Skip to content
L&L Synergy
Get in touch

HomeBlogAI meeting notes

Automation · Office AI meeting notes: how to record meetings without losing legal footing or trust

Meetings cost twice: once in the session and once in the minutes. AI can take notes, summarise and pull out tasks. Whether that works depends not only on the quality of the summary but also on who may record, where the data is kept and who checks the result.

Updated October 2026 · 5 min read · L&L Synergy

In short

  • An AI‑written record arises in four steps: recording, transcription, summary to a template, distribution with tasks. A person checks before it is sent.
  • Recording and transcription process personal data. All participants must be informed in advance. In Germany and Austria, unauthorised sound recordings are a criminal offence.
  • AI systems that infer the emotions of employees in the workplace have been banned under the AI Act since 2 February 2025, except for medical or safety reasons.
  • Where a works council exists, introducing such tools may be subject to co-determination.

How an AI‑written record arises

Four steps work behind the finished minutes. Each has its own sources of error.

Step What happens What to watch
Recording The sound of the meeting is recorded or turned into text directly Everyone informed, store only what is needed, delete the recording after transcription
Transcription Speech recognition turns the sound into text, usually with separate speakers Keep names and technical terms in a glossary, check samples
Summary A language model writes to your template: decisions, tasks, open points Fixed template, support decisions with the passage in the text, invent nothing
Distribution Minutes go to participants, tasks go into your task or ticket system Approval by the minute-taker, clarify access rights

Speech recognition does not have to run in the cloud. Whisper is a freely available speech recognition model released under the MIT licence that can be run on your own machines. Whether that is necessary depends on how sensitive the conversations are.

What works well and what does not

  • Decisions and tasks: who does what by when? This can be pulled out of a conversation well if the template asks for it.
  • Short versions of long meetings: anyone who was not there reads in two minutes what was decided.
  • Customer call notes: the note, key statements and next steps go straight into the customer file.
  • Limits: people talking over each other, strong dialects and technical jargon worsen the result. Language models can add content that was never said. OWASP lists false outputs as a risk in its own right (“Misinformation”). That is why a person checks.

Law and co-determination

  • Information and legal basis: Recording and transcription are processing of personal data. You need a legal basis (Art. 6 GDPR), must inform people (Art. 13 GDPR) and must limit the data to what is necessary and keep it only for as long as needed (Art. 5 GDPR). Say so in the invitation and at the start. External guests agree explicitly.
  • Recording law: Unauthorised sound recordings of non-public spoken words are a criminal offence in Germany (§ 201 StGB) and Austria (§ 120 StGB).
  • No emotion analysis: Art. 5 of the AI Act bans AI systems that infer the emotions of people in the workplace, except for medical or safety reasons. Switch off features such as “participant mood analysis”.
  • Works council: In Germany, technical equipment intended to monitor the behaviour or performance of employees is subject to co-determination (§ 87(1) no. 6 BetrVG). In Austria something similar applies to control systems that affect human dignity (§ 96(1) no. 3 ArbVG). Involve the works council before you start.
  • Provider: You need a data processing agreement with the provider (Art. 28 GDPR). Clarify server location, retention period and whether recordings or texts are used for training.

This overview is general information, not legal advice. Check your individual case with your data protection officer or lawyer.

How to introduce AI meeting notes

  1. Define the purpose: which meetings should be minuted? Not all of them, and above all no confidential sessions such as HR conversations.
  2. Write the rules: who may record, what notice goes in the invitation, how long is what stored?
  3. Define the template: decisions, tasks with owners and deadlines, open questions.
  4. Choose a tool: the feature in your video conferencing tool or a process of your own, with local transcription if you wish. Data protection check first.
  5. Pilot with one team: a few weeks, comparing the AI‑written minutes with a manual record.
  6. Define approval: one person checks the minutes before they are distributed.

Sources

  1. Regulation (EU) 2016/679 (General Data Protection Regulation), Art. 5, 6, 13 and 28: EUR-Lex
  2. Regulation (EU) 2024/1689 on artificial intelligence, Art. 5(1)(f): EUR-Lex
  3. Germany: § 201 StGB and § 87 BetrVG
  4. Austria: § 120 StGB and § 96 ArbVG
  5. OpenAI: Whisper, speech recognition model
  6. OWASP: Top 10 for LLM Applications 2025

Frequently asked questions

Do I have to tell all participants?

Yes. Say in the invitation and at the start of the session that AI is taking notes and what happens to the data. External guests should agree explicitly before recording starts.

May I minute HR conversations or advisory conversations this way?

We advise against it. For HR conversations, meetings with the works council or conversations under professional secrecy, special rules apply. Clarify such cases individually and, when in doubt, with legal advice.

How accurate is the transcription?

It depends on sound quality, language, dialect and technical terms. With a good recording, standard speech usually works well; talking over each other and jargon are harder. Test with your own meetings and keep a glossary.

Is the feature in Teams or Zoom enough, or do I need a system of my own?

For many cases the built-in feature is enough, once data protection and co-determination are settled. A process of your own pays off when minutes are to flow into your systems, such as tasks into the ticket system and notes into the customer file, or when data should not leave your premises.

Where should your next minutes write themselves?

Tell us briefly which meetings you record and where the results should go. We will get back to you personally with an assessment, free of charge and without obligation.

Get in touch

Read on

AI use cases for business: 12 examples AI with your own data Process automation AI for law firms AI for architects Construction AI and automation AI reporting and report automation AI for management and controlling