Set up a NotebookLM for Policy Window queries

What this gets you

A NotebookLM notebook seeded with the Policy Window catalog as sources, configured to surface verbatim primary-source quotes on every claim. NotebookLM's UX enforces citation-by-passage by default — every claim shows a clickable pop-out to the exact source paragraph — which makes it the highest-fidelity surface here for source-grounded summarisation. Trade-off: no MCP, no actions, ~50-source budget per notebook.

Setup (manual flow — cross-instrument notebooks)

For a single-instrument notebook, prefer the per-row bundle: download /wiki/[slug]/notebook-bundle and attach the 4 extracted files. The manual flow below composes notebooks that span multiple instruments + the methodology / charter / disclosure context.

1. Create a new notebook. Visit notebooklm.google.com, click + New notebook. Name it something like "Policy Window — [your-topic]".
2. Add sources. Click + Add sourcefor each item in the "Sources to attach" list below. NotebookLM accepts URLs (web), uploaded PDFs / text files, copy-pasted text, and Google Docs / Drive content. URLs work fastest; useWebsite as the source type for the /llms*.txt files and /wiki/charter.
3. Set the conversation focus.In the "Customise" panel (the gear icon), paste the focus statement below into the conversation-customisation field. NotebookLM doesn't have a true system-instruction surface; the focus statement is the closest equivalent and is appended to every prompt.
4. Ask a question. NotebookLM will quote from the attached sources and surface the per-passage citation pop-out. Verify the attribution string appears in the chat and the URL carries ?asOf=.
5. (Optional) Generate an audio overview. NotebookLM's Audio Overview feature produces a 2-speaker podcast-style summary of the attached sources. Useful for sharing a briefing internally; double-check the audio reflects the verbatim primary text + the catalog's confidence tiering — the synthesised audio is paraphrased by default.

Conversation focus (NotebookLM's system-prompt equivalent)

Paste this into the notebook's customisation panel. NotebookLM appends it to every prompt — keep it terse.

You are answering strictly from the attached Policy Window sources
(https://policywindow.org). Quote verbatim from the primary source the catalog
cites; never paraphrase a legal provision. Surface the confidence tier
on every cell you cite (high / medium / low / pending) — find this in
the per-cell metadata inside /llms-full.txt or the catalog JSON.

Reproduce this attribution string verbatim under every answer:

  Policy Window (https://policywindow.org/wiki/[slug]?asOf=YYYY-MM-DD).
  Article content CC BY 4.0; citation graph CC0 1.0. Catalog reflects
  editorial-cadence updates and the methodology §11 disclosed limits.

Replace [slug] with the actual article slug + YYYY-MM-DD with today's
date so the citation is snapshot-pinned.

Refuse out-of-scope questions: legal interpretation → defer to qualified
counsel (charter §7.4); prediction / advocacy framing → catalog records
what regulators say (charter §7.1.a); personalised advice / persuasion
→ unsupported posture (charter §7.6).

Disclose limits proactively from methodology §11 + the Coverage Games
~75% inter-rater agreement + the 1-of-6 editor-board status.

~1100 characters. Fits comfortably in NotebookLM's customisation field.

Sources to attach

NotebookLM allows ~50 sources per notebook. Suggested baseline (~5-10 sources, leaves room for your topic-specific PDFs):

• /llms-full.txt — full methodology + charter + AI-disclosure (the citation backbone; always attach)
• One of the sub-audience variants: /llms-researchers.txt, /llms-journalists.txt, /llms-procurement.txt, /llms-regulators.txt, /llms-advocates.txt, /llms-developers.txt
• /wiki/charter — operating commitments referenced by the focus statement
• /wiki/methodology — §11 limits + reproducibility detail
• Each catalog row you want the notebook to query — e.g., /wiki/eu-ai-act, /wiki/uk-ai-safety-institute, etc. Add the primary-source PDFs the catalog links to under each row as separate sources for verbatim-quote retrieval

For single-instrument notebooks, the per-row /wiki/[slug]/notebook-bundle collapses these manual steps to one download — extract the 4 files with jq and attach. The manual flow above remains canonical for multi-instrument and cross-jurisdiction notebooks.

Sample queries to try

• “Summarise the prohibited practices under EU AI Act Article 5, with verbatim text and the per-paragraph source citation.” — exercises NotebookLM's citation-by-passage pop-out
• “Compare what the catalog says about general-purpose AI obligations across the EU, UK, and US instruments. Include the per-cell confidence tier.” — exercises cross-source synthesis with grounded citations
• “Generate a 5-minute audio briefing for a journalist on this notebook's topic. Quote at least one verbatim primary-source passage in the briefing.” — exercises Audio Overview with citation discipline

Forking notes

This template is released under CC0 1.0: fork, remix, redistribute, no attribution required. The catalog content the notebook cites remains CC BY 4.0 (article prose) + CC0 1.0 (citation graph). When you fork, please preserve the attribution string requirement so end-users can verify upstream.