How to Write Meeting Follow-Ups in Ten Minutes Using Otter.ai and Claude

The follow-up email after a meeting is one of those tasks that should take five minutes and routinely takes two hours or never happens at all. If you manage a team or run back-to-back calls, you know the specific friction: the meeting ends, six other things land immediately, and the follow-up sits open in a tab until it is either rushed or quietly abandoned. This article is for managers, ops leads, and anyone who runs meetings and is responsible for what happens after them. By the end, you will have a two-tool workflow and a ready-to-use prompt that gets a clean, accurate follow-up out within ten minutes of any call ending.

Why This Matters

The problem is not that follow-ups are difficult to write. The problem is timing and accuracy together. You finish a call with a reasonable sense of what was decided, but the details blur fast. You remember the decisions but not who owns what. You have the action items but not enough context to make them useful to someone who was not in the room. Writing from memory produces follow-ups that are approximate at best. Notes taken during the call are patchy because you were also talking. The follow-up either gets delayed until the specifics are gone, or it goes out fast but thin. Neither outcome is good enough.

The System

The workflow uses two tools: Otter.ai for transcription and Claude for drafting. No other steps, no intermediate tools.

Record the meeting through Otter.ai and let it run the full transcript automatically. Do not touch it during the call. When the meeting ends, open the transcript and copy the full raw text. Do not summarise it yourself first. Do not edit it. Paste the complete transcript directly into Claude and use this prompt exactly as written.

PROMPT: Here is a transcript from a work meeting. Please write a follow-up message I can send to all attendees. Include: a one-sentence summary of what the meeting was for, the key decisions made, the actions agreed with the name of the person responsible where mentioned, and any open questions that were not resolved. Write it in plain, professional prose. Do not use bullet points. Keep it under 200 words.

One thing worth noting: asking Claude to summarise the transcript first and then write the follow-up as a second step produces worse results. The summary flattens the specifics and the follow-up becomes too vague to be useful. Keeping it as a single instruction with explicit output requirements produces noticeably better output. Run it as one prompt, not two.

This is the kind of practical AI workflow covered every week in the AI at Work newsletter on LinkedIn. Subscribe to get one tested system per week, directly in your feed.

What to Expect

The follow-up goes out within ten minutes of the call ending, and it is more accurate than anything written from memory. Action items are pulled from what was actually said, not from what you thought you heard. Claude will occasionally surface an open question you had mentally resolved in your head but never actually closed in the conversation. That turns out to be useful more often than you would expect.

The format is not perfect in every situation. For calls with heavy crosstalk, the Otter transcript gets messy and the output reflects that noise. For one-to-ones and smaller structured calls the workflow runs very cleanly. For large group calls with ten or more people the results are patchier. Adjust your expectations based on call size and structure, not on the prompt.

Start Here

Record your next meeting in Otter.ai. When it ends, copy the raw transcript and paste it into Claude with the prompt above. Do not adjust the prompt yet. Run it once exactly as written and read what comes back. If the output is roughly right, you have a repeatable workflow. If something is missing, you will know precisely what to add to the prompt next time. That is the whole starting point.

For more systems like this one, the AI at Work blog at navi-at-work.blogspot.com covers practical AI workflows tested for real business use.