AI meeting notes into tasks: a workflow for client projects

The short answer

Record the call, transcribe it, and ask the AI for four lists: decisions, to-dos with an owner, dates, and open questions. Then spend five minutes checking the lists against what you remember, move the client's to-dos to the client, and send a short recap. The tasks go into the project the same day, not into a notes document nobody opens again.

The recording is the part people skip, and it is the part that makes the rest work. Notes taken by hand during a call are thin exactly where the discussion was most interesting, because that is when the note-taker was talking.

Before the call

Three things make the transcript far more useful.

1. Put the agenda on the project before the call: the ticket or tickets you will discuss and the questions you need answered. The AI can then sort what was said under those headings.
2. Ask everyone at the start whether it is fine to record. In many countries you have to, and clients rarely say no when you say why: "so nobody has to take notes".
3. Decide who owns the call. One person on your side runs it, asks the closing questions and checks the list afterwards. Without an owner, every attendee assumes someone else wrote it down.

During the call

You do not need to take notes, but you do need to say a few things out loud, because the AI can only extract what was said.

• Say decisions as decisions: "So we go with the second route."
• Say owners and dates together: "Anna sends the product photos by Friday."
• At the end, read the promises back. Two minutes of "so I have: we do this, you do that, by these dates" turns a rambling call into a clean transcript.

That closing summary is also the most reliable thing in the transcript. If the AI's list and your spoken summary disagree, trust the summary.

After the call: the workflow

1. Get the transcript. Most call tools and project tools can transcribe a recording; check the names, which speech recognition often gets wrong.
2. Ask for the four lists. Use your tool's built-in feature or the prompt below.
3. Check the lists. This is the step you cannot skip; the checks are in the next section.
4. Put the tasks where the work happens. Team to-dos go on the tickets they belong to, with dates. Client to-dos go on a "waiting on client" list.
5. Send the client a recap with their to-dos and the decisions, the same day.

Here is a prompt for when your tool has no meeting feature:

Below is the transcript of a call with a client about the projects
listed at the top. Make four lists.

Decisions: what was agreed, one line each.
To-dos: one line each, with the owner and the date if one was said.
  Mark each owner as OURS or CLIENT.
  If no owner was said, write NO OWNER. Do not guess.
Dates: every date or deadline mentioned, with what it is for.
Open questions: anything raised and not settled.

Ignore small talk. Do not add tasks that were not said.
If a to-do belongs to one of the listed projects, name it.

Projects: [list]
Transcript: [paste]

Checking the AI's list

AI meeting notes go wrong in a few predictable ways. Check for each one.

• Invented owners. The model assigns a task to whoever spoke last. Anything marked NO OWNER needs one before it goes on the board.
• Client tasks on your list. "We need the logo files" said by your designer is a client to-do. AI often files it under the team.
• Misheard names. Client names, product names and file names come out of speech recognition wrong. A task for "Fristram" when the client is "Fristam" will never be found by search.
• Soft promises turned into tasks. "We could maybe look at a video version" is not a commitment. Delete it or move it to open questions.
• Several projects in one call. Clients with three jobs talk about all three. Make sure each task lands on the right project.
• Dates without a year or a weekday. "The 14th" next to a month boundary is a guess; write the full date.

The check takes five minutes on an hour-long call. It is still far quicker than writing the notes yourself, and the result is more complete.

The recap to the client

Send it the same day, short, with their to-dos first because those are the ones that slip.

Thanks for the call. What we agreed:

From you:
- Product photos for the launch page, by Friday 14 November.
- Sign-off on the second logo route, by Tuesday 18 November.

From us:
- First draft of the launch page on Thursday 20 November.

Decided: the launch moves to 1 December. Open: whether the video
version is in this phase. We will come back on that with a price.

The recap is also your record. If a date is later disputed, the recap the client did not correct is what you point to. Agreeing scope changes in writing like this is covered in how to write a statement of work, and if the call added work, put it on the next quote with the free quote template.

How this works in OIO

In OIO you record the call from a desktop browser onto a ticket or the project, and existing audio or video files can be uploaded to Transcribe. The recording is transcribed and summarised on the ticket, split into chapters, and the transcript can be searched or downloaded as a text file.

From the recording, Extract to note lets you tick what you want, the summary, deliverables, dates, to-dos, decisions or open questions, and adds those sections to the ticket's note. The date reader picks the kickoff and the deadline out of the call. Tickets from meeting reads the transcript and proposes tickets across all the projects the call discussed, and skips work that an open ticket already covers. Nothing is created until you confirm each one, which is where the checks above happen.

The assistant, transcripts and ticket tools are on the paid plans. For how this fits with briefs, feedback and client updates on the same job, see AI project management for agencies and using AI to write client updates. You can try it on one project for free.

Questions