In short
Before sending AI meeting notes to a client, check every commitment, number and name against the transcript, remove anything said off the record or meant for your side only, label the notes as AI-written and checked, and send text rather than the recording. Invite the client to correct anything.
Sending AI meeting notes to clients needs more care than sharing them internally, because the client will act on them and may hold you to them. Before you send: check every commitment, number and name against the transcript; remove anything said off the record or meant for your side only; label the notes as AI-written and say that you checked them; and send text, not the recording. Then ask the client to reply with corrections.
This guide is a checklist for that step. For the follow-up email itself, see the follow-up email after a meeting.
Send AI notes to a client
- Check the client allows it. Some clients and contracts restrict AI tools on their information. If you have not asked, ask before the first recording, not after.
- Check commitments against the transcript. Every "we will" and "you will", with the owner and the date. AI summaries are most likely to go wrong on who committed to what.
- Check numbers and names. Prices, quantities, dates and people's names are where transcripts and summaries both make mistakes. Check them against the recording if the transcript is unclear.
- Remove what should not go. Anything said off the record, internal pricing discussion, your team's asides, and any criticism of third parties.
- Rewrite in your own voice where it matters. A summary that sounds like a machine is fine for a recap; a commercial commitment should be in words you chose.
- Label it. One line saying AI drafted it from the transcript and you checked it.
- Send text, not audio. Paste it into an email or attach a document. Keep the recording on your side.
- Ask for corrections. "Let me know if I've missed or misstated anything" turns the client into a second checker.
Why client notes need more care
Internal notes that are slightly wrong get corrected in the next standup. Client notes become part of the relationship:
- They look like commitments. A recap that says "we will deliver by the 15th" reads as your promise, whether or not anyone said it.
- They get forwarded. The client's colleagues read them without having been on the call.
- They can be quoted back. In a dispute, what you sent is evidence of what you understood.
AI summaries fail in specific ways: commitments nobody made, dissent left out, hedges turned into facts, the wrong person given a task. Can you trust an AI meeting summary? describes each and how to check for it.
What to remove before sending
| Remove | Why |
|---|---|
| Anything said "off the record" | The client asked for it not to be written down |
| Your internal discussion before or after they joined | It was never meant for them |
| Internal pricing, margin or negotiation notes | Commercially sensitive |
| Comments about third parties | Unfair to people who were not there |
| Personal or health details someone mentioned in passing | Not needed for the work, and often sensitive |
| Unresolved guesses ("maybe we could…") written as plans | They read as promises |
Some AI notes are written for you, not for the client. A list of tough questions to follow up, or a note of where your side and theirs disagreed, is useful preparation and usually not something to send as it is.
Labelling notes for a client
Keep it short and put it at the top:
If you rewrote it completely, you can say so, or simply send it as your own recap. How to label AI-written notes you share has more wording, including for documents and tickets.
Text, not audio
Send the notes as text. Text is short, checkable and easy to correct. A recording holds everything everyone said, including the parts you just removed, and it is much harder to redact. If a client asks for the recording, treat it as a separate decision: it should only happen if everyone on the call knew it was recorded, and your contract and policy allow it.
The same applies in reverse: if you are recording client calls at all, say so at the start. Recording client calls on a Mac covers how to ask.
For consultants and agencies
If client calls are most of your meetings, this becomes routine. Two habits help:
- Agree how you will use AI at kickoff. Say what you record, where transcription happens and what is sent to an AI. Put it in the engagement letter if the client wants it in writing.
- Keep the transcript until the work is signed off. It is what you check a disputed recap against.
AI notes for consultants covers the rest of the workflow.
Doing this with Notey
Notey records the call on your Mac, transcribes it there, and writes notes from the transcript text when you ask.
- Follow-up email. One kind of note is a message to the other side confirming what was agreed and asking for what is still open, ready to send once you add their name. Check it with the steps above before sending.
- Summary and action items. It says so when nobody committed to anything, rather than inventing items.
- Tough questions and where you differed are written for you, to prepare the next conversation. Read them; do not forward them as they are.
- Check quickly. Every transcript line plays from its timestamp, so checking a number or a promise against the audio takes a click.
- Labels travel. Export the summary, any note or everything as Markdown, to the clipboard or a file, and the AI and edited-by-you labels go with it.
- Only text leaves your Mac, when you ask: the transcript and the names you gave people, sent to Notey's service and on to OpenAI as a processor, not used for training and not retained by Notey. The recording stays on your disk. How AI meeting notes are made explains the whole chain.
Frequently asked questions
Should I tell a client that meeting notes were written by AI?
Yes, as good practice. A short label such as "Summary drafted by AI from the call transcript and checked by me" tells the client what they are reading. Some clients and contracts have their own rules about AI tools, so check those too.
Can I send the client the call recording?
Only if there is a reason and everyone on the call knew it was recorded. Text is easier to check, correct and redact. Many teams never send audio outside the organisation by default.
What if the AI summary gets a client's commitment wrong?
Correct it before sending, using the transcript. If a wrong commitment has already gone out, send a short correction. Asking clients to reply with corrections catches errors either way.
Is it OK to use AI on client calls at all?
It depends on your contract, your client's policies and the law where you work. Some clients forbid third-party AI tools on their information. Ask at the start of the engagement, and tell people on the call that you record.
Should I send the client the full transcript?
Usually not by default. Send a checked recap, and share the transcript if the client asks and everyone on the call knew it was recorded. A transcript holds everything, including asides.