# Capture every voice memo into actions and notes: the AI memo to action build Source: https://getstartedwithai.net/tutorials/ai-voice-memo-to-action-build Site: Get Started With AI Tagline: Build AI automations that actually run your business. Audience: solo professionals, freelancers, consultants and AI implementers in training Contact: hello@getstartedwithai.co.uk This Markdown companion is generated from the prerendered page for AI crawlers. # Capture every voice memo into actions and notes: the AI memo to action build The voice memo is the fastest capture and the slowest output. Most people record one and never act on it. The build below closes that gap. Record the memo on your phone, the audio runs through Whisper, Claude parses the intent into structured actions, and the actions land in Notion or your task manager. Calendar invites get drafted, emails get drafted, follow-up tasks get queued. About 60 minutes to set up; pays back in the first week. The realistic time saving is 30 minutes per memo over manual processing. ## When this build pays back - Founders or owner-operators who think out loud while driving or walking. - Sales or service professionals who debrief clients on the way back from meetings. - Researchers and journalists who capture stray ideas before they evaporate. - Anyone who records 3 or more voice memos a week. ## The tooling stack - Phone voice recorder or AudioPen at £3 a month. - Whisper for transcription (OpenAI API at about $0.006 a minute or self-hosted). - Claude Pro at £16 a month: parses the memo into structured actions. - Notion, Asana, Linear or Todoist for the task manager. - Optional: Zapier or Make to wire the steps automatically. ## Step 1: Build the parsing project in Claude (30 minutes) 1. Open Claude Pro, Projects, New Project. Name it Voice Memo Actions. 2. Custom Instructions: paste the parsing rules. Extract: tasks (with assignee and due date), calendar invites, email drafts, ideas, follow-ups. Use UK English. 3. Project Knowledge: upload your standard task tags, project list and the names of your collaborators. 4. Test with a known memo. Confirm the output is structured cleanly. ## Step 2: Capture the memo (1 minute) 1. Open the phone recorder. 2. Talk through whatever is in your head: tasks, ideas, replies, decisions. 3. Aim for 2 to 5 minutes per memo. 4. Keep it natural. Stop trying to format the memo as you record. ## Step 3: Run the audio through Whisper (1 minute) 1. Upload the .m4a or .mp3 to Whisper via the OpenAI API or a local install. 2. Receive the transcript. 3. For automated runs, use a Zap or a Make scenario triggered by the file landing in a folder. ## Step 4: Parse the transcript into actions (2 minutes) 1. Open the Voice Memo Project. 2. Paste the transcript with this prompt: "Parse this voice memo into structured actions. Tasks with assignee and due date, calendar invites with date and time, email drafts with recipient, ideas, follow-ups." 3. Claude returns the parsed output. 4. Read once. Mark anything that looks misparsed. ## Step 5: Pipe into the task manager (one-off setup, 30 minutes) 1. In Zapier or Make, build a flow: Whisper transcript triggers a Claude parse, Claude output writes to Notion or Asana. 2. Per task: title, assignee, due date, project tag. 3. Per email draft: write to Drafts in Gmail or Outlook. 4. Per calendar invite: write to Google Calendar or Outlook with the proposed time. 5. Test with a real memo end to end. ## Step 6: Review the output daily (10 minutes) - New tasks queue in Notion or Asana for review. - Email drafts sit in Drafts for review and send. - Calendar invites queue in your invites list, not in your accepted calendar. - Approve, edit or reject each one in 10 minutes. - Decline any item the build got wrong. Tune the project instructions if it gets the same thing wrong twice. ## What you ship A working voice-memo-to-action pipeline that turns a 5-minute memo into structured tasks, drafted emails and calendar invites in about 60 seconds of run time. Saves about 30 minutes per memo over manual processing. Most users record more memos once the friction drops, which compounds the saving. ## What to build next Once the memo pipeline runs, the matching layer is the SOP capture build: turn a verbal walkthrough into a documented procedure. Pair this with the SOP AI capture build and the AI newsletter publish weekly build. ## Related builds More tutorials in Productivity. Each one runs in 60 to 90 minutes and ships a finished asset. ### Turn scattered notes into a searchable knowledge base: 5 Notion AI builds that earn their fee 30 to 60 minutes. Solo Pro. [Open tutorial](/tutorials/notion-ai-solo-pro-5-setups) ### Reclaim 3 hours of meeting follow-up: the closed-loop note pipeline 30 to 60 minutes. Solo Pro / Implementer. [Open tutorial](/tutorials/ai-meeting-notes-pick-right-tool) ### Reply to email at twice the speed, in your own voice: a scaled email build 30 to 60 minutes. Solo Pro. [Open tutorial](/tutorials/ai-email-writing-scaled-system) ### Run a project without the chase: a practitioner project management stack 60 to 90 minutes. Solo Pro / Implementer. [Open tutorial](/tutorials/ai-for-project-management-stack) ## Questions people ask before they start ### What is the best transcription tool in 2026? Whisper still leads on accuracy in technical and accented audio. AssemblyAI is faster for streaming. For most voice memos, Whisper via the OpenAI API at about $0.006 a minute is the lowest-friction option. ### Will the parsing miss the actions I cared about? Sometimes. Tune the parsing prompt over the first week. Most users hit 90 percent accuracy after 5 memos because the project instructions sharpen on each correction. ### Is this safe with confidential information? On the Pro tier of Claude or ChatGPT, inputs are not used for training and the build is safe for ordinary commercial work. For regulated work in legal or healthcare or any heavily-regulated financial context, use enterprise tier with a processor agreement, or run a local Whisper plus a local LLM. ### Do I need Zapier to run this? No. The manual version works in about 5 minutes per memo: record, transcribe, paste into Claude, copy actions to your task manager. Zapier collapses that to about 60 seconds, which is what makes the build worth keeping. ### How is this different from AudioPen? AudioPen does the transcription and a single rewrite. The build above structures the output into tasks, emails and calendar invites, and pipes them into your existing tools. AudioPen is the front end; this build is the back end.