AI meeting notes for accountants: where client data goes and what to check

The AI meeting assistant I'd point an accounting practice to is Granola. It doesn't need to join the meeting, and the notes are really good. Client data doesn't stay on your computer, though, because the transcript and notes are stored in the United States. So before you use it with a client, ask them, switch off the setting that lets Granola train on your notes, and have an accountant check the notes before they go on file.

Why Granola, and what does it cost?

I use Granola on my own calls. When a business owner books a first call with me on Google Meet, I record it, with their agreement at the start, so what I recommend afterwards is accurate. The recording isn't shared with anyone.

Granola is probably one of the best. That's my view from using it, not the result of a side-by-side test, and I don't take commissions on any software I recommend. What I like is that it uses your computer's audio, so nothing joins the call as an extra participant.

It isn't built for accounting. Granola's own guide for accountants says it "does not integrate natively with accounting practice management platforms like Canopy or TaxDome", so a checked note reaches the client file by export or by copy and paste.

The prices come from Granola's pricing page, checked on 12 October 2026. The rand amounts use R16.51 to the dollar, the XE mid-market rate on 11 October 2026.

I'm still on the free trial, and I'm moving to Business once that's done.

I haven't set Granola up for an accounting client yet. I've been testing it myself. My client uses Read.ai right now, but I think Granola is better, mostly because it has an iPhone and Apple Watch app. That makes it useful in real-life meetings too.

Where does the client's information go?

It goes to Granola's servers in the United States. Here's what Granola's security page says, checked on 12 October 2026:

• "Granola doesn't store the audio from meetings". On a Mac or Windows computer it transcribes as you talk. On a phone it holds the audio for a short time and then keeps only the transcript.
• It keeps the transcript and your notes. "Notes are stored in our US-hosted AWS Virtual Private Cloud".
• Transcription goes through providers "like Deepgram and Assembly", and the summary is written by AI providers "like OpenAI and Anthropic". Granola says it doesn't allow them to train their models on your data.
• "Granola trains on your anonymized data so we can keep making Granola better".

That last line is the one to act on. My rule is that client information only goes into accounts that don't use it to train AI models. In Granola the switch is under Settings, then Preferences, and it's called "Use my data to improve models for everyone". On the Basic and Business plans every person has to turn it off for themselves.

I've done that in my own account. "Use my data to improve models for everyone" is switched off.

For a South African practice, storing notes in the United States needs its own decision. Section 72 of the Protection of Personal Information Act says a responsible party "may not transfer personal information about a data subject to a third party who is in a foreign country unless" one of five conditions is met. Two of them are a binding agreement that gives adequate protection, and the data subject's consent. Granola says it has "a Data Processing Agreement available upon request", so ask for it before the first client meeting.

If you're a SAICA member, subsection 114 of your Code of Professional Conduct (2025 edition) asks for "appropriate action to protect the confidentiality of such information in the course of its collection, use, transfer, storage or retention". A notes tool is one more place where client information is collected and stored. Bookkeepers and members of other bodies should read their own code.

This isn't financial or legal advice. It's for education purposes only, and it doesn't decide whether your setup complies with the Act or with your code.

How do you get a client's consent?

You just ask them. I ask at the start of the call, before anything is recorded.

Section 4(1) of RICA says "Any person, other than a law enforcement officer, may intercept any communication if he or she is a party to the communication", unless it's done to commit an offence. That section covers a conversation you're part of. It doesn't cover where the transcript is stored afterwards, and it doesn't touch your duty of confidentiality. So ask every time, and say where the notes go.

Asking matters more here. Nothing joins the call, so the client sees no sign of Granola unless you tell them. Granola can post a message in the meeting chat that reads "Hey, I'm using www.granola.ai to transcribe and take notes", and it can show a watermark on your video. A chat message tells the client, but it doesn't ask them. Granola's own wording is that "customers remain responsible for determining what notice or consent is required for their use case and jurisdiction".

Here's wording you can adapt:

Do you mind if I use an AI notetaker for this meeting? It writes up notes from what we say and doesn't keep the audio. The notes are stored on the provider's servers in the United States, and I check them before they go on your file. If you'd rather I didn't, I'll take notes by hand.

On my own calls, this is how I ask:

So I plan to use a tool called Granola to record and transcribe and take notes on this meeting. It's a US based tool, so encrypted data is stored there, but not used for any AI training. Is that ok with you? If not that's fine and I'll take notes by hand.

Write the client's answer at the top of the notes. If they say no, close Granola and take notes by hand.

What should an accountant check before the notes go on file?

"A polished answer is not evidence that the information is correct." I wrote that about setting up AI tools around your own tasks, and it applies to meeting notes too, because they read well whether or not they're right.

Here's the checklist I'd give an accounting practice. I built it from the rules I set for AI summaries in my other work. It isn't a professional standard.

1. Figures. Check every amount, percentage, date and tax year against your own notes or the source document.
2. Who said what. Make sure each decision and promise sits with the person who made it.
3. Deadlines. A due date needs a basis. If nobody said a date in the meeting, the note shouldn't carry one.
4. Facts and suggestions. Keep what the client said apart from the next steps the tool proposes, and label those as suggestions.
5. Advice. Anything that reads like advice, a calculation or a conclusion is yours to write or approve. The notes don't replace your review and sign-off.
6. Gaps. If you remember something that isn't in the notes, add it yourself and mark it as added.
7. The copy left behind. Once the checked note is on the client file, decide how long the copy in Granola stays. Its plans include auto-deletion, set per user on Basic and Business.

Here's one Granola got wrong on my own notes. It assumed a date in the call was for this year instead of next year, so I corrected it after reviewing the notes. I also have it highlight any of the items above, so it's easier for me to check it quickly.

Try it next to your own note-taking before you rely on it. The stop rule that goes into my written quotes works here as well: agree up front what would count as not working, "for example if checking its output takes longer than doing the job yourself". If checking Granola's notes takes longer than writing your own, stop paying for it.

What about preparing for the meeting?

Meeting notes capture what happened. Preparation is about what you need to know before you walk in. I set that up separately, as a short brief before each meeting built from the calendar entry and the messages you choose. If the pressure in your practice sits somewhere else, such as chasing documents, my page on AI for accountants and bookkeepers shows where I'd start.

I'm Henno Fourie. I find where admin is costing an owner-run business time and money, fix it with AI around the tools the business already uses, one job at a time, and stay on as the person responsible for AI in that business. Start with a 30-minute call. It's free and online. Bring one admin job that eats your week, and I'll tell you whether it's worth fixing and what the next step costs.

How I wrote this: I drafted it with AI from my own notes and client work, then checked the facts and edited it myself.