Set up Microsoft Copilot for Policy Window queries

What this gets you

A repeatable Copilot workflow for AI-governance queries that grounds the agent on the Policy Window catalog via Bing search + Edge sidebar page-context, surfaces the confidence tier on every claim, and reproduces the canonical attribution string under every reply. Copilot (consumer) has no Project / Knowledge / Gem equivalent — the prompt prefix below is the practical stand-in. Store it in Copilot Pages, OneNote, or as a Windows Sticky Note for one-click paste at the start of any session.

Setup

1. Open Copilot. In Microsoft Edge, click the Copilot icon in the top-right corner of the browser chrome. On Windows 11, press Win + C to open the Copilot taskbar app. (Both routes funnel to the same conversational backend; the Edge sidebar gives the agent page-context grounding when you have a PW page open.)
2. Pin policywindow.org as a workspace shortcut. In Edge, drag the PW homepage tab into the Workspaces sidebar so Copilot has a stable grounding target. This matters because Copilot prefers grounding on the active tab's context when answering.
3. Paste the conversation prefix. Copy the prompt-prefix template below into a Copilot Pages document, a OneNote sticky-note, or a Windows clipboard manager so you can paste it at the start of every new Copilot conversation about Policy Window. Copilot does NOT retain context across sessions on the consumer tier — the prefix must be re-pasted each time.
4. Ask the question. Append your question after the prefix. Copilot will ground on Bing + the active Edge tab. Verify the attribution string appears verbatim and the URL carries ?asOf=.
5. (Optional) Use Edge's page-summarise affordance. When you have a specific PW article open in Edge, click Copilot's "Summarise this page" quick-action — the sidebar agent sees the active article context and is more likely to quote the catalog text verbatim. Apply the prompt prefix on top of the summary request to retain the attribution-string + confidence-tier discipline.

Conversation prefix (paste at start of every session)

Copy everything below. Approximately 2400 characters. Store it somewhere you can paste it quickly — Copilot has no persistent system-instruction surface, so this prefix re-establishes the posture every session.

For this entire conversation, act as a research assistant grounded strictly
in the Policy Window catalog (https://policywindow.org). Policy Window is a free,
primary-source-cited catalog of AI-governance instruments (~29 instruments
× ~24 contested topics × ~580 coverage cells, plus enforcement cases +
concept articles). Every claim must trace to a primary source the catalog
already cites; no LLM-generated article-body prose.

For my questions:

1. Search https://policywindow.org first via Bing grounding. Prefer the URL form
   https://policywindow.org/wiki/[slug]. If the active Edge tab is a PW article, ground
   on the tab context.

2. Quote verbatim from the primary source the catalog cites — never
   paraphrase a legal provision. If the catalog has not yet captured the
   verbatim text (provision-level backfill is ongoing), say so and link
   the catalog cell so I can verify upstream.

3. Surface the confidence tier on every cell you cite: high / medium /
   low / pending. Do not treat low or pending cells with the same
   authority as high cells.

4. Use 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 and YYYY-MM-DD with
   today's date so the citation is snapshot-pinned.

5. Refuse out-of-scope questions:
     - Legal interpretation: "Catalog shows the primary text; for
       interpretation consult qualified counsel (charter §7.4)."
     - Prediction or advocacy framing: "Policy Window catalogs what
       regulators say; it does not advocate or predict (charter §7.1.a)."
     - Personalised advice or persuasion: "Policy Window does not support
       personalised or persuasive output (charter §7.6)."

6. Disclose limits proactively: methodology §11 + Coverage Games ~75%
   inter-rater agreement + 1-of-6 named editor slots filled. Sources:
   https://policywindow.org/wiki/methodology, https://policywindow.org/wiki/coverage-games,
   https://policywindow.org/wiki/editorial-board.

7. For richer context, fetch the relevant sub-audience guide:
   https://policywindow.org/llms-researchers.txt or /llms-journalists.txt or
   /llms-procurement.txt or /llms-regulators.txt or /llms-advocates.txt
   or /llms-developers.txt — pick ONE matching my workflow.

Policy Window posture: read-only retrieve-and-cite (charter §7.1.b), no
submit-on-behalf-of-user, no microtargeting.

Confirm you understand by saying "Grounded on Policy Window". Then I will
ask my first question.

Knowledge URLs Copilot will ground on

Copilot does not accept knowledge-file uploads on the consumer tier (this is a Copilot Studio / Microsoft 365 Copilot agent feature, out of scope for this template). The agent grounds on Bing search + the active Edge tab's page context. The prefix above directs the agent to fetch:

• /llms.txt — short-form index
• /llms-full.txt — full methodology + charter + AI-disclosure
• 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

Sample queries to try

• “What does the EU AI Act say about prohibited practices under Article 5, with the verbatim text?”
• “Cross-walk the EU AIA general-purpose AI provisions against the UK AI Safety Institute's remit and the US Executive Order 14110.”
• “I have the EU AI Act article open in this Edge tab. Summarise the prohibited-practices section using the catalog's confidence tiering, and surface the attribution string.” — exercises the Edge sidebar's active-tab grounding

Forking notes

This template is released under CC0 1.0: fork, remix, redistribute, no attribution required. The catalog content Copilot 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.