Chapter 33: AI Summary Templates

In Chapter 32 you learned how AI Extraction turns a recording into a summary, transcript, action items, decisions, and chapters. That chapter covered what the AI produces. This chapter covers how it's shaped — the AI Summary Templates system that decides whether your recording becomes a chronological court record, a SOAP medical note, a Q&A deposition transcript, or a punchy podcast show notes page.

Without a template, every recording gets the same generic "Meeting Minutes" treatment — fine for a team standup, useless for a clinical consultation where you need Subjective · Objective · Assessment · Plan. Templates control the section headings, the writing style, and the constraints the AI follows. Pick the right one and the same 30-minute audio file becomes the right document for the job. Even better: SeaMeet can pick the right template for you automatically.

Chapter Objectives

After reading this chapter, you will be able to:

Understand what a summary template changes in the AI output
Browse the built-in library (48 templates across 8 categories)
Use the auto-pick classifier and understand when it auto-applies vs. just suggests
Pin a template per recording via the picker
Toggle auto-suggest from anywhere in the app
Re-render an existing summary with a different template via Regenerate
Configure auto-detection from Settings → AI Features → AI Summary Templates

What Is a Summary Template?

A template is a structured instruction set the AI follows when writing a summary. It declares:

What sections to include (and in what order)
What writing style to use (clinical, legal, conversational, terse)
What to omit if missing rather than infer
What constraints to apply (e.g., "preserve the witness's actual phrasing on contested points")

Side-by-side example — same recording, two templates:

Without a template (generic Meeting Minutes): "The witness was asked about the events of June 14. They described arriving at the warehouse around 9 PM and noticing the back door was open. Counsel objected to the line of questioning twice."

With the Deposition template: "## Witness identification — John Reyes, warehouse supervisor, deposed 2026-06-04. ## Substance of Q&A — Q: Where were you at 9 PM? A: At the loading dock. ## Objections — Form (twice). ## Key admissions — Witness confirmed back door was unsecured on arrival."

Same transcript, same AI, two very different documents. That's what templates do.

The Template Library

SeaMeet ships 48 built-in templates across 8 categories. The list is auto-discovered at startup — drop a new .md file into the templates folder and it appears in the picker without a code change.

Team & Management

Generic meetings, 1:1s, brainstorms, exec readouts.

Meeting Minutes (general) — the default fallback when nothing else fits
Executive Summary — topic-by-topic executive readout with decisions and actions
Issue / Decision / Action Items — surfaced issues, decisions, owned next actions
1:1 / Interview — themes, sentiment, growth, follow-ups
Brainstorm — ideas, threads, open questions
Team Weekly — last week, this week, and risks for a weekly team sync

Project & Agile

For engineering teams running sprints, postmortems, and delivery reviews.

Standup / Status — per-person updates, blockers, next
Sprint Retrospective — what went well, what didn't, concrete improvements
Project Progress — topic-by-topic project status with decisions and actions
Incident & Bug Review — timeline, impact, root cause, action items
Requirement Review — requirements clarified, open questions, action items
Execution Planning — plan, owners and timeline, risks for a delivery

Sales & Client

For deal motions — discovery through close.

Discovery / Needs Assessment — discovery, objections, next steps, decision-makers
BANT · MEDDIC Qualification — qualify the deal across BANT and MEDDIC criteria
Sales Customer Visit — discussion, next steps, decision-makers from a visit
Pipeline / Account Review — account status, risks, and next steps
Business Negotiation — terms discussed, concessions, agreements, next steps
Client Follow-Up — a ready-to-send follow-up email summarizing the call

HR & Recruitment

For hiring loops, performance reviews, and people operations.

Structured Interview Notes — competency scorecard with evidence and ratings
Candidate Evaluation — fit vs. role, strengths, gaps, recommendation
Job Interview Report — background, strengths, concerns, hire recommendation
Compensation & Performance — performance review with comp discussion
HR / Supervisor Consultation — concern raised, discussion, follow-ups
Exit Interview — reasons for leaving, feedback, follow-ups

Education & Training

For teachers, lecturers, and learners.

Lecture / Talk — thesis, key points, questions, references
Course Summary / Study Notes — key concepts and takeaways for review
Teacher Handout — topic-by-topic handout for students
Teaching Evaluation — strengths, areas to improve, recommendation
Research Seminar / Journal Club — thesis, methods, findings, critiques
Q&A Extraction — questions asked and answers given

Medical & Healthcare

Templates with the Specialized badge — they auto-append a disclaimer reminding the reader the output is AI-assisted and not a substitute for clinical judgment.

SOAP Note — Subjective, Objective, Assessment, Plan
Detailed Clinical Case — history, findings, assessment, plan
Medical Consultation — clinical encounter with history and plan
Outpatient Report — reason for visit, findings, plan
Psychotherapy Record — presentation, intervention, response, plan
Multidisciplinary Consultation — specialist input, consensus, shared plan

Legal & Compliance

Also marked Specialized — disclaimers auto-append.

Deposition / Interrogation — structured Q&A with objections and admissions
Court Record — chronological record of court proceedings
Legal Arbitration / Mediation — positions, agreements, compliance
Law Enforcement / Incident Record — incident, statements, actions taken
Client Consultation — facts, options, fee arrangement, next steps
Partners' Meeting — case assignments, firm matters, decisions

Events & Media

For panels, podcasts, press conferences, and investor calls.

Keynote Summary — summary and key takeaways from a keynote
Forum / Panel Summary — panel summary with points of agreement and debate
Press Conference Analysis — announcements made and the Q&A that followed
Investor / Earnings Call — earnings readout with guidance and concerns
Podcast Golden Quotes — extracted quotes with speaker attribution
Podcast Show Notes — topics, timestamps, and links for an episode

The Specialized badge. Medical and Legal templates show a small shield icon in the picker. When you pick one, the AI output ends with a one-paragraph disclaimer making it explicit that the document is AI-generated and not a substitute for professional review.

Auto-Pick: The Template Classifier

You don't have to memorize all 48 templates. Auto-pick reads your transcript and chooses the best fit.

How it works

After AI Extraction finishes transcribing, a tiny classifier prompt runs as a second pass:

SeaMeet builds a candidate list — one slug: description line per installed template.
It sends the first ~6,000 characters of the transcript plus that list to the AI.
The AI returns a single template slug and a confidence rating: high, medium, or low.
SeaMeet applies a fallback policy:
- High confidence → auto-apply. The summary is generated with the suggested template immediately.
- Medium confidence → pre-select but don't apply. You see an "Auto-detected: " chip; one click confirms.
- Low confidence → fall back to Meeting Minutes. No chip shown.
- Unknown slug (model hallucinated a name that isn't installed) → fall back to Meeting Minutes.

The excerpt cap exists for a reason: a 3-hour transcript wouldn't help the classifier — the meeting type is almost always clear from the first few minutes.

When auto-pick runs

Auto-pick runs in two places:

Per-recording (during AI Extraction) — when you generate the summary for a recording, the classifier picks the template before the summary is written. The choice is persisted on the recording's manifest so even months later, opening the recording surfaces the same template chip without re-running the classifier.
Per-note (during note synthesis) — when you synthesize a note that pulls in recordings, the same classifier picks the template for the synthesis.

The auto-detected chip

After extraction completes, look at the top of the AI Summary tab:

┌────────────────────────────────────────────────────────┐
│  AI Summary  │  Transcript  │  Actions  │  Decisions   │
├────────────────────────────────────────────────────────┤
│  [Regenerate ▼]   [✨ Auto-detected: SOAP Note]        │
│                                                        │
│  ## Subjective                                         │
│  Patient reports headache for 3 days…                 │
└────────────────────────────────────────────────────────┘

Click the chip to open the picker pre-selected on the suggested template — handy when you want to confirm the choice or swap it for a sibling.

The chip is hidden when the suggestion is general (the low-confidence fallback) — showing "Auto-detected: Meeting Minutes" would be a no-op.

The Picker UI — Suggested Zone + Library

Open the picker via the Regenerate split-button (more on that below) or by clicking the auto-detected chip. The picker has two zones.

Suggested zone (top). If a template was auto-detected for this recording, it appears with an ✨ icon under a "Suggested" header. The card is pre-selected so a single click on Regenerate confirms it.

Full library (below). Every installed template laid out in a responsive grid (1 to 4 columns depending on window width), grouped by category in fixed order: Team & Management, Project & Agile, Sales & Client, HR & Recruitment, Education & Training, Medical & Healthcare, Legal & Compliance, Events & Media, and Other (user templates). A search box at the top filters by name and description. Type "deposition" or "earnings" or "soap" and the list narrows instantly. Specialized templates show a shield badge inline with the name.

┌────────────────────────────────────────────────────────────────┐
│  Regenerate Summary                                     [×]    │
├────────────────────────────────────────────────────────────────┤
│  🔍 Search templates…                                          │
│                                                                │
│  ─ ✨ Suggested ─────────────────────────────────────────────  │
│   [ SOAP Note  🛡 ]   ← pre-selected, click to confirm        │
│    Subjective, Objective, Assessment, Plan                     │
│                                                                │
│  ─ 👥 Team & Management ────────────────────────────────────  │
│   [Meeting Minutes] [Exec Summary] [Issue/Dec/Action] [1:1 ]  │
│                                                                │
│  ─ 🚀 Project & Agile ──────────────────────────────────────  │
│   [Standup] [Retro] [Project Progress] [Incident & Bug ] …    │
│                                                                │
│  ─ 🩺 Medical & Healthcare ────────────────────────────────  │
│   [SOAP Note 🛡] [Clinical Case 🛡] [Medical Consult 🛡] …     │
│                                                                │
│  Or write custom instructions (overrides template): [______]   │
│  ☐ Save as a custom template                                   │
│                                                                │
│                              [ Cancel ]  [ Regenerate ]        │
└────────────────────────────────────────────────────────────────┘

The custom instructions textarea below the library is a power-user escape hatch — if you'd rather tell the AI exactly what you want for this one summary, you can override the template entirely.

Per-Recording Template — How It Persists

When the classifier picks a template (or you pick one manually), the slug is written into the recording's manifest. This means the auto-detected chip survives across sessions — open the recording weeks later and it's still there, no re-classification needed. And if you manually swapped the suggestion for a different template, that swap becomes the persisted choice. Override anytime by reopening the picker and re-running Regenerate; the summary is rewritten from the existing transcript (no re-transcription) and the new slug replaces the old one.

The Split-Button Regenerate (AI Summary tab)

At the top of the AI Summary tab, the Regenerate control is a split button: a primary pill plus a caret that opens a dropdown menu.

Primary click — re-renders the summary with the currently-applied template (the auto-detected one, or your last pick). Fast: seconds, not minutes — no re-transcription, just the summary call.
Caret menu — opens this menu:

┌──────────────────────────────────────┐
│  🎛  Choose template…                 │
│  🔄  Re-analyze recording             │
│  ─────────────────────────────────── │
│  ☑  Auto-suggest template            │
└──────────────────────────────────────┘

Menu item	What it does
Choose template…	Opens the picker (Suggested zone + full library).
Re-analyze recording	Runs the full pipeline from scratch: re-transcribes, re-classifies, re-summarizes. Slower; uses transcription quota. Use when the recording was edited or transcription quality was bad.
Auto-suggest template	Global toggle. When on, future syntheses run the classifier. When off, Meeting Minutes is used by default (skipping the classifier call entirely).

This same auto-suggest toggle appears in three places — the split menu here, the notes-side split menu, and Settings — and they all reflect the same global preference. Flip it from wherever you're working.

Settings → AI Features → AI Summary Templates

For a bird's-eye view, open Settings → AI Features → AI Summary Templates. The panel has two parts:

Auto-detect toggle (top)

☑  Auto-detect summary template
    Let the model pick the best template for each meeting. When off,
    Meeting Minutes is used by default and no classifier call is made.

Default: on. Turning it off makes every summary default to Meeting Minutes, saves the cost of the classifier round-trip, and skips the auto-detected chip on the AI Summary tab. You can still pick templates manually from the Regenerate picker.

Installed templates inventory (below)

A read-only grid showing every template currently installed, grouped by category, with name + description. Specialized templates show the shield badge. This is the easiest way to discover what's available without opening a recording — handy when you're deciding whether SeaMeet covers a meeting type you haven't recorded yet.

User templates (created via "Save as custom template" in the Regenerate modal) appear under the Other category with a user: slug prefix.

Regenerating with a Different Template

Already have a summary you don't love? Regenerate is a few clicks:

Open the recording in your library
Click the AI Summary tab
Click the caret next to the Regenerate button → Choose template…
Pick a template from the picker (or use the Suggested one at the top)
(Optional) Add custom instructions in the textarea — these override the template's defaults
Click Regenerate

The summary is rewritten in seconds. The previous version is automatically archived as a version snapshot before the new one is written — nothing is ever lost. You can browse previous versions from the version picker (see Chapter 34: Editing AI Results, Part B).

Tip: If the recording's transcript is solid but the wrong template was picked, always use "Choose template…" rather than "Re-analyze recording." Choose template is cheap (text-only) and fast (seconds). Re-analyze is expensive (it hits transcription quota) and slow (re-transcribing a 1-hour recording can take minutes).

MCP Exposure — Chat Agent Tools (Advanced)

For power users who use the in-app chat agent (the AI Agent panel, not the AI Summary tab), two MCP tools expose the templates system to the agent:

seameet_list_templates — returns the full installed library with slug, name, description, category, and specialized flag. Use when asking the agent "what summary types do you support?" or "which template would fit my Friday user-interview recordings?"
seameet_regenerate_summary — takes a recording's absolute filePath and a templateSlug (and an optional customPrompt) and re-renders the summary in place. After it runs, the recording's persisted summary is the new one — subsequent calls to seameet_get_artifact return the new version.

A typical agent conversation: "Find my June 4 customer call and regenerate it as a BANT/MEDDIC summary." The agent searches, confirms the bant-meddic slug, then calls seameet_regenerate_summary. The chip on the AI Summary tab updates automatically.

Troubleshooting

"The auto-detected template is wrong"

Symptom: The chip says "Auto-detected: Sprint Retrospective" but the meeting was a sales call.

Why: The classifier picks based on the first ~6,000 characters. If the call opened with team chitchat ("How was your sprint?") before getting into sales, the model latched onto the wrong signal.

Fix: Click the chip. The picker opens pre-selected on the wrong template — the full library is right there. Pick the correct one and hit Regenerate. Your pick is persisted.

"No chip appears after extraction"

Possible causes:

The classifier returned low confidence → fell back to Meeting Minutes silently (chip suppressed by design).
Auto-suggest template is off in Settings or the split-menu toggle.
The recording was processed by an older version of SeaMeet before templates shipped. Re-analyze to populate the suggestion field.

Fix: Check Settings → AI Features → AI Summary Templates. Turn the toggle on if needed, then Regenerate. If it's already on, you're in the low-confidence fallback — open the picker and choose manually.

"Template picker is empty"

Why: The IPC call to load templates failed silently — usually a corrupted install or a renderer that can't reach the main process.

Fix: Restart SeaMeet. If still empty, reinstall — built-in templates ship inside the app bundle.

Best Practices

Leave auto-detect on for mixed workflows. If you record a variety of meeting types, auto-detect saves you from picking every time. The classifier is wrong sometimes, but it's right far more often, and one click overrides it.

Turn auto-detect off for narrow workflows. If 100% of your recordings are the same type, turn auto-detect off and reach for Choose template… when you need an exception — you'll skip the classifier round-trip every time.

Specialized templates aren't legal/medical advice. Medical and Legal templates ship with a disclaimer in the output, but it doesn't replace professional judgment. Treat these summaries as draft notes that need a clinician's or counsel's review.

Custom prompts override the template. If you've picked a template but also typed something in "Or write custom instructions," the custom instructions take precedence. Use this when the template is almost right but needs a tweak.

Quick Reference

┌────────────────────────────────────────────────────────────────┐
│              AI SUMMARY TEMPLATES                              │
│              Quick Reference                                   │
├────────────────────────────────────────────────────────────────┤
│  Library size      │ 48 built-in templates, 8 categories       │
│  Auto-pick         │ Classifier picks a template per meeting   │
│  Confidence        │ high → auto-apply                         │
│                    │ medium → pre-select (chip)                │
│                    │ low/unknown → Meeting Minutes (no chip)   │
│  Persistence       │ Slug stored on recording manifest         │
├────────────────────────────────────────────────────────────────┤
│  Picker            │ Suggested zone + grouped library + search │
│  Pick a template   │ Regenerate ▼ → Choose template…           │
│  Reuse same        │ Regenerate (primary button)               │
│  Full re-analyze   │ Regenerate ▼ → Re-analyze recording       │
├────────────────────────────────────────────────────────────────┤
│  Settings panel    │ Settings → AI Features → AI Summary       │
│                    │ Templates                                 │
│  Auto-suggest      │ Toggle in: Settings, notes split-menu,    │
│                    │ AI Summary split-menu (all in sync)       │
├────────────────────────────────────────────────────────────────┤
│  Specialized badge │ Medical + Legal — disclaimer auto-appends │
│  Custom templates  │ "Save as custom template" in picker       │
│  Chat agent tools  │ seameet_list_templates                    │
│                    │ seameet_regenerate_summary                │
├────────────────────────────────────────────────────────────────┤
│  Versions          │ Previous summary archived on regenerate   │
│  Output language   │ Follows Settings → AI Extraction          │
└────────────────────────────────────────────────────────────────┘

Last updated: 2026-06-04

← Chapter 32: AI Extraction | Chapter 34: Editing AI Results →