What a good meeting write-up contains

The four sections, and what each one is for

Notes come out in a fixed order, and the order is fixed on purpose. Notes whose shape changes from one meeting to the next cannot be skimmed, compared or trusted, because you can never tell whether a missing heading means the meeting settled nothing or the model forgot to look.

The summary is two to six sentences on what the session was about and where it landed. It is not a topic list. The concrete detail belongs here - names, numbers, dates, systems - because that is what people come back for. A summary that says "the team discussed the migration" has thrown the meeting away and kept the calendar entry.

Decisions is one bullet per decision actually made, written as a decision rather than as a discussion that leaned one way. "We ship on the 30th and cut scope on search" is a decision. "The team talked about the date" is not.

Action items is one bullet per task, as who - what, with the deadline where one was given. That is the only section anybody will still be reading next week, so it is written to be lifted straight into a tracker.

Open questions is what was raised and left hanging. It is the section people forget to ask for and then need, because the thing that stalls a project is rarely a decision anybody made.

A section that says "None." is doing work

Decisions and action items are never dropped from a set of notes. If the meeting settled nothing, the heading is still there with "None." under it.

The distinction sounds pedantic until you have read notes with no Decisions heading and had to reopen the transcript to find out which of the two things happened. An absent section is ambiguous; an empty one is a statement. Open questions works the other way round, and is omitted when there were none, because a meeting with nothing outstanding is a good meeting rather than a suspicious one.

Where the owner of an action item comes from

This is the part most write-ups quietly get wrong, so it is worth being precise about what Dictera can and cannot know. Everyone except you arrives on one audio channel, so the transcript has two labels: you, and them. There is no voice fingerprinting and no attempt at one.

Names come from somewhere else entirely: the participants you type into Meeting Setup, or the invitee list read from a calendar invite. Even then, a name is used only where the meeting itself makes clear who is speaking or who is being addressed - somebody says a name, or a person names themselves. An action item whose owner the conversation never established is written Unassigned. That is a rule in the prompt rather than a preference, and it holds even when there is exactly one name on the list and attaching every task to it would look tidier.

If you name some of the people but not all of them, the model is told the list is incomplete. Without that, a model given eighteen names out of twenty treats "not listed" as "not present" and hands every commitment to whichever eighteen it was shown - a mistake that is invisible in the notes and still there weeks later when you go looking for who owes what.

"Next Friday" has to become a date

Meetings are full of relative time. Next Friday, end of the month, in two weeks, after the release. Written down as spoken, an action item stays legible for about a week - which is exactly as long as nobody needs to look it up.

The write-up is told when the session took place, so those references are resolved into actual dates. Where a reference is genuinely too vague to place, the speaker's own words are kept instead of a date being invented. That is the general rule everywhere in the notes: never invent a detail, an owner or a date, and where something is ambiguous, write what is known and mark the gap "(unclear)" rather than resolving it with a guess. A write-up you have to fact-check is worth less than no write-up at all.

What good notes leave out

Half of what makes a write-up readable is subtraction. Greetings, small talk, technical difficulties and scheduling chatter are dropped unless they produced something that belongs in a section - "we lost Ana for ten minutes" matters only if a decision was taken without her.

Sections are ordered by consequence rather than chronology, so the item that matters most is first, not the item that happened first. Nothing is padded to make a section look complete, and nothing is repeated in one section that another already says. Direct quotes appear only where the exact wording carries the meaning: a commitment, a figure, a definition. Everything else is written plainly.

When the recording is not a meeting

Summary, decisions and action items describes a working meeting and collapses on everything else people record. A lecture has no decisions and no owners. Forced into that structure, it manufactures action items out of "read chapter four".

So the structure follows the session. Left on automatic, a recording is read for what it is - meeting, lecture or talk, one-on-one, interview, sales call - and the notes say which shape they used, so a wrong guess is one click from corrected rather than a silent distortion. The headings change with it. A one-on-one gets feedback, commitments and concerns raised. A lecture gets key points, key terms, worked examples and the Q&A. A sales call gets what they need, objections, commercial points and next steps. An interview gets strengths, concerns and the answers worth remembering. The rules about not inventing anything are the same in all of them.

Notes are one rung on a ladder

Notes keep the outcome, and the outcome is a few per cent of a meeting. Every other question about a call - what was actually argued, what someone really said, where in the hour that was - needs the material a summary throws away, which is why Meeting Notes writes four different things from the same recording.

Full minutes keep the meeting: every point, argument and figure, in the order it happened, at roughly a third of the length. The transcript keeps the words, cleaned of disfluencies and mis-hearings, in the language they were spoken and never translated. Chapters keep the shape - a timeline of what was discussed when, with times taken from the transcript rather than estimated. Pick before you press Stop and that is what Stop produces; the other three stay a click away.

When the write-up is not quite right

A write-up is a document, not a reply, so it can be reshaped or edited. Notes and minutes can be cut to the action items alone, reorganised by topic, or turned into a recap email you can send the participants; a lecture offers key terms and a study guide instead. Those rewrites are deliberately not offered on the transcript or the chapters, because a reshaped transcript is no longer a record of anything.

You can also just fix it by hand - a misheard name, a line nobody agreed to - without regenerating the nine-tenths that were right. The model's own version is kept, so Revert still works in a session you reopen next month, and an exported file says that a human edited it.

What the notes still owe you weeks later

The test of a write-up is not how it reads on the day. File a session under the meeting it repeats and every open action item across that series can be gathered into one list. It reads back the notes you already have - no model runs, so it is instant and cannot invent a commitment nobody made - and it tells you when some meetings could not be read instead of quietly showing a short list.

That only works if the individual write-ups were honest. Every Unassigned owner, every "(unclear)", every "None." is what keeps the year-long view from becoming fiction. Notes, minutes, transcript and chapters all export to Markdown with the date, the participants and the model that wrote them, and all of it stays on your Mac - see how the data is handled for what leaves it if you point the assistant at a cloud provider. The rest of what the panel does is on the AI Assistant page, and if you are wondering why none of this appears while the call is still running, that decision has its own reasoning.