NotebookLM

The current sample contains 91 fully audited posts from 83 creators: 11 TikToks, 3 Instagram posts and 77 YouTube videos. Twenty-one posts exceed 200K views, seven exceed 500K, and two exceed 1M. The biggest recovered run is not an AI-news account: a 10.2M-subscriber Spanish math teacher earns 5.8M views by turning six Porfiriato sources into a cited notebook, podcast and mind map.

Nursing students supply the other repeatable breakout system, pairing personal results—passing med-surg, sleeping again, keeping a GPA high—with visible source-to-quiz, flashcard and Audio Overview workflows. The expansion also reaches teachers, language learners, researchers, product managers and tiny tutorial channels; creator size now runs from 2 subscribers to 14.5M, so the sample captures micro, mid and large accounts rather than equating reach with authority.

Across those audiences, the strongest script is consistent: name a source-heavy pain, show the uploaded evidence, then reveal a reusable output. Two audience-first videos use NotebookLM only as the explicit benchmark for Google Illuminate, and their low or medium product visibility is kept distinct from a NotebookLM demo.

No retained post or account exposes a NotebookLM-specific paid, affiliate, ambassador or standing creator-program relationship; the newest independent workflow and comparison rows are stored as `organic`, while the others remain `unknown`; official Google posts retain owned context without rewriting the evidence layer, and the dedicated `How to NotebookLM` channel explicitly says it is unaffiliated.

This is a platform-limited sample, not a census: all 60 exact rows in the bounded expansion queue are canonical and the two-week refresh adds a later agentic workflow, a high-frequency feature overview and a business-data demonstration, an automated trend-radar build, a source-grounded statistical dashboard, an Obsidian-to-Drive source bridge, and a beginner workflow from source import through Audio Overview, while lower-value creator-feed repeats and the duration-exclusion ledgers remain outside the sample.

Every retained row has a stable join, full transcript and manually reviewed frame set.