During your meets, Gemini can now take notes and summarize the key points of your videoconference.
Another AI use to gain efficiency. Since the launch of Gemini in Google Meet on March 12, users of the workspace suite can use generative AI to transcribe and summarize key exchanges. A long -standing feature that should have been released with the arrival of Gemini for Workspace. After 15 days of use, the JDN shares you a first return of experience and some tips to optimize the work of the AI.
How to activate Gemini in Google Meet?
Officially called “Take notes for me”, Google AI allows you to transcribe and synthesize the key elements of a meeting in a Google Doc. Its use is simple: just launch a Google Meet and activate the note intake in the side panel by clicking on “Start taking notes.” Google’s AI (without exactly knowing what model exactly) automatically transcribes audio content from Reunion live. Once the meeting is completed, a complete doc document is created by Gemini. It is shared in all participants (with publishing rights) and also added directly as a part of the Google Meet invitation.
The final document is rather well structured. The latter contains all the guests and their respective email addresses, a short summary of the exchanges and finally a detailed course of the meeting in the form of a chip list. All the major points in the meeting are transcribed there. The comments are also sourced with the exact timecode of their appearance in the discussion. Finally, a classic transcription which traces all the exchanges is available in the “transcription” tab of the final doc.
A functional AI despite 3 major limitations
According to our various tests, the final document is always operational and never gets lost (or almost) in secondary discussions. Beware, however, of private conversations before a meeting, the latter will sometimes be transcribed in the meeting notes in a more or less random manner. The “details” point of the document is always faithful to the discussions. It mainly takes up the factual elements addressed (example: presentation of results) and the tasks to come with the names of the actors involved. (Example: X must return its financial audit on Monday at 9 a.m.).
Overall the results are positive. However, three major problems arise regularly. First, the AI only takes into account (in its transcription from Reunion) the names of the participants connected to Google Meet. If several participants are connected to the same Google account or – as is the case in many companies – in a meeting room, the AI allocates all comments to one person. A real limitation which can quickly make the account unusable.
Second limitation: Gemini cannot (yet?) “See” the content broadcast on the screen. It is then impossible to obtain the transcription of data presented in screen sharing. Too bad when you know that Gemini has the capacity to carry out this type of operation (visual modality live). Finally, obstacle, and not least for international teams, taking note can only take place in one language. Consequently, it is impossible to switch from language to the risk that AI begins to completely derail (example: start a meeting in French and finish it in English with a non -French -speaking interlocutor).
Availability and advice
“Take notes for me” is available with Gemini for Workspace. Gemini features are now offered free for all workspace business and Enterprise accounts. To use AI in Meet, your administrator must also have activated the module. Personal accounts can access the tool by subscribing to a Google One AI Premium subscription for 21.99 euros per month. Finally, “smart features” must also be activated in your personal workspace account. For the time being, taking note by IA operates in French but also in English, German, Italian, Japanese, Korean, Portuguese and Spanish.
To maximize the results with Gemini in Meet, Google recommends activating the functionality for meetings equal to or greater than 15 minutes. Pretty logically, participants also have to have a good internet connection and that sound quality is sufficient for a good transcription. Google still recommends creating a Google Meet link (or an agenda event) by meeting. Duplicated events or links do not allow transcription.
Finally, to maximize the relevance of the final report and to avoid biases linked to the multi-user in a single session, we recommend that you conduct meetings with a user position. To optimize a little more, it can be good to present yourself quickly once before the first speaking. Example: “Laurent of the marketing service”. During the assignment of the tasks, it is also advisable to be precise and clear by giving for example the first name of the participant concerned. Example: “Marguerite, can you give us this file by Monday evening?”. Finally, a small oral summary point at the end of the meeting is a real plus so that the AI understands precisely the decisions that have been established.
In itself, Gemini will not replace a much more precise manual transcription for strategic meetings. But it is of real use for everyday meetings.
