Simple AI Auto-Generation of Meeting Minutes

■ Significance & features

In activities that involve the community, occasions to talk—meetings, interviews, debriefings, handovers—arise one after another. Yet what is exchanged there mostly remains only in the memory of those present. Things proceed with “who does what by when” left vague, and by the time you hand over to the next cohort, the reasons behind the seniors’ judgments are lost. In student activities, where members change over each year, this loss is especially large.

On the ground, the following also tends to happen. One is that, with no time to make records, sharing doesn’t occur and the same questions are repeated over and over. Asking a community member or a municipal official the same thing again also takes their time. Another is having no room to hold a dialogue while taking notes. When you are preoccupied with taking notes, you cannot listen deeply to the other person or ask follow-up questions on the spot.

This tool gathers the steps to make everything from a recording to minutes and handover materials with a single smartphone and generative AI. No special equipment or software purchase is needed. By turning what was said into a “written asset,” you can reliably share decisions with the team and pass experience to the next cohort. Above all, by leaving the recording to the machine, you can concentrate on the dialogue at the time.

The whole is four steps: “record → transcribe → draft with generative AI → have a human verify.” Not skipping the last, “have a human verify,” is the key to this tool. The AI draft is a “well-made provisional,” and figures, proper nouns, and owners and deadlines must always be finalized by a human against the source material.

■ How to use

(Four steps)

First, record. Place the smartphone at the center of the desk, within 1 m of the speaker. Reading out “date, meeting name, participants” aloud at the start lets that information become the basic information of the minutes as is. Break long meetings at around 45–60 minutes.

Second, transcribe. There are four routes—iPhone (the standard Voice Memos), Android (standard features on Pixel, Galaxy, etc.), online meetings (Zoom/Teams recording and transcript), and having only a recording file—and the steps for each are covered in the guidance materials.

Third, make a draft with generative AI. Always include in the prompt the sentence “do not guess at information not in the transcript.” Because the suitable form differs by use, use three types—minutes type, key-points memo type, and handover-material type. The prompts for each type are contained in the “worksheet / detailed explanation.”

Fourth, have a human verify. Check these five points against the source material: figures, proper nouns, owners and deadlines, the distinction between decisions and opinions, and the temperature of dissenting opinions. Whether the content is fit to share is also judged at this stage.

(Three checks before you begin)

Give notice of, and obtain consent for, the recording. Do not record without permission. Add a word at the start: “I’ll record this for the minutes.”

Decide where to store the data. Save the recording and transcript in a place decided by the team, not on personal devices.

Check whether the information may be entered into the AI. If it contains personal or unpublished information, check first whether it may be entered.

(Materials used)　*See the “Download” materials

The guidance is a participant-facing document to hand to the members of your activity. It walks through the concrete steps of the four steps, down to the app names.

The worksheet / detailed explanation is a document to keep at hand when actually making minutes. It contains a template for the items to write in the minutes (title, date and time, place, participants, decisions, action items, etc.), prompts by use, and a verification checklist.

■ Track record & usefulness

The steps in this guide are used in university classes and project settings to record meetings, record Q&A at debriefings, and create handover materials. It is especially effective in the following three situations.

・Presentations and debriefings: Q&A and comments tend to be let slip on the spot, but looking back later, they are full of hints for improvement

・Regular meetings: writing out decisions and action items each time prevents “he-said, she-said,” and lets you check the previous action items at the start of the next meeting

・Handover: have the AI make a draft and a question list from the materials at hand, and have your senior answer verbally only the things not in the materials. Using this procedure, a handover that used to depend on individuals takes shape in about half a day’s work

■ Next steps

(Making records an asset of the activity)

Once minutes have accumulated, line them up chronologically in the team’s shared folder and read them back once at the change of academic year. If they are in a state where you can follow the course of decisions, that itself becomes handover material for the next cohort.

Alongside, if you make and grow a “glossary” of proper nouns and technical terms, mis-conversions in transcription decrease and the work becomes lighter and lighter.

(References and their overview)

What this guide covers is, after all, simple auto-generation. If you are interested in more full-fledged minutes operations or in setting rules for AI use, proceed after checking the generative-AI usage policy set by your university or partner municipality.

■ Glossary

Transcription: converting audio into text data. It can be created with a smartphone’s standard features or the features of online meeting tools.

Prompt: the instruction to a generative AI. Write what, in what format, and how far it may go. In this guide, always include the constraint “do not guess at information not in the transcript.”

Hallucination: when an AI plausibly fabricates information not in the source material. Prevent it with both prompt constraints and human verification.