A transcript is a record of what was said. Minutes are a record of what was decided. If you try to make minutes by shortening a transcript evenly, you get a long, vague summary. You want a different shape.
What you need
- The transcript, saved before you start editing anything.
- The agenda, or a list of topics, if there was one.
- The list of who attended and who sent apologies.
- Any rules your organisation has about minutes. That might be a constitution, a board charter, or a lawyer's advice.
- A blank template with fixed headings. You'll fill it in, not invent a new layout each time.
What minutes are for
Minutes are described as the official written record of a group's meetings, and not transcripts of those proceedings1. The same source says that, following Robert's Rules of Order, minutes should mainly record what was done at the meeting, not what was said by members1. It adds that an organisation may have its own rules on content1.
That's one set of conventions, and a small business may not follow it. But the idea carries over. Someone reading in three months wants the outcome, not the chat.
A simple layout
Use the same headings every time:
- Meeting details: name, date, time, place or call link.
- Who was there: attendees and apologies.
- Decisions: one line each, in plain words.
- Actions: what, who, by when. See action items and owners.
- Open questions: things raised but not settled.
- Next meeting: date, if one was set.
Anything that's chat, tangents or jokes can go. Keep a brief note of discussion only where a reader will need the reason behind a decision.
Using the tool's summary as a first draft
If your tool writes a summary for you, treat it as raw material.
One reference says abstractive summarisation writes wording that isn't in the original2. That's useful for readability. It also means the tool can add things. A separate source warns that chatbots built on language models may slip falsehoods into what they write3.
There's a name for the safer way of working. One source describes it as Human Aided Machine Summarization, where a human post-processes software output2. That's what you're doing.
Try this routine.
- Paste or load the transcript and ask for decisions and actions only, not a retelling.
- Ask the tool to list anything it's unsure about.
- Read the result with the transcript open beside it.
- For each decision, find the line in the transcript where it was made.
- Delete anything you can't find.
- Fix names, numbers, dates and amounts by hand.
Common problems
The summary is too long. Ask for a set number of bullets per agenda item. If it's still long, cut by hand.
Decisions are missing. Often nobody said "decided" out loud. Search the transcript for phrases like "let's go with", "agreed" and "we'll do". Then ask the people involved.
The tool made a decision up. This is the one to watch. A line such as "the team approved the budget" must trace back to someone saying it. If it doesn't, remove it.
Opinions are written as facts. A person floating an idea isn't a decision. Label it as a suggestion or leave it out.
Sensitive remarks are included. Minutes get shared and filed. Take out personal comments, health details and anything the group wouldn't want circulated. The page on what to keep out of the recorder covers the bigger version of this.
Before you send
- Compare the draft to the transcript, not only to your memory.
- Ask one attendee to read it if the meeting was important.
- Mark it clearly as drafted with AI assistance and checked by a named person, if your business wants that. Some do, some don't.
- File the final minutes in one agreed place.
- Decide what happens to the recording and transcript. The storage page has the questions to ask.
If the minutes are for a formal body, such as a company board or an incorporated association, check what your governing documents require. Ask your accountant or lawyer for anything beyond that. The tool can draft, but it can't tell you what your legal obligations are.
Notes and sources
- Minutes, Wikipedia
- Automatic summarization, Wikipedia
- Hallucination (artificial intelligence), Wikipedia