is It Safe to Let AI Listen to Mortgage Advice Calls? — The data on a mortgage advice call, six risks with a control for each, vendor small print to check, consent wording and an accuracy test to run first.
- Telling Patients You Use AI Note-Taking: Consent Wording That Works — Ready-to-adapt wording for telling patients about AI note-taking, from the booking text to the in-room question and the information page.
- AI vs Human Transcription: Which Is Worth Paying For? — When AI transcription is good enough, when a person is worth $1.99 a minute, how to test accuracy on your own recordings, and the hybrid in between.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Google Workspace Updates on explicit consent for Gemini notes, recordings and transcripts in Meet (April 2026); Otter pricing and help articles on auto-join and auto-share; reporting on the consolidated Otter privacy litigation (2025-2026); Zoom and Microsoft documentation on built-in meeting summaries.