The prompt
Copy and paste this into your AI tool of choice. The prompt assumes you can attach or paste the input documents inline; substitute as needed for the tool’s interface.
You are a legal-AI governance specialist. The user will describe their law firm in structured form (size, practice areas, jurisdictions, tools currently in use). Your task is to produce a first-draft AI-usage policy for the firm, mapped to the relevant Model Rules of Professional Conduct.
The firm-description input fields the user will provide:
- Firm name
- Number of attorneys
- Number of non-attorney staff
- Practice areas
- Jurisdictions where the firm practises
- AI tools currently in use (or planned)
- Firm leadership structure (managing partner, executive committee, etc.)
- Existing technology-use policies (yes/no/partial)
- Particular constraints (insurance carrier requirements, client-imposed restrictions, etc.)
Produce the policy with these sections:
PREAMBLE
One paragraph stating the firm's commitment to competent and ethical use of AI in legal practice.
SECTION 1: SCOPE AND DEFINITIONS
- What "AI" and "generative AI" mean for the policy
- Who the policy applies to (lawyers, paralegals, secretaries, contractors)
- When the policy applies (any client-related work)
SECTION 2: APPROVED TOOLS
- List of currently approved AI tools by category (legal-research, drafting, document-review, embedded-PMS, general-purpose)
- Which tier of each approved tool is approved (consumer vs enterprise)
- Process for proposing additions to the approved list
- Process for removing tools
SECTION 3: COMPETENCE OBLIGATION (Rule 1.1)
- Each attorney's individual obligation to be trained on each tool used
- Training cadence (initial + quarterly)
- Training documentation requirements
- Process for situations where a tool changes materially
SECTION 4: CONFIDENTIALITY (Rule 1.6)
- Categorical rules: what categories of client information may go into which tier of which tool
- Mandatory redaction or anonymisation steps
- Specific prohibition on consumer-tier AI for client documents
- Vendor due-diligence requirements before any tool joins the approved list
SECTION 5: SUPERVISION (Rule 5.3)
- Supervising attorney's obligations for non-attorney use of AI
- Quality-control review of AI outputs before they leave the firm
- Documentation of supervision
SECTION 6: COMMUNICATION WITH CLIENTS (Rule 1.4)
- When AI use must be disclosed to the client
- Form of disclosure (engagement letter, separate notice, oral)
- Documentation of disclosure
SECTION 7: CANDOR TO THE TRIBUNAL (Rule 3.3)
- Verification requirements before filing any AI-assisted document
- Specific reference to the Mata v. Avianca standard
- Process if AI-fabricated content is discovered post-filing
SECTION 8: FEES (Rule 1.5)
- Treatment of AI savings under the firm's billing model (hourly, flat-fee, contingent)
- Specific prohibition on billing time AI did
- Disclosure of AI use as it relates to fees, where applicable
SECTION 9: DATA RETENTION
- How long AI conversations and outputs are retained
- Where they are stored
- When and how they are deleted
- Coordination with the firm's record-retention policy
SECTION 10: TRAINING
- Required training for each role
- Training records the firm maintains
- Annual review
SECTION 11: ENFORCEMENT
- Who is responsible for compliance
- Reporting violations
- Disciplinary process
SECTION 12: REVIEW AND AMENDMENT
- Quarterly review schedule
- Process for amending the policy
Constraints:
- The policy should be 8-15 pages of single-spaced text.
- Every reference to a Model Rule should cite the rule number and (if available) the relevant Comment.
- Where the firm's jurisdiction differs from the ABA Model Rules, note that the policy must be reviewed against the state's rules.
- Use plain English where possible; specialist legal language where necessary.
- Include placeholder text for jurisdiction-specific elements (e.g., "[STATE BAR OPINION REFERENCE]").
End the policy with:
- Effective date placeholder
- Amendment-history placeholder (table)
- Signature lines for the managing partner and ethics-counsel attorney
This is a FIRST DRAFT. The firm must review with its ethics counsel before adoption.Input
Input format
The user provides the firm description as a structured list. The prompt’s output adapts to the inputs (firm size affects supervision depth; practice areas affect confidentiality concerns; jurisdiction affects which state-bar opinions are cited).
Expected output
Output format
A 12-section policy document of 8-15 pages, in plain text or markdown. The output is structured for direct copy-paste into a Word document, with placeholder fields clearly bracketed.
Verification — what the lawyer must do after
- Review with ethics counsel before adoption. The output is a first-draft, not a finalised policy. The firm’s ethics counsel should review for jurisdiction-specific requirements, insurance-carrier alignment, and any client-imposed constraints.
- Verify state-bar citations. The policy will reference state-bar opinions; confirm the citations are accurate for your state and current.
- Update the approved-tools list quarterly. Vendor terms change; new tools emerge; existing tools update. The policy should be a living document reviewed at least every 90 days.
- Distribute, train, and document. A policy that exists but is not trained on does not satisfy Rule 5.3. The firm must distribute, train every staff member on it, and retain training records.
⚠ Risks and failure modes
- Generic-policy risk: A policy generated from a general prompt may miss firm-specific concerns. Read the output critically and edit for the firm’s actual practice.
- Jurisdiction-specific risk: The output uses ABA Model Rules. Several states have stricter rules (California, New York, Florida); these should be cross-referenced.
- Stale-vendor risk: The approved-tools list reflects a moment in time. Without quarterly review, the policy will diverge from actual practice.
- Compliance-theatre risk: A policy that exists but is not enforced is worse than no policy — it creates documented non-compliance. The enforcement section must be operative, not aspirational.
Vendor compatibility
Best on Claude or GPT-4 with a long context window. CoCounsel and Spellbook will produce defensible drafts with built-in legal-research overlay.
Citations and further reading
- ABA Model Rules of Professional Conduct.
- ABA Formal Opinion 512 — the implementation playbook the policy operationalises.
- IXSOR: State bar AI opinions, comparative tracker.
- IXSOR: AI training curriculum — the training section of the policy operationalises this curriculum.