AI meeting notes from a recording: transcript, chapters and next steps

Most AI note takers join a live call as a bot and send notes afterward. This guide is for the other case: you already have a recording. Maybe it's a walkthrough you made for a client, a demo you recorded instead of a meeting, or a screen-shared review of a draft. You want a summary, chapters someone can jump to, and a list of what was promised, without rewatching it.

What good meeting notes from a recording contain

• A summary of 2 to 4 sentences: what the recording is about and what was decided.
• Chapters with start times, so the client can jump to "pricing page" without scrubbing through the video.
• Next steps, each with an owner and a date if one was said: "You: send the revised homepage by Friday." "Client: send logo files."
• Decisions and open questions, kept apart. A decision is done; a question needs an answer before work continues.
• Nothing that wasn't said. A summary that adds a deadline nobody mentioned is worse than no summary.

Step by step: from recording to notes

1. Get consent before you record anyone else's voice. Say it at the start ("I'm recording this so I can send you notes") and get a yes on the recording. The legal side is below.
2. Record clean audio. Use a microphone close to you, close noisy apps, and if you're recording a browser tab that plays the other person's voice, share the tab's audio too.
3. Transcribe the audio with a speech-to-text service that returns times per segment. Times are what make chapters possible.
4. Summarize from the transcript only. Ask for the summary, chapters at times that exist in the transcript, and next steps as stated. Tell the model not to add anything that isn't in the text.
5. Check the next steps against the transcript: names, amounts, dates and who owns each one.
6. Send the recap, put the next steps on your task board, and turn any new work into a quote while it's fresh.

Recording consent: one-party and all-party states

Under federal law, 18 U.S.C. § 2511(2)(d), a person who is a party to a conversation, or who has the consent of one party, may record it, unless the recording is made to commit a crime or a tort. Several states go further. California Penal Code § 632 makes it illegal to record a confidential communication without the consent of all parties, by phone or in person, with fines up to $2,500 per violation in the base case. Other states have their own all-party rules.

When the people on a call are in different states, the safe default is the strictest rule: tell everyone you're recording and get their yes on the recording itself. This is general information, not legal advice. If recording is central to your work (coaching, interviews, research), check the law of each state where your clients are, or ask a lawyer. A recording can also hold other people's voices and personal details, so share transcripts only with people who should have them.

In SoloWOS: Transcribe and summarize

SoloWOS works on recordings you make with Screen Recording, which records your screen with your microphone and, in Chrome, the audio of a tab when you check "Share audio". It doesn't join live calls as a bot. Recording length depends on the plan: up to 3 minutes on Basic, 5 on Solo and 15 on Team. That suits client walkthroughs, demos and async updates more than an hour-long meeting.

1. Open the recording and click Transcribe and summarize. Owners, admins and members with access to Screen Recording see it.
2. SoloWOS shows how many transcription minutes it will use (the length rounded up to the next minute) and how many you have left this month. If there aren't enough, it says so before anything starts.
3. The recording joins a queue. Only its audio track is sent to the transcription provider (Deepgram or OpenAI) when it can be separated; the transcript is then sent to Anthropic to write a suggested title, a 2 to 4 sentence summary, chapters and the next steps mentioned. You can leave the page; you get a notification when it's done.
4. The panel shows the summary, a Use suggested title button, chapters that jump the video to that time, and the next steps. Each next step has Add to board, which adds it as a task, and Make a quote from this, which opens the assistant with the text ready to edit.
5. Below that, the full transcript has search and Copy transcript, and the player gets captions.

Minutes only count when the transcription works, and the summary doesn't use any other quota. If the recording has no speech, it says "We didn't hear any words in this recording." If only the summary fails, the transcript stays and you can try the summary again without using more minutes. Chapters only use times that exist in the transcript, and there are at least two of them. The transcript and summary are saved with the recording and only your workspace sees them; deleting the recording deletes them. On the shared page, viewers can turn on captions (off by default) and see the chapters. The summary is never public.

What to check before you send the recap

• Names. Speech-to-text often mishears names of people, brands and products. Fix them in your recap.
• Numbers and dates. "Fifteen" and "fifty" sound alike. Confirm any amount or deadline against the audio before it becomes a promise.
• Owners. A next step without an owner is a wish. Add "you" or "client" to each one.
• Jokes and asides. A model can turn "we could always add a blog, ha" into a next step. Delete what nobody meant.
• Sensitive details. If the recording shows a client's account numbers or health information, don't paste that part of the transcript into an email.

If you send updates by video often, presenting design work by video and async video client updates cover how to structure them, and the screen recording privacy checklist covers what not to show. The transcript features are part of the SoloWOS AI assistant; more posts are in AI for freelancers.