An AI meeting summary should leave a disagreement recognizable

Consider a fictional design meeting. One person wants an automatic AI reply because it saves time; another wants draft-only replies because the messages can create commitments. Both understand the proposal, but they have not agreed.

A summary that says “The team aligned on automation” erases the unresolved choice. Try a disagreement record instead:

1. Proposal: what action is being considered?
2. Each position: wording its owner can correct.
3. Shared facts: what both accept, kept separate from preferences.
4. Open choice: what has not been decided.
5. Next step: who has freely agreed to investigate or decide, and within what limits.

Ask each participant: “Does this represent your view, even if it does not endorse it?” Let them correct the record or decline attribution. Confirming an accurate summary is different from consenting to the proposal.

If no owner or next step has been agreed, write that explicitly. A polished summary should not invent consensus to create a tidier ending.

What phrase in a meeting summary would make you worry that your disagreement had been softened into agreement?