Deep Research in Action: A Case Study
How a team completed an industry report in hours using Deep Research.
A four-person market research team had one business day to deliver a competitor briefing. They imported 12 core Sources into NotebookLM and used Deep Research to fill gaps before exporting a Studio PDF.
Constraint: every bullet in the final deck had to be traceable — a requirement NotebookLM's citation model satisfies natively.
Timeline (actual run)
| Time | Action | Notes |
|---|---|---|
| 0:00 | Upload 8 PDFs + 4 URLs | Investor decks, 10-K excerpts, earnings call transcripts |
| 0:20 | Chat: scope questions | Market size, share shifts, regulatory risks |
| 0:45 | Deep Research pass | Added 6 web Sources recommended by NotebookLM |
| 1:30 | Studio Briefing Doc | Executive summary + SWOT with citations |
| 2:00 | PDF export + review | PM verified each citation in under 30 min |
Before vs after
| Metric | Manual process | NotebookLM Deep Research |
|---|---|---|
| Wall-clock time | 2–3 days | ~2 hours active work |
| Citable bullets | Manual footnotes | Auto-linked Sources |
| Gap detection | Analyst memory | Deep Research suggestions |
| Stakeholder format | Rebuild in slides | Studio export + light edits |
Lessons learned
- Start with a narrow research question — broad prompts inflate Deep Research runtime
- Prefer structured PDFs over scraped HTML when accuracy matters
- Run a second Chat pass asking "what contradicts our thesis?" before exporting
- Keep a "source quality" note: flag paywalled or outdated items early
Team quote: "Deep Research didn't replace our judgment — it replaced the boring assembly step so we could spend time on interpretation."