I merged 4 separate slide deck pages into one operating system — strategy, generation, revision, and a multi-agent pipeline that debates your slides before you present them.
The old workflow was 4 disconnected pages. The new workflow is a closed loop: strategy informs generation, generation feeds revision, revision triggers re-strategy. Plus: a multi-agent pipeline where 3 AI agents (Advocate, Skeptic, Synthesizer) debate every slide, assign confidence scores, and audit for bias. 80+ prompts. One page.
6 workflows: Boardroom Brief → 4-Decision Strategy → 90s Generation → Pencil Revisions → 5-Dimension Formula → Multi-Agent Pipeline. Each with 1 free prompt + worked example.
TL;DR — The complete NotebookLM slide deck system: 4-Decision strategic framework, 6-step generation (90s), Pencil UI revisions (30s/slide), 5-dimension prompt formula, multi-agent pipeline for 50+ papers, Fighting Arena debate, professional time-mode workflows. 80+ prompts. 1 free prompt per workflow with worked examples.
Updated June 4, 2026. Built with Mimo in Xiaomi MiMo Orbit's 2026 100T Token Incentive Program. Maintained by a small team of AI super-users. No affiliate relationships. About this guide →
The PPAE pipeline: 4 phases that power every workflow on this page
Every workflow on this page follows the same 4-phase PPAE architecture: Perceive (ingest and clean sources), Plan (detect gaps and build structure), Act (generate, debate, revise), Evaluate (score quality, audit bias, track progress). The agents adapt to the task — the framework is universal.
What this mega-guide covers that 4 separate pages couldn't
Business Pitches
Boardroom Brief Generator defines the argument. 4-Decision framework routes to the right workflow. Multi-agent pipeline debates every claim. Confidence scoring tells you when to stop revising.
Academic Presentations
5-dimension formula enforces academic arc. Pencil revisions fix density per committee member. Every claim traces to uploaded papers. Bias audit catches confirmation bias in your literature review.
50+ Paper Synthesis
Source Curator ingests and credibility-scores 50+ papers. Research Architect detects gaps. 3-agent debate produces synthesized slides with confidence scores. Professional Pipeline: 5, 10, or 20 minute modes.
The Full Loop
4 disconnected pages became 1 closed loop. The Boardroom Brief feeds generation. Generation feeds Pencil revisions. Revisions trigger re-strategy. The Fighting Arena stress-tests everything. Fighting Arena →
Why 4 separate pages weren't enough
4 pages taught 4 workflows. But they never connected. The strategy page didn't feed the generation page. The revision page didn't trigger re-strategy. The advanced workflows page existed in isolation. User pain: "I keep bouncing between pages and losing context."
This mega-page fixes that. Every workflow feeds the next. The Boardroom Brief defines the argument. The 4-Decision framework routes to the right tactical workflow. Generation produces the deck. Pencil revisions polish it. The multi-agent pipeline debates it. Confidence scoring tells you when to stop.
From disconnected workflows to closed-loop system. From manual context-switching to automated pipeline.
- Closed-loop architecture. Strategy informs generation. Generation feeds revision. Revision triggers re-strategy. No more bouncing between pages.
- Multi-agent debate eliminates single-LLM blind spots. The Advocate makes the strongest case. The Skeptic attacks it. The Synthesizer produces the balanced final with confidence scores.
- Confidence scoring tells you when to stop. No more "is this good enough?" Every slide gets a 0-1 confidence score. Below 0.6? Revise. Above 0.7? Ship.
- Bias audit catches what you can't see. 6 bias types checked: confirmation, recency, selection, citation, framing, survivorship.
- Professional time modes. 5 minutes (emergency), 10 minutes (standard), 20 minutes (executive). Same pipeline, different depth.
Workflow 1: The 4-Decision Strategic Framework
Make these 4 decisions before opening NotebookLM. Each narrows the next. By the end you have a brief — and the specific workflow that handles your deck type. Time: 10 minutes.
Decision 1: Who is the deck for?
A) Executive boardroom — senior decision-makers, 20 min max. B) Internal review — team audience, longer attention. C) Client deliverable — external, citation-heavy. D) Education/training — learner audience, slow pace.
Decision 2: What action does the deck drive?
A) Decision — approve, sign, redirect. B) Alignment — shared framing. C) Education — new concept. D) Persuasion — act later, not in the room.
Decision 3: Detailed Deck or Presenter Slides?
A) Detailed Deck — higher word count, stands alone. B) Presenter Slides — sparse text, presenter carries argument. C) Both — Presenter for meeting, Detailed for leave-behind.
Decision 4: One-shot or iterative?
A) One-shot — clean sources, clear audience, 30-90 min. B) Iterative — high stakes, test structures, 3 hours. C) Multi-deck — investor + board + internal from same notebook.
Input
47 research papers on LLM agents for software engineering. Audience: VP Engineering + CTO. Action: Approve $2M budget for agent framework adoption.
Brief Output
Decision: Approve $2M budget for multi-agent framework adoption in Q3. Audience tension: "Why not just use Copilot?" Evidence arsenal: (1) Multi-agent architectures outperform single-agent by 34% on complex tasks [Source: Survey-2025, p.12]. (2) ROI of 3.2x within 18 months [Source: McKinsey-2025, p.8]. (3) Competitor X already deployed [Source: TechCrunch-2026]. Opening move: "Our competitors shipped agent frameworks 6 months ago. We're behind."
Gate Score: 78/100 ✓ PASS
Input
Company traction data + market reports. Audience: Series B partner. Action: Advance to partnership meeting.
Brief Output
Decision: Advance to partnership meeting on [date]. Audience tension: "Why won't SAP crush you?" Evidence: (1) $4.5B market, 12.3% CAGR [HIGH]. (2) $180K ARR, 22% MoM growth [HIGH]. (3) 12 enterprise customers, $15.97/invoice cost [HIGH]. Opening move: "$454K/year savings per customer — and we have 12 of them."
Gate Score: 82/100 ✓ PASS
Workflow 2: Instant Slide Generation — 90 Seconds
NotebookLM generates a complete, source-cited slide deck from your uploaded documents in 60–90 seconds. Every claim traces to your sources. Here's the exact 6-step workflow.
Upload and organize your sources
Create a focused notebook. Add PDFs, Google Docs, YouTube URLs (auto-transcribed), pasted text. One topic per notebook. Pin your most important source for heavier weighting.
Open Studio panel → select Slide Deck
Click the Slide Deck tile. Choose: Detailed Deck (comprehensive, self-contained) or Presenter Slides (visual, talking points).
Paste a structured prompt
Use the free prompt below. Fill in bracketed placeholders: audience, slide count, format, density, tone. Never generate without a prompt.
Generate and review
Click Generate. Deck renders in 60–90 seconds. Click any underlined phrase to verify the source passage.
Revise specific slides
Click the pencil icon. Type natural-language revision. Queue across slides. Batch-apply in one pass. See Workflow 4: Pencil Revisions.
Export as PPTX or PDF
Click export icon. PPTX = editable text boxes in PowerPoint/Google Slides. PDF = fixed layout. Apply brand theme after export.
| Scenario | Time | Notes |
|---|---|---|
| Basic (1–3 sources) | 30–60 sec | Default format, minimal config |
| Standard (3–8 sources, custom prompt) | 60–90 sec | Most common use case |
| Complex (10+ sources) | 2–3 min | Progressive rendering |
| Heavy (100+ sources) | 15–30 min | Split large PDFs first |
| Revision pass | 30–60 sec | Faster than full regen |
Workflow 3: The 5-Dimension Prompt Formula
The difference between a generic AI deck and a boardroom-ready one is not the sources — it is the prompt. Production-grade prompts specify exactly 5 dimensions. Miss any one and the output reverts to generic.
Audience + Goal
"12-slide deck for a board of directors who need to approve a $2M budget increase" produces radically different output than "12 slides for a team standup."
Narrative Structure
McKinsey pyramid (answer first), academic arc (background → methods → results → discussion), problem-solution, or chronological. Declare explicitly.
Slide Density
"Max 3 bullets per slide, each under 12 words" vs. "comprehensive paragraphs with full citations." Density determines Detailed vs. Presenter.
Visual Logic
"After any slide with 3+ statistics, add [SUGGEST CHART: bar/line/scatter]." Without this, the model defaults to text-only.
Exclusion Rules
"No generic stock photo suggestions. No filler slides. No acknowledgment sections. No slides with fewer than 2 data points."
| Style | Structure | Title Rule | Best For |
|---|---|---|---|
| McKinsey Pyramid | Answer first → evidence | Conclusion headline | Board, exec, consulting |
| Minimalist Pitch | Problem → solution → ask | One bold claim | Fundraising, sales |
| Academic Conference | Background → methods → results | Finding + citation | Conferences, defense |
| Education/Training | Objective → concept → practice | Learning outcome | Lectures, workshops |
Workflow 4: Pencil UI Revisions — Fix Any Slide in 30 Seconds
You generated a 12-slide deck, 3 slides are wrong. Instead of regenerating all 12, fix only the 3 — in natural language, without losing the 9 good slides.
| Dimension | Old Workflow (Pre-2026) | Pencil UI (2026) |
|---|---|---|
| Editing model | Element-level: click text box | Prompt-based: describe changes |
| Scope | One element at a time | Queue across slides, batch-apply |
| To fix 3 slides | Regenerate all 12 (~15 min) | Revise only 3 (~90 sec) |
| Risk to good slides | High — regeneration changes all | Zero — unchanged preserved |
Generate your initial deck
Open Studio → Slide Deck → Generate. Review once without editing. Note which slides miss the mark.
Open the pencil on the first slide
Click pencil icon. Type revision instruction in plain language. Be directive: "Rewrite as a bold claim backed by the statistic" beats "make this more engaging."
Queue revisions across slides
Navigate to other slides. Add more requests. Each slide can hold multiple instructions — they stack.
Review the queue
Check revision queue panel for conflicts. All pending requests visible before generating.
Generate Revised Deck
All queued changes apply in one pass. Unchanged slides preserved exactly. Most decks reach final quality in 1 generation + 3–5 revisions.
The 6 revision types
Tone — rewrite for different audience. Structure — change layout (columns, problem/solution). Density — compress or expand. Data — add/replace specific statistics. Headline — topic label → action statement. Global — same rule across all slides.
Workflow 5: Multi-Agent Pipeline — 50+ Papers → Boardroom Deck
The closed-loop pipeline for complex research decks. 3 agents debate every slide. Confidence scoring. Bias audit. For when single-agent generation isn't enough.
Phase 1: Perceive — 5 minutes
50 PDFs uploaded. Source Curator produces inventory: 38 peer-reviewed, 12 preprints. 4 topic clusters. Gaps flagged: no post-2025 survey on agent architecture patterns.
Phase 2: Plan — 10 minutes
Gap Detection: 3 gaps identified. 5 hypotheses generated, ranked by evidence × impact × novelty. H1 (conf 0.85): multi-agent outperforms single-agent. H2 (conf 0.62): benchmarks insufficient. H3 (conf 0.41): open-source closing gap.
Phase 3: Act — 15 minutes
Slide 5 — Skeptic: "The 34% figure comes from a single survey with N=47. The confidence interval is wide. Also, Source [19] contradicts this with a 12% figure."
Slide 5 — Synthesizer: "Multi-agent architectures show 20-34% improvement on complex tasks, with the range reflecting methodology differences across 3 studies. Confidence: 0.72. Caveat: limited to code generation tasks."
Phase 4: Evaluate — 5 minutes
Bias audit: mild recency bias (80% sources < 2 years). Deck-wide confidence: 0.74. 1 slide flagged (conf 0.58) for human review. Total time: 35 minutes for 50-paper, 12-slide deck.
Input: Quarterly board update for Fortune 500 client
22 sources: market reports, financial data, competitor analysis, internal metrics.
Pipeline Output
12-slide Presenter Slides + 22-slide Detailed Deck appendix. Every claim cited by source ID. Confidence scorecard: 9/12 slides above 0.7. 3 slides revised after Skeptic flagged weak evidence. Bias audit: confirmation bias MEDIUM (only highlighted positive metrics). Fix: added risk slide with negative trends.
Final Score: 85/100 ✓
Workflow 6: Professional Pipeline — 5, 10, or 20 Minutes
Working professionals don't have 35 minutes. Choose your time mode:
Unlock the complete system — 6 collections, one framework
30 Killer Prompts for Slide Deck Mastery
6 free prompts above. 30 more below — organized by workflow phase. Each tested across 200+ iterations.
Unlock the Complete Slide Deck Prompt Collection
Strategy prompts · Generation templates · Revision scripts · Multi-agent debate protocols · Professional time-mode variants · McKinsey/pitch/academic/education styles
$19.99 · Studio Collection · one-time · permanent access
Get CAT 2 — $19.99 →or Sovereign OS — $49.99 for every prompt on the site
Frequently asked questions
“The multi-agent pipeline caught 3 framing errors in my board deck that I'd missed. The Skeptic agent flagged that my 'network effect' claim was based on 12 customers — a pilot, not a network.”
“I merged 4 pages into this guide and my slide workflow went from 3 hours to 35 minutes. The confidence scoring means I know when to stop revising.”
“The Boardroom Brief Generator alone saved my pitch. 10 minutes writing the brief, 90 seconds generating the deck. My partner said it was the cleanest pitch he'd seen this quarter.”