Analyze case law at scale with a precedent agent

Step 1: Define the legal question preciselyWord

Write the issue as a one-sentence legal question with the jurisdiction named explicitly — "Is a non-compete enforceable in [State] when the employee was terminated without cause?" not "research non-compete law."

Save this framing at the top of the Word doc the agent will eventually populate; a precedent search run against a vague question returns a vague pile of tangentially-related cases, not usable arguments.

Step 2: Build the analysis agent's document scopeCopilot Studio

In Copilot Studio, open the agent's Knowledge settings and add the case law database or the matter's specific document set as a connected source — don't leave it pointed at a generic, firm-wide repository, or it will surface precedent from unrelated practice areas.

Set the agent's instructions to require, for every case it surfaces, the specific holding and the facts that make it applicable, not just a topical match.

Prompt idea:

Find precedent cases in this jurisdiction addressing whether a non-compete clause is enforceable when the employee was terminated without cause. For each relevant case, state the holding, the key facts that made it applicable, and how it supports or undermines our position.

Step 3: Draft the brief from validated precedent onlyCopilot Studio

After the attorney marks up which surfaced cases are actually on point (strike the rest), paste only that validated citation list back into the agent and have it draft the brief section from those cases alone.

Re-running the draft against the full unvetted list — even "just to see" — risks an unverified or weakened case slipping into the final argument.

Step 4: Have an attorney own the final argument and citationsWord

Run every citation through your firm's standard cite-checking tool (Westlaw KeyCite, Lexis Shepard's, or equivalent) to confirm each case still represents good law before the brief goes anywhere near a filing.

Use Word's Track Changes while the attorney edits the drafted argument, so the revision history shows exactly what the agent produced versus what a human verified and changed — that distinction matters if the drafting process is ever questioned.

Check the work

• Independently verify every cited case actually exists and says what the agent claims — citation fabrication is a known failure mode for AI-assisted legal research.
• Confirm cited cases are still good law (not overturned or superseded) in the relevant jurisdiction.
• Have the reviewing attorney confirm the case's key facts are actually analogous to the current matter, not a surface-level topic match.

Source: Microsoft Copilot Scenario Library — Legal (2026)