AI meeting assistants compared: Otter, Fireflies, tl;dv or just Copilot?

This article is an overview based on vendor information, documentation and publicly available reporting, not a scored test of our own. Prices and features change quickly; the vendor's own terms always take precedence. We only score what we have used ourselves for weeks: those tools are listed under AI tools compared.

What the tools do, and where they get it wrong

All the assistants discussed here work on the same pattern: a participant, usually a bot with its own name, joins the video call, produces a transcript with speaker attribution and then summarises it. Added to that are a list of tasks, often with owners, and a search box for questioning the conversation afterwards.

Transcription is good by now, but not flawless. The typical weak spots are personal names, company names and technical terms, which the model bends into a familiar word, plus moments when several people speak at once. German meetings with English sprinkled in are a case of their own: almost every tool expects one language per conversation and stumbles at the switch.

The summary inherits those errors and adds its own: irony is taken literally, a passing idea becomes a decision, an objection at the margin disappears. Anyone who sends the minutes unread will, now and then, be sending out decisions that nobody took.

Otter, Fireflies, tl;dv: three bots compared

Otter.ai is the oldest of the three services and strongly geared to the English-speaking world. According to the vendor, German is transcribed, but only as one of a handful of languages and only one per conversation. Data is held in the United States by the company's own account. On price, the Pro tier starts at around 17 US dollars a month, considerably less on an annual plan; Business sits at 20 to 30 US dollars per user, as of September 2026.

Fireflies.ai advertises over 100 languages with automatic detection, a multi-language mode in the higher tiers and a very large number of connections to CRM and project tools. According to the vendor, Pro costs around 10 US dollars per user per month on an annual plan and Business around 19, noticeably more when billed monthly. Processing takes place in the US; Fireflies offers EU storage only on the Enterprise tier, and even then processing still happens in the US.

tl;dv comes from a GmbH in Aachen, hosts on European servers by its own account and runs the AI processing in the EU or the US depending on your settings. For German teams that is the most interesting point, and one you should actually verify in the settings. The free tier is usable but tightly limited; according to the vendor, Pro is around 18 US dollars per user per month on an annual plan, and the sales-oriented Business tier around 59.

The built-in options: Copilot in Teams, Gemini in Meet

If you already use Microsoft 365, Copilot in Teams gives you summary, task list and questions to the conversation without an extra bot. It requires the Copilot licence, according to the vendor around 30 US dollars per user per month on an annual plan, with a cheaper Business variant around 20 US dollars for companies with up to 300 seats. The follow-up after the meeting works only if transcription was switched on; an "only during the meeting" mode works with speech data that is not stored once the call ends.

The real advantage lies in administration: admins set by policy whether and how Copilot is allowed in meetings, the transcripts stay in the existing Microsoft tenant, and the data processing agreement is the one the company already has. Microsoft cites the EU Data Boundary for Copilot, though with exceptions that are worth reading up on in the contract documents.

Google solves it similarly: Workspace tiers from Business Standard upwards, according to the vendor around 14 US dollars per user per month on an annual plan, include Gemini, and the "take notes for me" feature writes minutes into the organiser's Drive folder. All participants see an icon as soon as notes are being taken. German is supported, but only one language per meeting.

The legal position: section 201 of the Criminal Code, consent, GDPR

Germany is stricter here than the United States, where most of the tools come from. Section 201 of the German Criminal Code (StGB) makes it an offence to record the non-public spoken word of another person without authorisation, or to use such a recording, punishable by up to three years' imprisonment or a fine. A video call is a non-public conversation. The recording is authorised when all participants know in advance what is happening and agree; simply staying in the call is not consent.

Whether a pure live transcript with no audio storage even counts as a recording under the provision is disputed among lawyers. In practice that helps little: most tools at least buffer audio, and the data protection authorities regularly require a separate consent for recordings of video conferences. So plan on explicit agreement from everyone, external participants included, and document it.

Then there is the GDPR: transcript, summary and speaker attribution are personal data. You need a legal basis, information for the data subjects under Article 13, a data processing agreement with the vendor, a settled processing location and a deletion deadline. And where a works council exists, a tool that attributes statements to individual employees is subject to co-determination. The order of checks is in our article on AI tools and the GDPR.

Rules to write down beforehand

One sentence in the invitation is enough to start with: the meeting will be transcribed, with which tool, for what purpose, and anyone with objections should say so in advance. That is repeated at the start of the call, and the bot carries a name that makes it recognisable as a bot. With external participants, especially in job interviews, customer meetings or advisory conversations, the rule is: when in doubt, without.

Decide who may start an assistant, how long recordings are kept, where they live and who reads the minutes before they go out. HR conversations, health matters and anything involving lawyers belong on an exclusion list. How to keep such rules short is described in our guide to choosing AI tools.

For the follow-up, a general assistant is often enough: paste the raw transcript into Claude or ChatGPT and have it sorted into decisions, open questions and owners, and you frequently get better minutes than the built-in summary. If you file minutes in Notion, Notion AI can do that right there; for the English version to partners abroad, DeepL is the more reliable choice than the meeting tools' own translation feature.

Who an AI meeting assistant is worth it for

For teams with many internal meetings where nobody has kept minutes so far, the benefit is immediate: decisions become traceable, tasks stop getting lost, absentees can read up instead of asking around. If you already work in Microsoft 365 or Google Workspace, start with the built-in feature. It is already covered by contract, centrally controllable and visible to every participant, which makes consent considerably simpler.

A separate service is worth it when there is no suite, when many meetings happen on other people's platforms, or when sales features such as CRM integration and conversation analytics are needed. Then tl;dv makes the shortlist for its European hosting option, Fireflies for its language range and integrations. Otter remains of interest mainly to English-speaking teams.

These tools are unsuitable for conversations where confidentiality is the whole point: HR conversations, mediation, meetings with clients or patients. And for teams that have no time to read the minutes before sending them, an assistant is not a time saver but a risk.

In short: Copilot in Teams or Gemini in Meet are the simplest choice for most companies, because contract, administration and visibility are already in place. Among the separate services, tl;dv has the best data position for Europe, Fireflies the widest language range. In every case: inform all participants beforehand, obtain consent, read the minutes before sending.