Why we built meetnotes: AI minutes and data requirements
In short: AI meeting notes can reduce writing work, but information may be processed by providers the organisation has not assessed. We built meetnotes to make the easy path and the defensible path the same path. Here's what we learned along the way, and what you should demand of any tool you pick.
The problem doesn't start with the tool. It starts with the meeting
A common challenge may look like this: Minutes are a chore nobody wants. Whoever writes them isn't really in the meeting. They arrive three days late, if they arrive. And half the decisions live only in the heads of the people who were there.
So someone finds a free AI tool and solves their own problem. It works. Colleagues copy them. Without checking the terms, the organisation may lack clarity about storage, processing and any use of content for training.
That isn't the employee's fault. It's the predictable result of the easy path and the defensible path not being the same path.
The entire design premise for meetnotes was to make them the same path.
Four requirements we fixed before writing a line of code
1. Data belongs in the EU — and you must be able to say where it is
It's worth being pedantic here, because the sub-processors are typically where it breaks: the interface runs in Europe, but transcription is sent somewhere else. When a head of IT security asks where the data sits, having two answers doesn't work.
So here is ours, without hedging: meetnotes data is stored in the EU region, and we don't train on customer content. Some sub-processors may process data outside the EU/EEA, and that happens under standard contractual clauses. The full list is in the privacy policy.
It isn't the cleanest answer anyone could give. But it is an answer, and one you can check. You should be able to get that from any vendor — us included.
2. No training on customer content
A set of minutes can contain a salary negotiation, a customer's price level, or a strategy that hasn't been announced yet. Training data is hard to claw back. So it's easier not to — and to put that in the terms, not in a support email.
3. Deletion has to mean deletion
Deleting a meeting must remove the audio file, the transcript and the minutes. Not hide them in the interface. It sounds trivial, but it's the difference between being able to answer a subject access request and not. With us, deleted meetings sit in the trash for 30 days and are then removed permanently; delete your account and all content goes immediately.
4. Minutes are structured fields, not nice prose
This is the least legal and most underrated decision in the whole project. A language model can happily write flowing minutes. But minutes aren't prose — they're data. Decisions, tasks, owners and deadlines have to be separate fields you can follow up on, filter and export. Otherwise you've automated the production of text nobody acts on.
What meetnotes does
meetnotes records the meeting, transcribes it while you talk, and delivers minutes with a summary, decisions and tasks — each with an owner and a deadline. On top of that:
- A mind map of the meeting's topics, which makes it fast to get the picture without reading the minutes
- Attachments count — add the agenda, slides or a spreadsheet, and the model knows the numbers, names and terminology up front instead of guessing
- Meeting language and minutes language are chosen separately — hold the meeting in Danish, send the minutes in English to an international board
- Export to PDF, Word and Markdown, plus a calendar file for the next meeting
- Trial options and current product terms are available on the product website
See the product and current pricing at meetnotes.io.
What to demand — whichever tool you choose
We obviously think meetnotes is a good choice. But this is the list we'd use if we were sitting on the buying side of the table:
| Requirement | What to ask for |
|---|---|
| Geography | A written answer on where data is processed and stored — including sub-processors and their regions |
| Training | That the terms state your content is not used for model training |
| Data processing agreement | Handed over within days, without a negotiation |
| Deletion | That deleting removes audio, transcript and minutes — ask them to confirm it |
| Structure | Decisions and tasks as fields, not as paragraphs |
| No emotion analysis | Tools that score participants' mood or engagement don't belong in a workplace — and are prohibited under the EU AI Act |
| Exit | What happens to your data if you cancel? Can you get it out in a usable format? |
Note the last one in particular. It's almost never asked at purchase, and it almost always becomes relevant.
How to roll it out without starting a fight
The technology is the easy part. What decides whether it works is how you introduce it.
- Start with one team and one meeting type. Project or steering group meetings. Not leadership meetings, not performance conversations.
- Write a meeting policy before you start — not afterwards. Above all, it should say which meetings you don't record.
- Involve employee representatives. It costs half an hour and saves a conflict.
- Make it obvious. In the invitation and out loud at the start of the meeting. And give people a real, consequence-free way to say no.
- Keep a human in the loop. The minutes are a draft until the chair has approved them. Say so on the document.
- Measure something real. Not "hours saved", but: do the minutes go out the same day? Does every task have an owner and a date? Can an absent colleague act on it?
The full legal walkthrough — lawful basis, transparency, data processing agreements and retention — is here: Can you record a meeting and let AI write the minutes? (in Danish).
About Q-analytics
Q-Analytics ApS builds AI solutions for organisations where data can't simply be sent anywhere. Alongside meetnotes we develop QWorker, which helps advisors and companies analyse and respond to public tenders, and MemoryRaven.
Data location, models and subprocessors must be assessed for each product and its particular use.
Want to know more? See meetnotes.io, or get in touch to discuss AI and data requirements in your organisation.
Written by Q-Analytics ApS (CVR 42163627, Allerød, Denmark). The sections on regulation are general orientation, not legal advice. Last updated 31 August 2026.
Read more on meetnotes.io (Danish):