Notebook LM · A Practitioner's Take

Every AI tool is guessing.
One isn't.

ChatGPT, Gemini, Copilot — they blend your documents with whatever else they were trained on, then hand you an answer that sounds certain. Notebook LM doesn't. It answers only from the sources you give it, and shows you the receipt.

01 — The Problem

Most AI tools blend your information with information you didn't ask for.

you: summarize our Q3 retention strategy from the deck
Generic AI · Typical Response
The Q3 strategy focuses on reducing churn in the mid-market segment through quarterly business reviews — which aligns with 2024 benchmarks suggesting 8–12% churn is typical for SaaS. The deck also outlines a referral incentive program launching in October, similar to approaches used by Dropbox and Airbnb in their early growth stages.
From your actual document Plausible-sounding filler the model added
01

Can't trace it

Your content gets mixed with the model's training data. You can't tell which is which without re-reading the source yourself.

02

Can't trust it

Hallucinations and real facts read the same — which is exactly the problem for client-facing work.

03

Can't scale it

Every answer still needs a human to verify. That defeats the point of using AI to move faster.

02 — What Makes It Different

Notebook LM is grounded.
It only knows what you upload.

Generic AI assistants
Open world
  • Answers pull from training data + your docs, mixed
  • No reliable way to tell which sentence came from where
  • Will invent citations if you ask for them
  • Same answer sounds confident whether it's right or wrong
Notebook LM
Closed · Sourced
  • Answers come only from the sources you've uploaded
  • Every claim ships with a numbered citation you can open
  • Tells you when your sources don't contain the answer
  • You review the receipt before you ship the work
Notebook LM · Grounded Response 04 sources · 04 citations
The Q3 retention strategy targets mid-market accounts with quarterly business reviews led by the CS team1. A referral program launches October 14 with a $500 credit per qualified referral2. Success metrics are NRR ≥ 108% and a churn ceiling of 6%3, tracked monthly by RevOps4.
[1]Q3-Strategy-Deck.pdf · slide 12
[2]Referral-Program-Brief.docx · §2.1
[3]FY25-OKRs.gdoc · line 47
[4]RevOps-Dashboard-Spec.pdf · p.3
03 — The Agency Playbook

Three ways agencies are using it right now.

01
Client Knowledge Base

Every client account, instantly searchable.

Brand guides, past campaigns, contracts, meeting recordings — all in one notebook. Your team asks in plain English and gets answers grounded in that client's actual material, cited line by line. Onboarding a new AM drops from weeks to an afternoon.

Internal QueryAcme · NB
What tone rules did Acme sign off on in '24?
No em dashes. No "game-changing"1. Active voice, 14-word sentence cap2. Legal flagged "guarantee" as banned in May3.
[1]Acme-Brand-V4.pdf · p.8
[2]Copy-Principles.gdoc
[3]Legal-Memo-05-22.pdf
02
Pitch Intelligence

Walk in as the best-prepped person in the room.

Load the prospect's 10-K, earnings calls, press releases, competitor teardowns. Ask "what are they actually worried about?" Notebook LM surfaces the themes with quotes, pages and dates. Your deck writes itself around real evidence, not a Perplexity summary.

Pitch PrepProspect · NB
Top 3 concerns flagged by their CFO this year?
DTC margin compression1, retention after the Q1 promo2, and rising CAC on paid social3 — all raised on the Q2 earnings call.
[1]Q2-Earnings-Transcript · 14:32
[2]Q2-Earnings-Transcript · 22:05
[3]Q2-Earnings-Transcript · 38:40
03
Training & Onboarding

Your SOPs, answering questions for you.

Playbooks, process docs, Loom transcripts, training wikis — all searchable. New hires ask the notebook, not Slack. Every answer points to the exact SOP or timestamp. Your senior team gets their afternoons back.

New Hire · Day 2Internal · NB
How do I kick off a new paid-social audit?
Start with the intake form1, pull 90 days of platform data, then run the 11-point audit checklist2. Lead time is 5 business days3.
[1]SOP-Paid-Audit.pdf · §1
[2]Audit-Checklist.gdoc
[3]Onboarding-Wiki · SLAs
A note from Digital Velocity

Stop pasting your work into a black box.

If the AI you're using can't show you where its answer came from, you're not using AI — you're gambling with it. Notebook LM is the first tool that earned a seat in our client-facing work. Load your sources. Ask the hard questions. Ship with confidence.

Scroll to Top