AI Extraction - Let SeaMeet Do the Note-Taking
Chapter 28: AI Extraction — Let SeaMeet Do the Note-Taking
Imagine hiring an assistant who sits in on every meeting, writes the summary, pulls out the action items, flags the key decisions, and timestamps every topic change—while you just focus on the conversation. That's exactly what AI Extraction does. Point it at any recording in your library and it returns five structured outputs: a full transcript, a concise summary, action items, key decisions, and timestamped chapters.
No typing. No rewinding. No guessing what was decided.
Chapter Objectives
After reading this chapter, you will be able to:
- Understand what AI Extraction produces and when to use it
- Set up prerequisites
- Generate your first extraction
- Navigate and use each of the five output tabs
- Use screenshot OCR to auto-extract text and copy it to your clipboard
- Regenerate results and understand how previous versions are preserved
- Export results in multiple formats
- Troubleshoot common extraction problems
What Is AI Extraction?
AI Extraction is a post-recording analysis process. After your recording is complete, you trigger it with a single click. The AI engine processes the audio and returns five structured outputs:
| Tab | What it contains |
|---|---|
| Transcript | Timestamped, word-for-word text of everything said |
| Summary | A concise narrative of what was discussed |
| Actions | Tasks identified from the conversation, with assignees and due dates where mentioned |
| Decisions | Key choices that were made during the meeting |
| Chapters | Time-indexed topic segments — like chapters in a book |
Think of it like this: The recording is the raw footage. AI Extraction is the editor who watches it and turns it into something structured and usable.
AI Extraction works on any recording in your library—whether or not you used live transcription during the recording. If live transcription ran, Extraction can still be a useful second pass, producing summaries and structured outputs that live transcription alone doesn't provide.
Before You Begin
Requirements
- AI Features enabled: Settings → AI → toggle on
- API key configured: Settings → AI → API Key field (green checkmark = valid)
- Active internet connection during extraction
- A recording in your library — any audio or video file SeaMeet recorded or imported
How long does extraction take?
Processing time scales with recording length:
| Recording length | Approximate time |
|---|---|
| Under 15 minutes | 20–40 seconds |
| 15–60 minutes | 1–3 minutes |
| 1–2 hours | 3–6 minutes |
| Over 2 hours | 5–10 minutes |
These are estimates. Actual time depends on server load and your internet speed.
Generating Your First Extraction
-
Open a recording from your Recording Library
- What you see: The recording detail panel opens on the right side, showing the player and recording info.
-
Click the AI Insights tab in the detail panel
- What you see: The AI Insights panel appears with a "Generate Summary" prompt if no extraction has been run yet. A ✨ icon marks the button.
-
Click ✨ Generate Summary
- What you see: The button label changes to a spinner. A progress bar or status message ("Analysing audio…", "Generating summary…") updates as the AI works through the content.
-
Wait for results
- What you see: When complete, the five tabs populate with content. The tab bar at the top of the AI Insights panel shows: Transcript · Summary · Actions · Decisions · Chapters.
-
Browse the tabs (see sections below for what to look for in each)
Understanding Each Tab
Transcript
The Transcript tab shows a complete word-for-word text of the recording, divided into segments with timestamps and speaker labels.
┌──────────────────────────────────────────────────────────┐
│ Transcript │ Summary │ Actions │ Decisions │ Chapters │
├──────────────────────────────────────────────────────────┤
│ Speaker 1 ► 0:00:08 │
│ "Good morning everyone. Let's get started with │
│ the Q3 budget review." │
│ │
│ Speaker 2 ► 0:00:21 │
│ "Thanks. I've prepared the slides. Should we start │
│ with revenue or expenses?" │
│ │
│ Speaker 1 ► 0:00:30 │
│ "Revenue first. Alice, can you walk us through it?" │
└──────────────────────────────────────────────────────────┘
Clicking a timestamp (the ► 0:00:08 link next to each segment) jumps the audio/video player to that exact moment. This makes it fast to verify a quote or find a specific discussion without scrubbing manually.
Speaker labels default to "Speaker 1", "Speaker 2", etc. You can rename these—see Chapter 29.
Summary
The Summary tab contains a concise narrative of what happened in the recording. It is not a bullet list of transcribed words—it's a synthesised overview that captures the main points discussed.
Example (for a 45-minute budget meeting):
The team reviewed Q3 revenue performance, which came in 8% above forecast. Expense overruns in the marketing budget were identified as the main concern going into Q4. Alice proposed a 15% reduction in paid advertising spend, which was accepted pending final approval from finance. The group agreed to reconvene the following Tuesday to review the revised projections.
The summary is designed to be shareable—you can copy it into an email, a project management tool, or a document without editing.
Actions
The Actions tab lists tasks that were explicitly or implicitly assigned during the recording.
┌──────────────────────────────────────────────────────────┐
│ Actions │
├──────────────────────────────────────────────────────────┤
│ □ Reduce paid advertising spend by 15% │
│ Assigned to: Alice · Due: Before Tuesday meeting │
│ ► 0:32:14 (in recording) │
│ │
│ □ Send revised Q4 projections to the team │
│ Assigned to: Finance team · Due: Friday │
│ ► 0:41:08 (in recording) │
│ │
│ □ Schedule follow-up meeting for Tuesday │
│ Assigned to: (unspecified) · No deadline mentioned │
│ ► 0:43:55 (in recording) │
└──────────────────────────────────────────────────────────┘
Each action includes:
- The task — what needs to be done
- Assignee — who was asked to do it (if mentioned)
- Due date — when it's needed (if mentioned)
- Timestamp link — jump to where in the recording this was discussed
Note: The AI extracts actions from what was said. If something was discussed but no clear assignment was made, it may or may not appear here. Always verify against the recording or transcript.
Decisions
The Decisions tab lists the key choices and agreements reached during the recording.
┌──────────────────────────────────────────────────────────┐
│ Decisions │
├──────────────────────────────────────────────────────────┤
│ ✓ Approved 15% reduction in Q4 paid advertising spend │
│ ► 0:32:14 │
│ │
│ ✓ Q3 revenue performance accepted as final (no recount)│
│ ► 0:12:45 │
│ │
│ ✓ Tuesday follow-up meeting confirmed │
│ ► 0:43:55 │
└──────────────────────────────────────────────────────────┘
Decisions are distinct from actions: a decision is what was agreed, while an action is what someone must do as a result.
Chapters
The Chapters tab breaks the recording into time-indexed topic segments, like chapters in a book or chapters on a DVD.
┌──────────────────────────────────────────────────────────┐
│ Chapters │
├──────────────────────────────────────────────────────────┤
│ ► 0:00:00 Introduction & housekeeping 3 min │
│ ► 0:03:10 Q3 Revenue Review 12 min │
│ ► 0:15:22 Expense analysis — Marketing 10 min │
│ ► 0:25:41 Q4 Budget Proposals 18 min │
│ ► 0:43:30 Wrap-up & next steps 4 min │
└──────────────────────────────────────────────────────────┘
Clicking any chapter jumps the player to that topic's start time. If you want to re-listen to just the budget proposals discussion, click ► 0:25:41.
Chapters are particularly useful for long recordings where you want to find a specific topic quickly.
AI-Suggested Filename Rename
After AI Extraction completes, the AI may suggest a more descriptive filename for your recording based on the content of the meeting. A banner appears at the top of the AI Insights panel:
┌──────────────────────────────────────────────────────────────┐
│ New file name: Q3 Budget Review — Oct 15 Team Meeting │
│ [ Rename ] │
└──────────────────────────────────────────────────────────────┘
To apply the suggested name:
- Click Rename — the recording file and its library entry are renamed immediately.
- The suggestion banner disappears after renaming.
To keep your original name:
- Ignore the banner — it does not auto-apply and disappears if you navigate away.
Note: The suggestion is based on the meeting content. For a generic or short recording, the suggested name may be similar to your original. You always have final say.
Gemini Model Selection
SeaMeet fetches the available Gemini models from Google's API dynamically — the model list stays current as Google releases new versions.
Which model is used:
- SeaMeet automatically filters for models that support text, vision, and audio (required for meeting analysis)
- Only capable, current models are shown — older or limited models are excluded
- The list is sorted with the latest/most capable models first
You can see and change the active model in Settings → AI → Model. The default is the most capable available model. There is rarely a reason to change this unless you need to reduce costs or work around a specific model limitation.
Screenshot OCR — Auto-Extract Text and Copy to Clipboard
AI Extraction isn't limited to audio recordings. When you take a screenshot with the "Extract text (OCR) and copy to clipboard" option enabled, SeaMeet automatically runs OCR on the captured image and copies the extracted text to your clipboard.
How it works
- Before capture: Check the "Extract text (OCR) and copy to clipboard" checkbox in the Capture Hub or the Region Selector overlay
- Take the screenshot (Full Screen or Area Capture)
- Wait ~15–30 seconds — the AI pipeline processes the image in the background
- Text is copied to clipboard — a notification confirms "Extracted text copied to clipboard"
- Paste anywhere — Cmd+V / Ctrl+V pastes the extracted text
The OCR pipeline runs entirely in the background. You can continue using SeaMeet or switch to other apps while it processes. The clipboard is updated automatically when the extraction finishes.
Where to find the setting
The OCR auto-copy option appears in three places:
| Location | What it controls |
|---|---|
| Capture Hub (Screenshot panel) | Per-capture toggle — checkbox below capture mode selector |
| Region Selector overlay | Per-capture toggle — checkbox near "Capture Screen or Press Enter" |
| Settings → AI Extraction & Summary | Default on/off — "Auto-extract text (OCR) on screenshot" toggle |
All three locations stay in sync. Toggling in any one location updates the others.
Requirements
- Gemini API key configured (or active subscription with managed key)
- AI Features enabled in Settings
If no API key is available, the checkbox is greyed out and displays "(API key required, subscribe here)" with a link to the subscription portal.
What gets extracted
The OCR pipeline extracts all visible text from the screenshot — including:
- UI labels, menus, and buttons
- Document and web page content
- Error messages and dialog text
- Code and terminal output
- Handwritten text (if legible)
The extracted text is also saved in the screenshot's AI Extraction panel under the Text (OCR) tab, so you can review or re-copy it later.
Regenerating
If you're not satisfied with the results, or if you want a fresh analysis after renaming speakers, you can regenerate.
-
Click ⟳ (regenerate) button in the tab bar
- What you see: The button spins. A brief "Processing…" notice replaces the content.
-
Wait (same duration as the initial generation)
- What you see: All five tabs refresh with new results.
-
Review the new results
Important: Before overwriting your results, SeaMeet automatically archives the previous extraction as a version snapshot. Nothing is ever lost. You can browse previous versions at any time—see Chapter 29, Part B.
Exporting
The tab bar includes export buttons that let you take the results outside SeaMeet:
[Transcript] [Summary] [Actions] [Decisions] [Chapters] · [MD] [SRT] [JSON] [⟳]
| Button | Format | Best for |
|---|---|---|
| MD | Markdown | Notes apps (Notion, Obsidian), documentation |
| SRT | SubRip subtitle | Adding captions to video files (iMovie, Premiere, YouTube) |
| JSON | JSON data | Developers, importing into other tools |
Which tab is active determines what gets exported:
- On the Summary tab → MD exports the summary text
- On the Transcript tab → SRT exports the timestamped transcript as subtitles
- On the Actions tab → MD or JSON exports the action item list
Tip: Export before regenerating if you want a standalone copy of the current results—though this is not strictly necessary since version history preserves past results automatically (see Chapter 29).
Troubleshooting
"Extraction failed"
Symptom: The spinner stops and an error message appears: "Extraction failed" or "Something went wrong."
Try these in order:
- Check your internet connection — try loading a web page
- Verify your API key is still valid (Settings → AI → look for green checkmark)
- Wait 2–3 minutes and try again — the AI service may be temporarily overloaded
- If the recording is very long (2+ hours), try splitting it or allow more time for processing
"Summary seems incomplete"
Symptom: The summary is short, misses key topics, or seems generic.
Possible causes:
- Audio quality was poor — transcription errors lead to extraction errors. Check the Transcript tab for garbled segments.
- Recording captured too little voice — if system audio dominated (music, video playback), the AI may have had little speech to work with
- Very short recording — recordings under 2 minutes produce limited summaries
What to try:
- Re-run the extraction with ⟳ after renaming speakers and reviewing the transcript
- Improve microphone placement for future recordings
"Action items are blank"
Symptom: The Actions tab is empty even after a meeting that clearly had tasks assigned.
What's happening: The AI extracts actions from explicit language ("Can you…", "I'll handle…", "Let's make sure…"). Implicit or informal commitments may be missed.
What to try:
- Check the Transcript tab — if the conversation was informal, the AI may not have identified clear action language
- This is expected behaviour for casual, exploratory conversations rather than structured meetings
Best Practices
Higher quality audio → better results Every extraction step depends on transcription accuracy. A recording made with a quality microphone in a quiet room will produce dramatically better results than one with background noise and echo. Invest in a decent headset microphone for regular meeting recordings.
Record both microphone and system audio If you're recording a virtual meeting (Zoom, Teams, Google Meet), use "Both" audio mode so system audio (other participants) is captured alongside your microphone. An extraction from a one-sided conversation produces one-sided results.
Rename speakers before regenerating Speaker names appear in the Summary and Actions tabs. Renaming "Speaker 1" to "Alice" before you click ⟳ produces a much more readable summary. See Chapter 29 for how to rename speakers.
Longer recordings produce richer summaries A 5-minute meeting will produce a short summary. A 45-minute strategic discussion will produce a detailed, structured summary with clear sections. AI Extraction is most valuable for substantive recordings.
Use Chapters for navigation For recordings over 20 minutes, always check the Chapters tab first. It gives you a map of the discussion, making it far faster to jump to the section you care about.
Quick Reference
┌────────────────────────────────────────────────────────────┐
│ AI EXTRACTION │
│ Quick Reference │
├────────────────────────────────────────────────────────────┤
│ Trigger │ AI Insights tab → ✨ Generate Summary │
│ Regenerate │ ⟳ button in tab bar │
│ Previous results │ Safely archived — see Chapter 29 │
├────────────────────────────────────────────────────────────┤
│ Transcript │ Full text + timestamps, click to jump │
│ Summary │ Narrative overview of discussion │
│ Actions │ Tasks + assignees + due dates │
│ Decisions │ Agreements and choices made │
│ Chapters │ Time-indexed topic map, click to jump │
├────────────────────────────────────────────────────────────┤
│ Export: MD │ Markdown — for notes apps │
│ Export: SRT │ Subtitles — for video editors │
│ Export: JSON │ Structured data — for developers │
├────────────────────────────────────────────────────────────┤
│ Screenshot OCR │ Auto-extract text + copy to clipboard │
│ OCR setting │ Capture Hub checkbox or Settings → AI │
├────────────────────────────────────────────────────────────┤
│ Requires │ Internet + AI Features on + API key │
│ Works on │ Any recording or screenshot │
└────────────────────────────────────────────────────────────┘
Last updated: 2026-03-27
← Chapter 27: Live Transcription | Chapter 29: Editing AI Results →
Published: