{"course":{"slug":"ai-at-work","title":"AI at Work","url":"https://addaly.com/learn/ai-at-work","level":"beginner","track":"work"},"lesson":{"slug":"meetings-and-the-inbox","title":"Meetings and email","url":"https://addaly.com/learn/ai-at-work/meetings-and-the-inbox","position":35,"of":73,"minutes":8,"answer":"From a meeting, extract decisions, owners and deadlines, never the discussion, and never let AI hit send.","keyIdea":"From a meeting, extract decisions, owners and deadlines, never the discussion, and never let AI hit send."},"body":{"format":"markdown","text":"Meetings and email are where most professionals lose most of their week. They are also where AI help is most oversold, so let us be precise about what works.\n\n## Meeting notes: the recording is not the point\n\nAutomatic transcription is now good in most major languages and cheap or free in most video tools. That is not the win. A transcript of a 60-minute meeting is 9,000 words nobody will read. You have converted an hour you cannot get back into a document you will not open.\n\nThe win is the layer above it. From a transcript, ask for exactly three things:\n\n1. **Decisions.** What was decided, and by whom. Not what was discussed: decided.\n2. **Actions.** Who owes what, by when. Name, task, date. If a date was never said, write \"no date agreed\" rather than inventing one.\n3. **Open questions.** What was raised and not resolved.\n\nThat structure works for a hospital handover, a school department meeting, a client call, a municipal committee. Notice what it excludes: the discussion. Almost nobody needs the discussion. They need to know what they now owe.\n\nThen read it. Attributions get scrambled: transcripts confuse speakers, especially on bad audio, in accented English, or when people talk over each other. \"Priya agreed to cover the shortfall\" when it was actually Rahul is not a small error; it is the kind that turns into a dispute.\n\n## Before you record: ask\n\nRecording and transcribing a meeting is not a neutral act, and consent rules vary widely by country and by state. Some places require every participant's consent. Some workplaces forbid it outright. And beyond the law, there is the plain fact that people speak differently when a machine is listening: the half-formed idea, the honest doubt, the thing said carefully off the record. Say at the top of the call that you are recording, and say where the notes will go. If someone objects, take notes by hand.\n\nDo not record a meeting where clients, patients, students, or personal grievances are discussed unless your organisation has explicitly approved the tool for that. See the next lesson.\n\n## Email: triage over composition\n\nEveryone reaches for AI to *write* emails. The bigger gain is upstream.\n\n- **Reading a hostile or confusing thread.** \"Here is a 14-message chain. What is actually being asked of me, and what has already been agreed?\" This is compression on something painful, and it works well.\n- **Deciding what needs you.** \"Sort these subject lines into: needs a decision from me, needs a reply, needs nothing.\" Rough, but rough is fine for triage.\n- **The difficult reply.** Not \"write this for me\", \"here is my angry draft; keep every point but remove everything I would regret.\" You keep the substance, lose the damage.\n\n## Two rules I would not bend\n\n**Never let it send.** Auto-reply and auto-send is where quiet disasters live: a wrong commitment, a wrong price, an apology for something you did not do. Draft, always. Send, never.\n\n**Do not paste a whole thread into a public chatbot without thinking.** Email threads carry other people's names, salaries, complaints, medical details and contract terms. You did not write most of it and it is not yours to share. That is the next lesson, and it is the one that matters most."},"check":{"question":"A team lead sets up an AI notetaker that joins every call and emails a full summary to all attendees automatically. Within a month there is a serious complaint. What most likely went wrong?","options":[{"text":"Something sensitive or misattributed was distributed before any human read it, and the automation removed the checkpoint where that gets caught","correct":true},{"text":"The summaries were too long, so people stopped reading them and missed their action items","correct":false,"tempting":"Identifies a real annoyance with automated notes, but ignored summaries produce drift, not complaints."},{"text":"The notetaker's transcription accuracy was too low for meetings with mixed accents","correct":false,"tempting":"Transcription error is the underlying mechanism, but the complaint comes from unreviewed distribution: accurate transcripts of sensitive discussion cause the same problem."},{"text":"Attendees were not told a bot would be present, which is a consent failure","correct":false,"tempting":"A genuine and common failure, but consent was arguably handled by the visible bot joining the call: the harm here is what went out unchecked."}]},"prev":{"slug":"charts-that-say-what-you-meant","title":"Charts, and the choices you did not make","url":"https://addaly.com/learn/ai-at-work/charts-that-say-what-you-meant"},"next":{"slug":"transcription-and-consent","title":"Recording, transcription and the consent question","url":"https://addaly.com/learn/ai-at-work/transcription-and-consent"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Meetings and email', in AI at Work. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-at-work/meetings-and-the-inbox (Accessed: 31 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Meetings and email","courseTitle":"AI at Work","module":{"index":4,"slug":"working-from-your-own-material","title":"Working from your own material"},"answer":null,"keyIdea":"From a meeting, extract decisions, owners and deadlines, never the discussion, and never let AI hit send.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"p","runs":[{"text":"Meetings and email are where most professionals lose most of their week. They are also where AI help is most oversold, so let us be precise about what works."}],"lede":true},{"t":"h2","id":"meeting-notes-the-recording-is-not-the-point","runs":[{"text":"Meeting notes: the recording is not the point"}]},{"t":"p","runs":[{"text":"Automatic transcription is now good in most major languages and cheap or free in most video tools. That is not the win. A transcript of a 60-minute meeting is 9,000 words nobody will read. You have converted an hour you cannot get back into a document you will not open."}]},{"t":"p","runs":[{"text":"The win is the layer above it. From a transcript, ask for exactly three things:"}]},{"t":"ol","start":1,"items":[[{"text":"Decisions.","b":true},{"text":" What was decided, and by whom. Not what was discussed: decided."}],[{"text":"Actions.","b":true},{"text":" Who owes what, by when. Name, task, date. If a date was never said, write “no date agreed” rather than inventing one."}],[{"text":"Open questions.","b":true},{"text":" What was raised and not resolved."}]]},{"t":"p","runs":[{"text":"That structure works for a hospital handover, a school department meeting, a client call, a municipal committee. Notice what it excludes: the discussion. Almost nobody needs the discussion. They need to know what they now owe."}]},{"t":"p","runs":[{"text":"Then read it. Attributions get scrambled: transcripts confuse speakers, especially on bad audio, in accented English, or when people talk over each other. “Priya agreed to cover the shortfall” when it was actually Rahul is not a small error; it is the kind that turns into a dispute."}]},{"t":"h2","id":"before-you-record-ask","runs":[{"text":"Before you record: ask"}]},{"t":"p","runs":[{"text":"Recording and transcribing a meeting is not a neutral act, and consent rules vary widely by country and by state. Some places require every participant’s consent. Some workplaces forbid it outright. And beyond the law, there is the plain fact that people speak differently when a machine is listening: the half-formed idea, the honest doubt, the thing said carefully off the record. Say at the top of the call that you are recording, and say where the notes will go. If someone objects, take notes by hand."}]},{"t":"p","runs":[{"text":"Do not record a meeting where clients, patients, students, or personal grievances are discussed unless your organisation has explicitly approved the tool for that. See the next lesson."}]},{"t":"h2","id":"email-triage-over-composition","runs":[{"text":"Email: triage over composition"}]},{"t":"p","runs":[{"text":"Everyone reaches for AI to "},{"text":"write","i":true},{"text":" emails. The bigger gain is upstream."}]},{"t":"ul","items":[[{"text":"Reading a hostile or confusing thread.","b":true},{"text":" “Here is a 14-message chain. What is actually being asked of me, and what has already been agreed?” This is compression on something painful, and it works well."}],[{"text":"Deciding what needs you.","b":true},{"text":" “Sort these subject lines into: needs a decision from me, needs a reply, needs nothing.” Rough, but rough is fine for triage."}],[{"text":"The difficult reply.","b":true},{"text":" Not “write this for me”, “here is my angry draft; keep every point but remove everything I would regret.” You keep the substance, lose the damage."}]]},{"t":"h2","id":"two-rules-i-would-not-bend","runs":[{"text":"Two rules I would not bend"}]},{"t":"p","runs":[{"text":"Never let it send.","b":true},{"text":" Auto-reply and auto-send is where quiet disasters live: a wrong commitment, a wrong price, an apology for something you did not do. Draft, always. Send, never."}]},{"t":"p","runs":[{"text":"Do not paste a whole thread into a public chatbot without thinking.","b":true},{"text":" Email threads carry other people’s names, salaries, complaints, medical details and contract terms. You did not write most of it and it is not yours to share. That is the next lesson, and it is the one that matters most."}]}],"check":{"question":[{"text":"A team lead sets up an AI notetaker that joins every call and emails a full summary to all attendees automatically. Within a month there is a serious complaint. What most likely went wrong?"}],"options":[{"text":[{"text":"Something sensitive or misattributed was distributed before any human read it, and the automation removed the checkpoint where that gets caught"}],"correct":true,"tempting":null},{"text":[{"text":"The summaries were too long, so people stopped reading them and missed their action items"}],"correct":false,"tempting":[{"text":"Identifies a real annoyance with automated notes, but ignored summaries produce drift, not complaints."}]},{"text":[{"text":"The notetaker’s transcription accuracy was too low for meetings with mixed accents"}],"correct":false,"tempting":[{"text":"Transcription error is the underlying mechanism, but the complaint comes from unreviewed distribution: accurate transcripts of sensitive discussion cause the same problem."}]},{"text":[{"text":"Attendees were not told a bot would be present, which is a consent failure"}],"correct":false,"tempting":[{"text":"A genuine and common failure, but consent was arguably handled by the visible bot joining the call: the harm here is what went out unchecked."}]}],"order":[3,2,0,1],"why":null},"prev":{"slug":"charts-that-say-what-you-meant","title":"Charts, and the choices you did not make","minutes":8},"next":{"slug":"transcription-and-consent","title":"Recording, transcription and the consent question","minutes":9}}}