I’m secretary to four boards, and much of my work follows a meeting from the notice, through the roll call, to the minutes. Before asking whether AI could write minutes, I start with a more basic question: what are minutes for?
New Jersey’s Open Public Meetings Act answers it. A public body’s minutes must show “the time and place, the members present, the subjects considered, the actions taken, the vote of each member, and any other information required to be shown in the minutes by law,” and they must be “promptly available to the public,” with limited exceptions.1 So minutes are the public’s record of what a body decided and how each member voted. A summary of the discussion doesn’t do that job.
Where a tool could help, with review
- Turning a recording into a rough transcript someone can search.
- Drafting routine parts, such as attendance and the agenda items taken up.
- Listing each motion it thinks it heard, with the time in the recording, so a person can check it.
- Flagging what it couldn’t make out: an unclear speaker, a vote it couldn’t hear or a motion that may have been amended.
What has to stay with a person
- Deciding what motion was actually made and seconded.
- Recording each member’s vote, and never guessing one that wasn’t clear.
- Noting recusals.
- Telling discussion apart from action.
- Presenting the draft to the body, which approves its own minutes.
The most useful thing a tool could do here is say what it doesn’t know. A draft that marks “vote unclear at 1:14:32” is worth more than a smooth paragraph that fills the gap with a guess. Once the minutes are approved, a wrong vote is part of the official record.
The record doesn’t stop at the minutes
Under New Jersey’s guidance for AI and machine learning systems, records created by government systems that use AI are public records.2 A recording, a transcript, a draft, the corrections and the approved minutes are different things. The people responsible for records should decide where each one is kept, and for how long, before any tool is added.
How I judge a tool
I don’t go by how fast a tool produces a first draft. I look at the total time to approved minutes, including checking and corrections, and at the number of material errors that get past review. If a tool doesn’t lower both, it isn’t helping.
Sources
-
Senator Byron M. Baer Open Public Meetings Act, N.J.S.A. 10:4-14 (minutes of meetings). Accessed September 24, 2026. ↑
-
New Jersey Division of Revenue and Enterprise Services, Guidelines for Developing Retention and Disposition Policies for Artificial Intelligence/Machine Learning Systems, February 2025, revised July 2025. Accessed September 24, 2026. ↑